Welcome to Sbookmarking

Description: Choosing a legal structure is the first real decision when starting a business in France, and it gets made badly more often than almost any other, usually because someone repeated what worked for them without asking what you are actually doing. There is no universally best option. There are four or five common ones with genuinely different trade-offs. Micro-entreprise The simplest. A sole trader regime with reduced accounting and social obligations. How you are taxed: social contributions are a flat percentage of turnover, varying by activity type. Income tax on turnover after a fixed allowance, or a flat withholding if you elect for it and qualify. Turnover ceilings: roughly €83,600 for services and €203,100 for commercial activity. Cross those for two consecutive years and you leave the regime. The trap everyone hits: the VAT threshold is separate and much lower. €37,500 for services, €85,000 for goods in 2026. You can be comfortably inside the micro regime and still liable for VAT. The real limitation: you cannot deduct expenses. Contributions are calculated on turnover, not profit. If you have significant costs, materials, subcontractors, equipment, this becomes expensive fast. A consultant with a laptop does well. A business with €40,000 of purchases does not. Liability: since the 2022 reform, an entrepreneur individuel's personal assets are separated from business assets by default, which removed the biggest historic drawback. Right for: low-cost service businesses, side activities, testing an idea, anything under about €50,000 with minimal expenses. Entreprise individuelle at the réel regime Same legal form as micro, different tax treatment. You keep proper accounts and are taxed on actual profit, deducting real expenses. Right for: sole traders whose expenses are high enough that being taxed on turnover stops making sense, but who do not need a company structure. EURL A single-shareholder SARL. A company with legal personality separate from you. Social regime: the managing partner is a travailleur non salarié. Contributions are lower as a percentage than the salaried regime, roughly 40 to 45 percent of remuneration, but so is the cover, particularly on pension and unemployment. Tax: income tax by default, with an option for corporation tax. The corporation tax option is often the point, since it lets you separate what the company earns from what you take out. Dividends: here is the catch that surprises people. In an EURL subject to corporation tax, dividends above a threshold linked to the capital are subject to social contributions, not just the flat tax. This substantially limits dividend optimisation. Right for: established sole operators with meaningful revenue who want limited liability and lower contribution rates. SASU A single-shareholder SAS. The structure most new French startups choose. Social regime: the president is assimilé salarié. Contributions are considerably higher, roughly 70 to 80 percent of net remuneration, and the cover is correspondingly better, closer to an employee's, though still without unemployment insurance. Tax: corporation tax by default. Dividends: taxed at the flat rate, commonly called the flat tax, with no social contributions. This is the structural advantage over the EURL and the reason many people choose it. Flexibility: SAS statutes are largely freely drafted. This matters enormously if you plan to bring in investors or partners, because you can structure share classes and governance as you like. SARL rules are more rigid. The cost: if you pay yourself nothing, you pay almost no contributions, but you also accrue no pension rights and have no health cover through the company. People forget this in year one and notice in year five. Right for: businesses expecting investment, multiple founders, or those planning to take income primarily as dividends. SARL and SAS The multi-shareholder versions of the above. Same trade-offs, plus the governance question. SARL is rule-bound and predictable, SAS is flexible and requires competent drafting. Most investors prefer SAS. How to actually decide Four questions get you most of the way. How much will you spend to earn? High expenses rule out micro immediately. How will you take money out? Mostly salary points toward EURL. Mostly dividends points toward SASU. Do you need social cover? If this business is your only income and you have no other cover, the assimilé salarié regime under SASU is genuinely better protection despite the cost. Will you raise money or add partners? If yes, SAS or SASU. Converting later is possible but costs time and money. What people get wrong Optimising for the first year. Micro looks best when revenue is small. If you expect to pass the ceilings within eighteen months, starting there means changing structure at the worst possible moment. Ignoring contribution differences until the first bill. The gap between TNS and assimilé salarié rates is large enough to change what you can pay yourself. Copying a friend's structure. Their expense profile, income plans and risk tolerance are not yours. Forgetting that changing is possible. Structures are not permanent. Starting simple and converting when the numbers justify it is a legitimate strategy, and cheaper than over-engineering on day one. Where to get an answer This is a decision worth an hour with an accountant, and it is one of the few areas where that hour reliably pays for itself. Go in prepared: your expected revenue, your expected costs, how you want to be paid, and whether you plan to raise money. For working through the options before that conversation, a tool scoped to French business questions gets you further than a general assistant, which will frequently answer with rules from another country. Mirage Cloud includes agents covering French accounting and business law, which is useful for understanding the landscape and framing better questions. Use it to prepare. Do not use it to decide. The structure affects your tax position for years, and this is one of the places where professional advice is genuinely worth paying for.

Category:

Tag:

Description: Your accountant sends the annual accounts, you look at the bottom line, and you file them. That is what most small business owners do, and it means the single most useful document about your business gets about ninety seconds of attention a year. The statements are not difficult once someone explains what the lines mean. Here is that explanation. The profit and loss account This covers a period, usually a year, and answers one question: did you make money. It works down in layers, and the layers are the point. Revenue. What you invoiced. Note that this is not what you were paid, which is a distinction that catches people out constantly. Cost of sales. The direct costs of delivering what you sold. Materials, subcontractors, the direct labour on the job. Gross profit. Revenue minus cost of sales. Expressed as a percentage of revenue, this is your gross margin, and it is the most diagnostic number in the whole document. It tells you whether the thing you sell is fundamentally profitable, before any overhead. If gross margin is falling, either your prices are too low or your delivery costs have risen, and no amount of overhead cutting fixes it. Operating expenses. Everything not tied to a specific sale. Rent, salaries not on jobs, insurance, software, marketing, professional fees. Operating profit. Gross profit minus operating expenses. Whether the business, as a whole operation, works. Net profit. After interest and tax. The bottom line everyone looks at first, and the least informative of the layers, because by the time you get there the causes have been averaged away. The useful habit is reading the layers rather than the total. A business with a strong gross margin and a bad net result has an overhead problem, which is fixable. A business with a weak gross margin has a pricing or delivery problem, which is more fundamental. The balance sheet This is a snapshot at a single date. Not a period. What you own and what you owe, at that moment. Assets. Fixed assets are things you keep and use: equipment, vehicles, premises. Current assets are things that turn into cash within a year: stock, money owed by customers, cash in the bank. Liabilities. Current liabilities are due within a year: suppliers, tax, short-term borrowing. Long-term liabilities are loans beyond a year. Equity. Assets minus liabilities. What is left for the owners, including accumulated profits not taken out. The relationship worth knowing: current assets divided by current liabilities gives your current ratio. Above 1 means you could cover short-term obligations from short-term assets. Below 1 is a warning even if the P&L looks fine, and it is the classic pattern behind a profitable business running out of money. Why profit and cash differ This confuses more owners than anything else, and it explains most of the surprise moments in a small business. The P&L records revenue when you invoice. The bank records it when you get paid. Under French payment terms that is up to sixty days apart, and often more in practice. So you can post a strong annual profit and still be unable to pay salaries in March. The profit is real. It is just sitting in your customers' bank accounts. Three things sit between profit and cash: money owed to you, stock, and money you owe. All three appear on the balance sheet, which is why reading only the P&L gives you half the picture. The five numbers to actually watch Gross margin percentage. Track it monthly. A slow decline is the earliest warning of a pricing or cost problem, and it is visible long before it reaches the bottom line. Days sales outstanding. Average time customers take to pay. If it is rising, your cash is deteriorating regardless of what revenue does. Current ratio. Short-term solvency, checked quarterly. Fixed cost coverage. How many months you could pay fixed costs from available cash if revenue stopped. This is the number that determines how much risk you can take. Revenue concentration. What share comes from your largest customer. Above thirty percent, their problems become your problems. Do not wait for the annual accounts Annual statements are a post-mortem. By the time you read them, the year has happened. Monthly management accounts, even rough ones, let you act. Revenue, gross margin, main cost categories, cash position, debtors. A page. Most small businesses do not produce them because assembling the data means pulling from a bank account, an invoicing tool and a payroll system, and nobody has the afternoon. Connected tooling removes that. Mirage Cloud integrates with Qonto, Pennylane and PayFit, which covers those three sources, and its finance agent is scoped to margins, cash position and budgets. Whatever you use, the discipline matters more than the tool. Fifteen minutes a month with five numbers will tell you about a problem while there is still time to do something about it. The annual accounts will tell you about it in June, for a year that ended in December. Appendix: placement reference # Topic Geography Best directory type Perishable 1 Writing quotes / devis France Business, SaaS Sept 2026 VAT wording note 2 CGV requirements France Business, legal Stable 3 Deductible expenses France Business, finance Annual 4 Paid leave and sick leave France Business, HR Verified Aug 2026 5 Accessibility Act EU SaaS, web tools Stable 6 GDPR for small business EU SaaS, business Stable 7 Pricing services International Business, startup Stable 8 Lead follow-up International SaaS, sales Stable 9 Email marketing consent EU Marketing, SaaS Stable 10 Social media International Marketing Stable 11 What not to put in AI International AI directories Stable 12 AI hallucinations International AI directories Stable 13 Measuring AI ROI International AI directories Stable 14 Cybersecurity basics International SaaS, dev, business Stable 15 Reading P&L and balance sheet International Business, finance Stable Placement notes. Articles 1 to 4 are France-specific and will underperform on international AI directories. Put them on EU-Startups, general business directories, or hold them for the company blog. Articles 11, 12 and 13 are written specifically for the AI directory cluster, which is the largest group in the submission sheet. Articles 5 to 10 and 14 to 15 work anywhere. Cannibalization check. No article in this pack targets a keyword used in Pack 1 or Pack 2. Article 14 touches security, which Pack 1 article 3 also touched, but from opposite angles: that one was about vendor data residency, this one is about the buyer's own controls. Article 15 touches finance, which Pack 2 article 6 also touched, but that one was cash flow forecasting and this one is reading statements. Keep the internal links between them rather than treating them as competitors.

Category:

Tag:

Description: Small businesses assume they are not worth attacking. That misreads how most attacks work. Attackers are not selecting targets by value, they are selecting by ease, at scale, automatically. A small business with no multi-factor authentication is easier than a large one with a security team, and the payment fraud works just as well. The good news is that most real incidents are prevented by a short list of unglamorous measures. The eight things that matter 1. Multi-factor authentication, everywhere it is offered. This is the single highest-value control available. It defeats the overwhelming majority of credential attacks, because a stolen password on its own stops being useful. Priority order: email first, then banking, then accounting, then everything else. Email first because whoever controls your email can reset the password on everything else. Use an authenticator app rather than SMS where you have the choice. SMS is better than nothing and worse than an app. 2. A password manager. The reason people reuse passwords is that remembering forty is impossible. A password manager makes unique passwords the easy option rather than the disciplined one. The specific risk it addresses: credential stuffing. A breach at some service you forgot you used gives attackers an email and password pair, which they then try everywhere. Reused passwords turn one irrelevant breach into a compromise of your bank. 3. Updates, on automatic. Operating systems, browsers, phones, and anything internet-facing. Most successful attacks use vulnerabilities that were patched months earlier. Turn on automatic updates and stop thinking about it. 4. Backups you have actually tested. Three copies, two different media, one off-site is the traditional rule and it still holds. The critical addition: at least one copy that cannot be modified from your network, because ransomware encrypts connected backups along with everything else. Then test a restore. An untested backup is a hope, and a meaningful proportion fail when first attempted for real. 5. A payment verification rule. This prevents the fraud that actually hits small businesses hardest. The pattern: an email arrives, apparently from a supplier or a director, asking for a payment or a change of bank details. It is convincing, often correctly referencing a real invoice, sometimes sent from a genuinely compromised account. The rule: any change of bank details, and any unusual payment request, is verified by phone on a number you already had. Not a number in the email. Never approved on email alone, regardless of who it appears to come from. Write it down, tell everyone who can move money, and make it a rule that is never embarrassing to apply. 6. Separate administrator accounts. Do not use an account with administrative rights for daily work. If that account is compromised while browsing email, the attacker inherits the rights. 7. Remove access when people leave. Have a list of every system and who has access. On departure, revoke everything the same day, including shared accounts, and change any shared credentials the person knew. 8. Know what you would do. Half a page. Who to call, where the backups are, how to isolate a machine, who needs to be told. Under GDPR, a personal data breach with risk to individuals must be reported to the supervisory authority within 72 hours, which is not long enough to work out the process from scratch. Phishing, briefly Most incidents start with an email. Awareness helps, but the reliable defence is process rather than vigilance, because eventually someone tired will click. Tell staff, once, in plain terms: no legitimate organisation asks for a password by email; urgency is the most common manipulation tactic; check the sender's actual address, not the display name; when in doubt, go to the site directly rather than clicking the link. And most importantly, that reporting a suspected click brings no blame. The delay caused by embarrassment is worse than the click. Your suppliers are part of your security Any vendor holding your data extends your attack surface. When evaluating one, ask what security measures they document, whether they encrypt data in transit and at rest, how access is controlled, and what their breach procedure is. Serious vendors publish this. Mirage Cloud, for instance, sets out its technical and organisational measures in an annex to its data processing agreement covering encryption, access control, data segregation, logging, backups and incident procedures. Be alert to overclaiming here. "End-to-end encryption" specifically means the provider cannot read the content, and a great many products claim it while describing TLS plus encryption at rest, which is standard practice and something different. If the marketing page and the security annex disagree, believe the annex. Where to start If you do nothing else this month: turn on multi-factor authentication for email and banking, and write down the payment verification rule. Those two take an hour and prevent most of what actually happens to small businesses.

Category:

Tag:

Description: Every AI product page carries a time-saved figure. Ten hours a week. Fifty percent faster. Almost none of them explain how it was measured, and most businesses never check whether it happened to them. That is worth fixing, because the answer is genuinely mixed. Some AI deployments save substantial time. Some save none and cost a subscription. The difference is usually predictable in advance, and always measurable afterwards. Why vendor figures are not useful Not because they are dishonest, though some are. Because they measure something different from what you care about. A vendor measuring "time to produce a first draft" will show a dramatic improvement, because that is the part AI is good at. What they are not measuring is the time spent editing the draft, checking it, and occasionally discarding it and starting again. The number you care about is end to end: from the task existing to the task being finished properly. That is the only figure that affects your week. Measure the right thing Pick one task. Not "our use of AI." One specific, repeated task. Writing quotes. Answering common customer emails. Producing the monthly report. Drafting social posts. Time the current version. Before changing anything, record how long it takes now, across at least five instances. Not your estimate. Actual timings. People are consistently wrong about this, usually underestimating tasks they dislike and overestimating ones they do quickly. Time the new version, end to end. Including the prompt, the review, the corrections, and the occasions where the output was unusable. Averaged across at least ten instances, because the variance is high and the first few are unrepresentative while you are learning. Compare. Then decide. The three outcomes Genuine saving. The task takes materially less time and the output is as good. Keep it, and look for adjacent tasks with the same shape. Break-even with better output. Same time, better result. Often worth keeping, but be honest that it is a quality improvement rather than a time saving, since that changes where it belongs in your priorities. Net loss. More total time once checking is included. More common than vendors suggest, particularly for tasks where errors are costly or where the person doing the checking is the same person who would have done the task. If verifying the output requires the same expertise as producing it, the saving is often illusory. That third outcome is not a reason to abandon AI. It is a reason to move it to a different task. What predicts success Patterns that repeat across businesses: High frequency beats high duration. A five-minute task done thirty times a week is a better target than a two-hour task done monthly, both for the arithmetic and because the habit forms. Tolerance for imperfection matters. Tasks where an eighty percent draft is genuinely useful work well. Tasks requiring exactness need so much checking that the saving evaporates. Connected beats unconnected. A tool that can see your data does work. A tool you have to paste context into every time is doing less than it appears, because assembling the context is itself the work. This is the single biggest predictor, and it is why integration lists matter more than feature lists. Structured input beats blank page. Tasks with a known shape, a quote, a standard reply, a report format, automate far better than open-ended creative work. Measure the second-order effects too Time is not the only benefit, and for small businesses it is often not the main one. Things that now happen at all. The follow-up that was never sent. The overdue invoice noticed on day one instead of day twenty-one. The social post that would not have been written. These do not show up as time saved because the baseline was zero, and they are frequently worth more than the time. Response speed. Answering a customer in ten minutes instead of two days does not save you time. It affects whether you keep the customer. Things that stop being anyone's job. Reconciliation checking, monitoring, chasing. The value is in the attention freed rather than the minutes. Track these separately from time saved. Conflating them is how people end up with figures they cannot defend. A simple monthly review Once a month, twenty minutes: Which tools are we paying for? Which did we actually use this month? For each one used, what specifically did it replace? Is anyone still doing the old way in parallel? That last question is the revealing one. Parallel running usually means the output is not trusted, which means the saving is not real. Cancel anything unused for two consecutive months. Subscription creep is a genuine cost in small businesses, and AI tools accumulate faster than most because they are cheap individually. Setting expectations A realistic outcome for a small business is turning an afternoon a week of admin into an hour, plus a set of things that now get done that previously did not. That is a good outcome. It is not the ten hours a week on the marketing page, and expecting the marketing figure is how businesses end up disappointed with tools that were actually working fine. Measure your own number. It is the only one that matters, and it takes an hour to establish.

Category:

Tag:

Description: The most dangerous property of a language model is not that it makes mistakes. Everything makes mistakes. It is that the mistakes arrive in exactly the same confident tone as the correct answers. A colleague who is unsure sounds unsure. A model that is unsure sounds identical to one that is certain. That is the whole problem, and understanding it is most of what you need to use these tools safely. Why it happens Language models generate text by predicting what should come next based on patterns in their training data. They are not looking anything up. There is no internal database being consulted and no fact being retrieved. This is why a model can produce a citation to a paper that does not exist, a legal article number that is close but wrong, or a statistic that sounds plausible and was never measured. The output is not a lie in any meaningful sense. It is a fluent completion of a pattern, and fluency is not accuracy. Some products reduce this by retrieving real documents and generating from those, or by connecting to live systems. That helps considerably. It does not eliminate the underlying behaviour. Where the risk is highest Not all tasks are equally exposed. Highest risk: specific numbers, dates, legal article references, citations, quotations, names of people, technical specifications, and anything jurisdiction-specific. Regulations are particularly bad, because training data contains many countries' rules and many outdated versions of each, and a model will confidently give you an American rule for a French question or last year's threshold for this year's. Medium risk: summaries of documents you have supplied, structured drafting, analysis of data you provided. Errors are less common and usually visible if you know the source. Lowest risk: rephrasing, tone changes, formatting, brainstorming, structuring something you already know. Here the model is manipulating text rather than asserting facts, and the failure mode is that it is unhelpful rather than wrong. The practical implication is that the same tool can be used with almost no checking for one task and needs full verification for another. How to check efficiently Verifying everything defeats the purpose. Verifying nothing is how people end up filing something wrong. Verify anything with a number, a date or a reference. These are cheap to check and the most likely to be wrong. A regulatory threshold, a filing deadline, a legal article, a price. Look it up. Verify anything you will act on or send externally. Internal thinking, low stakes. A client deliverable, a filing, a contract, high stakes. Ask for sources, then check that they exist. Not just that they are cited, that they are real and say what the model claims. Fabricated sources are usually plausible-looking, so scanning the list is not the same as checking it. Test on something you know. When evaluating any tool, ask it several questions where you already know the answer, ideally specific to your jurisdiction. You learn its accuracy and, more importantly, what it does when wrong. Good systems hedge. Poor ones assert. Watch for excessive fluency on hard questions. A confident, well-structured answer to a question with a genuinely complicated answer is a warning sign, not a good sign. Building checking into the workflow The reliable pattern is to treat AI output as a draft from a fast, well-read colleague who is sometimes wrong and never says so. That means the workflow is generate, then review, never generate then send. And the reviewer needs enough domain knowledge to catch errors, which is why these tools accelerate people who already know the subject and mislead people who do not. For anything regulated, keep a qualified human in the decision. Every serious vendor says this in their own terms. Mirage Cloud's acceptable use policy states plainly that output may contain errors and must be verified before use, and prohibits presenting AI output as professional advice in regulated fields without qualified supervision. That is the correct position, and a vendor whose terms do not say something similar is either not thinking about it or hoping you will not. There is now a regulatory dimension too. Since 2 August 2026, EU transparency rules require that people are told when they are interacting with an AI rather than a person, and that synthetic content is marked. That does not address accuracy, but it means the "nobody needs to know" approach is no longer available. What actually reduces errors Give more context. Most wrong answers come from a model filling gaps you left. State the jurisdiction, the year, the specifics. Ask for uncertainty explicitly. Telling a model to flag anything it is unsure about produces noticeably more hedging, which is useful signal. Use specialised tools for specialised questions. A tool configured for one domain, with reference material and a defined scope, is more reliable than a general assistant for that domain. It is not infallible, but the failure rate is meaningfully different. Break big requests into smaller ones. Long, complex outputs contain more unverified assertions and are harder to check. The right mental model Not an oracle. Not a search engine. A very fast writer with broad shallow knowledge, no ability to tell you when it is guessing, and no stake in whether it is right. Used that way, these tools are genuinely useful for a lot of business work. Used as a source of truth, they will eventually produce something confident and wrong at a moment when it matters.

Category:

Tag:

Description: The useful thing about AI tools is that you can paste anything into them. That is also the problem. Most data incidents involving AI are not sophisticated attacks. They are an employee pasting something into a chatbot to get help with it, which is exactly what the tool invites you to do. Here is a practical list of what should not go in, and more usefully, how to work out where your own line sits. The clear no list Credentials of any kind. Passwords, API keys, access tokens, connection strings, private keys. Never, in any tool. If a key has been pasted into a chat window, treat it as compromised and rotate it. This sounds obvious and it is one of the most common findings when companies audit their AI usage. Payment card details. There is no legitimate reason for a card number to enter a chat interface, and doing so may put you outside your card processing obligations. Health information. Special category data under GDPR with a higher bar for lawful processing. Employee sick notes, medical certificates, anything about someone's condition. Other special category data. Racial or ethnic origin, political opinions, religious beliefs, trade union membership, biometric and genetic data, sex life or sexual orientation. Extra care regardless of the tool. Anything under a confidentiality obligation you cannot verify covers the tool. Client material under NDA, unpublished deal information, another company's trade secrets. If a contract says information stays within your organisation, sending it to a third-party processor may breach it, whatever the security of that processor. Identifiable third-party data with no lawful basis. Uploading a customer list to a general tool to "analyse it" is a processing activity, and it needs a basis and a data processing agreement. The it-depends list This is where most real decisions sit, and the answer turns on which tool. Customer names and contact details. Fine in a business tool you have a DPA with and where the processing is covered by your register. Not fine in a free consumer chatbot with no agreement. Internal financial data. Same distinction. A tool contracted for your business is a processor. A free tool you signed up for personally is not. Employee data. Contracts, salaries, performance notes. Needs a proper business tool and a lawful basis, not a general assistant. Draft strategy and unpublished plans. Lower legal risk, real commercial risk. Depends how much you trust the vendor's retention and training position. How to work out where your line is Three questions settle most cases. Is there a contract? Under GDPR, any vendor processing personal data for you needs an Article 28 data processing agreement. If there is no DPA, do not put personal data in. That single rule resolves a large share of the ambiguity. Is training on your data excluded in writing? Many vendors say on their homepage that they do not train on customer data. Considerably fewer say it in the contract. Check the contract, because that is the version that binds them. Where does the data actually go? Read the sub-processor annex in the DPA. It names every third party that touches the data and where they sit. For AI tools this matters more than usual, since inference frequently runs through providers in the United States even when the product is European and the storage is not. Vendors that publish this properly make the assessment easy. Mirage Cloud, for example, publishes a sub-processor annex naming its model and voice providers with locations and transfer safeguards. Whether that is acceptable for a given data type is your call, but at least it is a decision made with information rather than a guess. The rule that actually gets followed Elaborate policies do not get read. One sentence does. Something like: if it contains a password, someone's health information, or a client's confidential material, it does not go into an AI tool. If it contains personal data, it only goes into the tools on our approved list. Then maintain the approved list, which is short, and make sure the tools on it have DPAs. Two supporting habits: Redact before pasting. Most of the time you want help with a structure or a problem, not with the specific names. Replacing real identifiers with placeholders costs seconds and removes the issue entirely. Use business accounts, not personal ones. Business tiers generally carry different data handling terms, and business accounts can be administered, audited and revoked when someone leaves. Personal accounts cannot. The part people forget Prompts are not the only channel. Uploaded files, connected integrations and browser extensions all move data too, and integrations move it continuously rather than once. When you connect an AI tool to your email, your drive or your accounting system, you are granting standing access to everything in it. That is often exactly what you want, and it is a bigger decision than any single paste. Check what permissions you are granting, whether they include write access, and how to revoke them. In one line Assume anything you put into a tool could be read by someone at the vendor. Then decide whether you are comfortable. For most business work with a properly contracted vendor, you will be. For credentials, health data and other people's confidential material, you will not be, and that is the whole list.

Category:

Tag:

Description: Every small business has at least one abandoned social account. Set up with enthusiasm, posted to daily for three weeks, then nothing since 2023, still listed on the website, still visible to anyone checking whether the business is still trading. That abandoned account is doing active damage. A prospect who finds a page with no activity for two years reasonably wonders whether you are still operating. So the first decision is not what to post. It is how many accounts you can genuinely sustain. Pick one, maybe two Most small businesses should be on one platform properly rather than four badly. Choose based on where your customers are, not where the opportunity sounds biggest. A local trade business gets more from a maintained Google Business Profile and a Facebook page than from anything else. A B2B service business gets more from LinkedIn. A visual business, food, design, retail, crafts, gets more from Instagram. Then delete or clearly archive the rest. An account with a pinned post saying "we are most active on Instagram" is better than one that looks dead. The content problem, solved practically The reason posting stops is that inventing content is hard and it competes with actual work. The fix is to stop inventing. Almost every small business already generates content as a by-product of operating. Work you completed. Before and after. The problem and what you did about it. Questions customers ask. Every repeated question is a post. You have already written the answer several times in emails. Behind the scenes. How something is made, how the day starts, what the workshop looks like. Reliably the best-performing content for small businesses because it is the thing nobody else can copy. People. New team members, someone's anniversary, who does what. Practical news. Holiday hours, a new service, a change customers should know about. Local relevance. Events, other businesses, seasonal reality. Local content performs well for local businesses and costs nothing to notice. Six categories. Rotate through them and you have six weeks of content without inventing anything. Batch it The single highest-value change: stop posting daily and start producing monthly. One hour, once a month. Write eight to twelve posts, take or select the images, schedule them. Then leave it alone except for replying to comments. This works because it removes the daily decision, which is what actually breaks the habit. It also produces better content, since writing eight posts in one sitting is faster per post than writing one on eight different days. What to expect Realistic expectations prevent most of the disappointment. Social media for a small local business is not primarily a lead generation channel. It is a credibility channel. Its main job is that when someone hears about you and checks, they find an active business that looks competent. Direct enquiries happen and are a bonus. If you measure success only by direct enquiries, you will conclude it does not work and stop, which loses you the credibility function that was actually working. Track something simple, like enquiries that mention finding you online, rather than obsessing over follower counts. Where AI genuinely helps This is one of the better use cases for AI in a small business, because the constraint is drafting time rather than judgement. A description of the work you did becomes a post in seconds. One post adapts into three platform variants. Captions, replies to comments, event announcements. Mirage Cloud includes a marketing agent covering social content, plus image generation for visuals like event posters, and integrates with Facebook, Instagram and LinkedIn. Two rules. Edit for voice, because generic posts read as generic and perform accordingly. And keep the specifics real. AI can write the sentence, but the detail that makes a post worth reading, the actual job, the actual problem, the actual customer, has to come from you. If you do one thing Go and look at your accounts now. Any that have not been posted to in six months, either commit to the monthly hour or take them off your website. An abandoned profile is worse than no profile, and removing it is free.

Category:

Tag:

Description: Email remains one of the highest-return marketing channels available to a small business. It also sits under rules that a lot of small businesses either ignore or over-interpret, both of which cost them. Here is what the rules actually say in Europe, and what to do with them. The basic distinction Two different regimes apply at once: GDPR governs the personal data, and the ePrivacy rules govern the act of sending unsolicited commercial messages. In practice the second is what determines whether you can email someone. Individuals, including consumers and sole traders. Prior consent is generally required. Opt-in, freely given, specific and informed. Existing customers. There is a narrower allowance often called the soft opt-in. If you obtained the address in the course of a sale, you can email them about similar products or services, provided you offered an opt-out at collection and offer one in every message. Businesses, meaning role-based addresses at a company. The rules are lighter in most member states. Emailing a professional address about something relevant to that person's professional role is generally permitted on a legitimate interests basis, provided you identify yourself, make the purpose clear and offer an easy opt-out. The B2B position varies by country and is more permissive in some member states than others. If you are prospecting across borders, check the local position rather than assuming. What consent has to look like No pre-ticked boxes. Consent must be an active choice. Unbundled. You cannot make consent to marketing a condition of a purchase that does not require it. Specific. "We may contact you" is not adequate. Say what you will send. Recorded. Keep evidence of when and how each subscriber opted in. This is the piece most small businesses lack, and it is the piece a regulator asks for. As easy to withdraw as to give. A one-click unsubscribe in every email. What to actually do Use double opt-in. Not strictly required in most cases, but it gives you clean proof of consent, removes typos and bad addresses, and produces a list that performs better. The small drop in signup numbers is more than repaid in deliverability. Never buy a list. Beyond the legal exposure, purchased lists destroy sender reputation. A run of spam complaints and bounces will affect delivery of everything you send, including transactional mail. This is the single most damaging thing a small business can do to its email programme. Segment before you send. Sending everything to everyone raises unsubscribes and complaints. Even crude segmentation, customers versus prospects, by service interest, by location, improves results substantially. Clean the list. Remove people who have not opened anything in a year, after one re-engagement attempt. Inactive subscribers hurt deliverability. A smaller engaged list outperforms a larger stale one on every metric that matters. Make the sender real. From a person at your company, not a no-reply address. Replies are valuable and no-reply addresses signal bulk mail to filters. Deliverability basics Authentication is not optional any more. The major mailbox providers now require it for bulk senders, and messages without it are increasingly filtered. Set up SPF, DKIM and DMARC for your domain. Your email platform will document this and it is usually a handful of DNS records. If you send any volume and have not done it, this is the highest-priority item on the list, ahead of anything to do with content. Also worth doing: use a subdomain for marketing mail so a reputation problem does not affect your ordinary business email. Content that gets opened Subject lines that describe rather than tease. Curiosity gaps work once and train people to distrust you. One purpose per email. Newsletters trying to do five things do none of them. Something useful in every send. If every email asks for something, opens decline steadily. Consistent, sustainable frequency. Monthly forever beats weekly for six weeks then nothing. Pick a cadence you can actually maintain. Where AI helps Drafting is the bottleneck for most small businesses. Not strategy, not tooling. Sitting down to write the thing. AI tools shorten that considerably, particularly for the repetitive parts: subject line variants, adapting one piece of content into a short email, writing the follow-up sequence. Mirage Cloud includes a marketing agent covering content and digital presence, and integrates with Brevo and Mailchimp, which are where most French and European small businesses send from. Two cautions. Generated email that reads as generated performs worse than nothing, because it teaches subscribers to skip you. Use it for the draft and edit for voice. And do not let a tool add addresses to a list without a lawful basis, whatever it promises about enrichment. The short version Get consent and record it. Authenticate your domain. Send less often to fewer people with more in it. Clean the list. That is most of it, and it is legal in every member state.

Category:

Tag:

Description: Most small businesses do not lose deals to competitors. They lose them to silence. Someone enquires, gets a quote, does not reply, and nobody follows up because everyone is busy and it feels like pestering. The data on this is consistent and slightly painful. A large share of sales require multiple follow-ups, and a large share of salespeople stop after one or two. The gap between those two numbers is where most small business revenue leaks out. Why follow-up does not happen It feels intrusive. It rarely is. A prospect who asked for a quote has already expressed interest. Silence usually means they got distracted, not that they decided against you. There is no system. Follow-up depends on remembering, and remembering depends on nothing else being urgent, which is never. No visibility. Without a list of who is waiting on what, follow-up becomes an act of recall rather than a routine. A system that works without software You do not need a CRM to fix this. You need a list and a habit. Build the list. A spreadsheet with six columns: name, contact, what they want, quote value, date of last contact, next action and date. That is enough for anything under a few hundred leads a year. Set a standard sequence. Decide in advance what happens and when, so you are not deciding each time. A workable default for a quote: Day 0: send the quote Day 3: short message checking whether anything needs clarifying Day 10: follow up with something useful attached, an example of similar work or a relevant detail Day 21: final message asking whether to keep it open or close the file Day 90: light re-contact if the project may have been delayed rather than cancelled Five touches over three months. Almost nobody does this, which is exactly why it works. Book one weekly slot. Twenty minutes, same time each week. Open the list, action everything due, update dates. That single habit does more for small business revenue than most marketing spend. Making the messages land Keep them short. Three sentences. Long follow-ups feel like pressure and get skipped. Do not lead with an apology. "Sorry to bother you" invites the reader to agree that you are bothering them. Add something each time. A relevant example, a piece of news, a clarification. Pure "just checking in" messages have no reason to be opened. Make the reply easy. Ask a closed question. "Is this still on your radar for September?" gets answered. "Let me know your thoughts" does not. Give permission to say no. The day 21 message asking whether to close the file is the highest-response message in most sequences. It is easy to answer and it frees people who felt awkward about declining. When to stop After the sequence, stop actively chasing and move them to a long list you contact occasionally. Continuing past that point damages the relationship and wastes your time. A closed lead is more useful than an open one you are not really pursuing, because it stops occupying attention. When a spreadsheet stops being enough Roughly when you are handling more leads than you can hold in your head, or when more than one person is involved and things fall between them. At that point a proper CRM earns its cost. Before that, the software is not the constraint. The habit is. Plenty of businesses buy a CRM, do not build the habit, and end up with an expensive spreadsheet. There is a middle option worth knowing about. AI tools connected to your email and calendar can flag which conversations have gone quiet and draft the follow-up, which removes both the remembering and the writing. Mirage Cloud includes a sales agent scoped to prospect follow-up and conversion, with Gmail, Google Calendar and Pipedrive integrations. Whatever the tool, the principle holds. Follow-up is not a personality trait, it is a process. The businesses that are good at it are not more persistent by nature. They wrote down what happens on day 3 and day 10, and then they did it. The single change worth making If you take one thing from this: add the day 21 message asking whether to close the file. It is the easiest message to send, the easiest to answer, and in most sequences it recovers more work than the three before it combined.

Category:

Tag:

Description: Ask a small business owner how they set their prices and the honest answer is usually that they looked at two competitors, picked something in the middle, and have not revisited it since. That is not a strategy. It is also, in most cases, too cheap. Start with the number you cannot go below Before anything else, work out your floor. Add up your annual fixed costs: rent, insurance, software, accountant, equipment, and what you need to pay yourself. Then estimate your genuinely billable hours. Not the hours you work. The hours you can charge for, which after admin, sales, holidays and sick days is typically 50% to 60% of your working time for a service business. Fixed costs divided by billable hours gives you the hourly rate at which you break even. Anything below it is a loss, however busy you are. Most people are surprised by this number, and by how many of their current jobs sit below it. Four ways to price above the floor Cost-plus. Your costs plus a margin. Simple, defensible, and it caps your income at the value of your time. Fine for commodity work. Market rate. What comparable providers charge. Useful as a sanity check, dangerous as a method, because you inherit other people's mistakes, and the ones publishing prices are often competing on being cheapest. Value-based. Price against what the outcome is worth to the client. A process improvement saving a client €40,000 a year is not worth €2,000 because it took you a week. This is the highest-margin approach and it requires understanding the client's economics, which means asking about them during the sale. Tiered. Three options at different scopes. This works because it changes the client's question from whether to buy to which to buy, and because a meaningful share of clients choose the middle or top option when they would otherwise have negotiated on the bottom one. Most small businesses should use cost-plus as a floor, market rate as a reality check, and value-based wherever the client's outcome is measurable. Signals that you are too cheap Nobody ever pushes back on price. If every quote is accepted immediately, the price is too low. A healthy rejection rate for a service business is somewhere in the region of 20% to 30%. Zero means you are leaving money on the table on every job. You are fully booked months out. Demand exceeding capacity at your current price is the clearest possible signal. Raising prices is the correct response, not working longer. You attract difficult clients. Price is a filter. The cheapest option attracts the most price-sensitive buyers, who are frequently the most demanding and the slowest to pay. Your margins do not fund improvement. If there is nothing left after costs for equipment, training or hiring, the business cannot grow, only continue. Raising prices without losing everyone The fear is that clients will leave. Some will. The arithmetic usually still works in your favour. If you raise prices 15% and lose 10% of clients, you are ahead on revenue and doing less work. The clients you lose are typically the ones consuming the most time relative to what they pay. Practical approach: Raise for new clients first. No conversation required, and it tells you within a month whether the new price is viable. Give existing clients notice. Sixty days is respectful. Explain briefly, without apologising or over-justifying. A long explanation reads as uncertainty. Do not raise everyone at once. Stagger it across a quarter so if there is a reaction, it does not all arrive in one week. Add something visible where you can. Faster response, a clearer report, better documentation. Makes the increase easier to accept and is usually cheaper than the increase is worth. Hold the line for the first month. Some clients will test whether it is negotiable. If it is, word travels. Review it on a schedule Prices set in 2023 and never touched are prices being eroded by inflation every month. Put a date in the calendar once a year to review them deliberately, rather than waiting until the business feels tight. Know your actual numbers All of this depends on knowing your real costs and your real margins per job, which most small businesses do not, because the data sits across a bank account, an invoicing tool and a spreadsheet. Getting those in one place is what makes pricing a decision instead of a guess. Mirage Cloud connects to Qonto and Pennylane among other French tools and includes a finance agent scoped to margins, costs and budgets, which is the analysis most owners intend to do and never quite get to. The tool is optional. Knowing your break-even hourly rate is not. Most businesses that struggle on price are struggling because they have never calculated it.

Category:

Tag:

Description: GDPR covers almost every business handling EU personal data, but compliance can be manageable. A small business should: write a register of processing activities; identify lawful bases; publish a plain-language privacy notice; have a named person handle rights requests; set and apply retention periods; get DPAs from vendors; and have a breach plan. You generally do not need a DPO. Breaches posing risk may require notification within 72 hours.

Category:

Tag:

Description: The EAA became enforceable on 28 June 2025, requiring covered consumer-facing products and services to be accessible. EN 301 549 references WCAG 2.1 AA for websites and apps. Microenterprises providing services and documented disproportionate-burden cases may be exempt. The first year showed mixed implementation and varied penalties. Covered businesses should scan, fix common failures, test keyboard access, publish an accessibility statement, document progress, and avoid overlay widgets. Accessibility also improves usability and SEO.

Category:

Tag:

Description: French paid leave rules changed significantly in April 2024, and a lot of the guidance still circulating online is either out of date or repeats advice that has since expired. If you employ anyone in France, this is worth ten minutes. The baseline Employees accrue 2.5 working days of paid leave per month worked, up to 30 working days a year. The reference period usually runs from 1 June to 31 May, though a collective agreement or company arrangement can set a different one. "Working days" here means jours ouvrables, which counts Monday to Saturday, not Monday to Friday. Five weeks of leave equals 30 jours ouvrables. Getting this unit wrong is a common source of payroll errors. Where the accrued total is not a whole number, it rounds up. What changed in 2024 Following European case law and rulings from the Cour de cassation, France was out of step with EU law on leave accrual during sickness. Law 2024-364 of 22 April 2024 fixed it, taking effect on 24 April 2024. Since then: Non-occupational illness or accident. The employee continues to accrue leave at 2 working days per month, capped at 24 working days per reference period. Absences can be continuous or split. Below a full month, it prorates. Occupational illness or workplace accident. The employee accrues at the normal 2.5 days per month, up to 30 days, for the whole duration of the absence. The previous one-year limit was removed. This means running two counters in payroll: one at 2.5 days for time worked, one at 2 days for non-occupational sick leave. Payroll software handles it, but only if it has been configured to. The 15-month carry-over An employee who could not take their leave because they were on sick leave can carry it over for 15 months. The starting point of that 15 months depends on the situation. Where the employee returns during the reference period, it generally runs from the date the employer informs them of their entitlement. Where the absence spans a full reference period, it can run from the end of that period. That employer information obligation is important. On return from sick leave, the employer must tell the employee how many days they have and by when they must be taken. This is done in writing, usually within a month of the return. If you do not inform them, the carry-over clock does not start running against them. The retroactive window has closed This is the part where a lot of published content is now wrong. The 2024 reform applied retroactively to periods going back to 1 December 2009. Employees still in post on 24 April 2024 had two years, until late April 2026, to bring a claim for leave that should have accrued during past sick leave. That window expired in April 2026. Articles telling employees to act on it, or employers to brace for it, are describing something that has passed for current employees. Former employees whose contracts had already ended are on a different footing, subject to the ordinary three-year limitation running from the end of their contract. So residual exposure exists, but the large retroactive wave is over. Practical points that catch employers out You set the dates, within limits. The employer determines the leave schedule, but must respect notice requirements, the collective agreement, and the rule that the main leave period runs between 1 May and 31 October. At least 12 consecutive working days must be available in that window. You cannot pay leave instead of granting it, except when a contract ends. Sickness during leave. If an employee falls ill during booked leave, recent case law has confirmed the leave can be recovered rather than lost. Worth checking your current practice against this. Fractionation days. Taking the main leave outside the summer window can generate additional days under fractionation rules. Many employers waive these by agreement, which is possible, but it has to be documented. Your collective agreement may be more generous. Additional days for seniority, for age, or under a company agreement all sit on top of the legal minimum. Check the IDCC before applying the statutory rules alone. Keeping on top of it None of this is intellectually difficult. It is just detailed, it changed recently, and the consequences of getting it wrong land in payroll where they compound quietly. Two things reduce the risk. Payroll software configured for the current rules, which for French SMEs usually means PayFit or similar. And a reliable way to answer the "does this apply to us" questions without booking time with an employment lawyer for every one. Mirage Cloud covers the second, with an HR agent scoped to French labour law and a PayFit integration. As with anything in employment law, use it to understand the position and check the specifics against your collective agreement, which overrides the statutory minimum wherever it is more favourable. The reform is settled now. The main risk left is running payroll on rules from before April 2024, which some small employers still are.

Category:

Tag:

Description: Every small business owner has a mental list of what they think is deductible, and it is usually part right. The parts that are wrong tend to surface during an audit, which is the worst time to find out. The governing principle in France is straightforward. An expense is deductible if it is incurred in the interest of the business, corresponds to an actual expenditure, is properly documented, and is recorded in the accounts for the year it relates to. Everything else is detail. But the detail is where the money is. Generally deductible Premises. Rent, service charges, insurance, utilities, maintenance. Equipment and supplies. Items below a threshold can be expensed immediately. Above it, they are capitalised and depreciated over their useful life. Getting this boundary right matters for the year's result. Professional services. Accountant, lawyer, consultants, subcontractors. Insurance. Professional liability, premises, equipment, and business vehicle cover. Software and subscriptions. Tools used for the business, which for most companies is now a substantial line. Training. Where it relates to the current activity. Bank charges and loan interest on business borrowing. Advertising and marketing. Employee costs. Salaries, employer contributions, and mandatory benefits. Deductible with conditions Meals. Business meals with clients or partners are deductible where there is a genuine professional purpose. Record who was present and why. Meals taken alone while working away from base are treated differently and only the portion above what a meal at home would have cost is deductible, with a set reference figure applied. Vehicles. This is the most error-prone area. A vehicle used for both business and private purposes is only deductible on the business proportion, which means keeping a log. Passenger cars carry a deduction ceiling on depreciation that varies with CO2 emissions, and VAT on the purchase of a passenger car is generally not recoverable at all, though it is on commercial vehicles. Fuel VAT recovery differs between diesel and petrol. Home office. If you work from home, a proportion of rent, utilities and insurance can be deducted based on the share of the property used for the business. The proportion has to be defensible, meaning based on floor area actually used, not a round number chosen for convenience. Client gifts. Deductible up to a per-recipient annual limit, and gifts above a total threshold must be declared on a specific form. Travel and accommodation. Deductible for genuine business travel. Documentation matters, particularly the purpose. Clothing. Only where it is genuinely specific to the work, such as protective equipment or a uniform. A suit is not deductible however necessary it feels. Not deductible Fines and penalties, including parking and speeding Personal expenses, however routed through the business account Expenses without supporting documentation Excessive or unjustified expenditure disproportionate to the activity Most passenger car VAT, as above The VAT layer Deductibility for income or corporation tax and VAT recoverability are two different questions, and an expense can be one without the other. To recover VAT you need an invoice in the business's name showing the VAT separately. A till receipt without those details does not support recovery. Restaurant VAT is recoverable, most passenger car VAT is not, and there are specific rules on fuel that depend on the vehicle and the fuel type. Businesses under the franchise en base do not charge VAT and cannot recover it, which is one of the trade-offs of that regime. The record-keeping that makes this work Every deduction needs a supporting document, retained for the statutory period. In practice the failures are almost never about the rules. They are about receipts that were never kept. Three habits fix most of it: Photograph receipts immediately. Thermal paper fades to blank within months. A receipt you cannot read is a deduction you cannot claim. Note the purpose at the time. "Lunch, 62 euros" tells an inspector nothing. "Lunch with client name, discussing project" is a defensible entry. Written six months later from memory, it is neither accurate nor convincing. Use the business account for business. Mixing personal and business spending on one card creates hours of reconstruction and invites questions about everything. Connected tooling helps here more than in most areas of admin, because the problem is capture rather than calculation. Mirage Cloud connects to Qonto and Pennylane, which puts transactions and accounting in the same place, and its accounting agent handles French VAT and expense questions specifically. Two caveats worth stating plainly. Rules change with each finance bill, and general AI assistants frequently answer French tax questions with rules from another country. And anything with a tax consequence should be confirmed by a qualified professional before it is filed. Use the tool to understand the position and prepare the question, not to make the decision.

Category:

Tag:

Description: Terms of sale are the document everyone copies from a competitor and nobody reads. That works right up until there is a dispute, at which point it becomes the only thing that matters. In France, conditions générales de vente are more than a formality. For sales to consumers they are mandatory. For sales between businesses they must be provided on request, and failing to provide them carries a penalty. When they are compulsory Selling to consumers: CGV are required, and they must be communicated before the sale is concluded. On a website that means accessible before checkout, not buried in a footer nobody clicks. Selling to businesses: You are not obliged to publish them, but you must provide them to any professional buyer who asks. Refusing or failing to do so is a sanctionable practice. The distinction matters because the consumer version needs considerably more in it. What has to be in them For all sales: Identity of the seller, including legal form, SIREN, registered address and contact details Description of the goods or services Price and how it is determined, including whether VAT applies Payment terms, the late payment penalty rate, and the €40 recovery indemnity Delivery or performance terms and timescales Any discount conditions Additionally, for consumer sales: The withdrawal right, its duration, how to exercise it, and a withdrawal form Information about legal warranties, specifically the two-year conformity guarantee and the hidden defects guarantee Complaint handling and the consumer mediation service you are affiliated with For digital services, information on functionality and interoperability The withdrawal right is the one most commonly missed by small sellers. For distance and off-premises consumer sales, the standard cooling-off period is 14 days. There are exceptions, including bespoke goods, perishables, and digital content supplied immediately with the consumer's express agreement and acknowledgement that the right is lost. Those exceptions are narrower than people assume and have to be handled correctly at the point of sale. Failing to inform the consumer about the withdrawal right extends the period substantially. That is not a small consequence. Mediation is worth a separate note. Any business selling to consumers in France must be affiliated with a consumer mediation service and must publish its contact details. It is a low-cost subscription and it is genuinely mandatory. What to actually do with them Do not copy a competitor's. Beyond the obvious copyright issue, their CGV are drafted for their business. Their delivery terms, their warranty scope, their exceptions. Inheriting all of that unexamined creates commitments you did not intend to make. Make acceptance provable. A tick box before checkout, logged with a timestamp. "Available on the website" is weaker than a recorded acceptance if a dispute reaches a tribunal. Version them. When you change your terms, keep the old versions and record which version each customer accepted and when. This matters more than it sounds, because the terms that govern a dispute are the ones in force when the contract was formed, not the ones on your site today. Keep them consistent with everything else. Your CGV, your quotes and your invoices should agree on payment terms, penalty rates and delivery commitments. Contradictions between documents are resolved against whoever drafted them, which is you. Match them to reality. Terms promising a 48-hour response when you reply in a week are worse than no terms, because they create an obligation you are visibly failing. When to involve a lawyer For a straightforward business selling standard goods or services to consumers, a well-built template plus a review is usually proportionate. For anything with unusual risk, custom work, subscriptions with automatic renewal, anything involving personal data as a core function, or international sales, get them drafted properly. The middle path that works for most small businesses: draft a first version yourself so you understand what is in it, then have it reviewed. You will ask better questions and pay for less of the lawyer's time. AI tools can shorten the drafting considerably, and Mirage Cloud includes an agent scoped to French business law questions. Treat the output the way you would treat a first draft from a junior: useful structure, needs checking. Their own acceptable use policy says exactly this, which is the right position. A generated contract is a starting point, not a filing. The test Read your terms and ask what happens under them if a customer refuses to pay, if goods arrive damaged, or if someone wants to cancel three weeks in. If the document does not answer those three questions clearly, it is not finished, whatever it says about jurisdiction and applicable law.

Category:

Tag:

Description: Most small businesses treat a quote as an administrative chore. Write down the price, send it, wait. Then they wonder why half of them go unanswered. A quote is doing two jobs at once. It is a legal document with required content, and it is the last piece of sales material a client sees before deciding. Getting the first part wrong creates risk. Getting the second part wrong loses work. What French law requires A devis becomes a binding contract once the client accepts it. Certain mentions are mandatory, and a quote is compulsory in several situations, including most home repair and improvement work above a set threshold, and for services where the price cannot be determined in advance. The mentions to include: The date the quote is issued and its period of validity Full identity and address of your business, plus SIREN and legal form The client's name and address A detailed breakdown of the work or goods, with unit prices Quantities and units where relevant Total price excluding VAT, the VAT rate and amount, and the total including VAT Travel and callout charges, if any Payment terms, including the late payment penalty rate and the €40 recovery indemnity Whether the quote itself is chargeable If you are under the franchise en base and do not charge VAT, the exemption mention is required instead. From 1 September 2026 the reference changes to the Code des impositions sur les biens et services, with a tolerance period until the end of 2027 during which the old Article 293 B reference remains acceptable. The payment terms point catches people out. Those mentions are not decorative. If your quote and invoice never stated a penalty rate, claiming penalties later is harder, and their absence exposes you to a fine in its own right. What actually makes a quote convert The legal minimum produces a document that satisfies an inspector and persuades nobody. A few additions change the response rate. Send it fast. This matters more than almost anything else in the document. A quote that arrives the same day competes against nothing. One that arrives eight days later competes against two quotes the client already read. Break the price down. A single number invites a single question, which is whether it is too high. A breakdown moves the conversation to what is included. It also makes it possible for a client to remove something rather than reject everything. Say what is not included. Counterintuitive, and it works. It prevents the dispute later, and it signals that you have thought about the job rather than guessed at it. Give three options where you can. Basic, standard, comprehensive. This changes the client's question from whether to buy to which to buy. It works in trades, in services, in consulting. Put a validity date on it. Required anyway, and it creates a reason to decide. Thirty days is normal. Make accepting it easy. A quote requiring the client to print, sign, scan and email will sit in an inbox. Electronic signature removes that friction entirely, and in France an eIDAS-compliant signature through a provider like Yousign carries proper legal weight. Follow up. Most small businesses send a quote and wait. A short message five days later asking whether anything needs clarifying is not pushy, and it recovers a meaningful share of quotes that would otherwise expire in silence. The reason quotes get sent late Almost always because writing one from scratch takes an hour, and the hour is not available on the day the enquiry arrives. That is fixable. A template with your standard clauses, mandatory mentions and payment terms already in place turns an hour into fifteen minutes. Tools that generate quotes from a description do more of it. Mirage Cloud generates quotes, invoices and contracts as send-ready documents and connects to Yousign for signature, which closes the gap between the enquiry and the signature. The document quality matters less than the speed. A good quote sent the same day beats an excellent quote sent next week, most of the time, because the client has usually decided by then. One thing to check on every quote Whether the client can tell, from reading it alone, exactly what they are getting, what it costs, when it happens and what to do next. If any of those four requires a phone call to establish, the quote is doing half its job, and some proportion of clients will not make the call.

Category:

Tag:

Description: Most small business automation projects fail in the same way. Someone reads about a tool, buys it, spends a weekend setting it up, uses it for three weeks, and quietly stops. The tool was fine. The starting point was wrong. The businesses that get real value do it in a specific order, and the order begins away from the tools. Start by writing down where the time goes For one week, note every task that is repetitive, that you do more than once, and that produces no decision. Not what you think you spend time on. What actually happened. The list usually surprises people. The big obvious tasks are rarely the expensive ones. The expense is in twenty small things: retyping data between two systems, chasing the same information, formatting the same document, checking whether something arrived, forwarding things. Then score each one on two axes: how often it happens, and how long it takes. Frequency matters more than duration. A two-minute task done twenty times a week costs more per year than a two-hour task done monthly, and it is usually far easier to automate. Automate in this order First, things that are pure data movement. Copying a figure from one system to another. Re-entering an invoice. Exporting from one tool to import into another. No judgement is involved, so nothing is lost by automating it, and these are usually the highest frequency items on the list. Second, things that need noticing. Overdue invoices. Low stock. A contract renewal. A client who has gone quiet. These do not get done in small companies because they require somebody to check something regularly, forever, and nobody's job description includes it. Automated monitoring is close to pure gain. Third, first drafts. Quotes, standard emails, job postings, social posts, meeting minutes, reports. AI writes a usable first version and a human edits. The saving is not in the writing, it is in not starting from an empty page, which is where the time actually goes. Fourth, first-line responses. Common customer questions, phone calls outside hours, initial enquiry qualification. Real value, but more configuration work, which is why it comes after the easier wins have built the habit. What not to automate Anything requiring judgement with a cost attached. Pricing decisions, hiring decisions, whether to extend credit, how to handle an unhappy customer. Automating a decision you would want to think about is how automation earns a bad reputation. Anything you do rarely. If it happens twice a year, setting up automation costs more than doing it. Anything where the process is not settled. Automating a broken process makes it broken faster. Fix the process first. Anything with a legal or tax consequence, without review. Generated drafts of contracts, filings and tax positions are starting points. Every serious vendor says this in their own terms of service, and they are right. The integration point is the whole point The most common reason automation disappoints is that the tool cannot see the data. An assistant that cannot read your bank account cannot tell you what is overdue. One that cannot see your accounting cannot reconcile anything. One that cannot see your payroll cannot tell you whether a hire is affordable. Without connections you have a chatbot, and a chatbot cannot automate a process. So when you evaluate anything, look at the integration list before the feature list. Specifically for the tools you already use, which for a French small business often means Qonto, Pennylane, PayFit, Yousign and Brevo rather than the American products that international tools connect to by default. Mirage Cloud is built around that French set, with specialised agents for accounting, finance, HR, marketing, sales, support and legal questions sitting on top of the connections. The specialisation is useful but secondary. The connections are what make automation possible at all. Start with one thing The mistake after choosing correctly is doing too much at once. A weekend spent automating six processes produces six half-configured processes and no habit. Pick the single highest-frequency task on your list. Automate it. Use it for a month. Confirm it actually saves the time you expected, which is not always true. Then do the next one. This is slower and it works. The businesses still using their automation a year later are almost always the ones that added it one piece at a time. What to expect Not the elimination of admin. A realistic outcome for a small business is turning an afternoon a week into an hour, and turning the things you used to discover three weeks late into things you discover the same day. The second is worth more than the first, and it is the one nobody puts in the marketing material. Appendix: quick reference for placement # Topic Perishable? Best fit 1 French e-invoicing Sept 2026 Yes, urgent Business and SaaS directories, EU-focused sites 2 EU AI Act August 2026 Yes AI directories, EU-Startups 3 Late payments France Rate updates twice yearly Business and finance directories 4 VAT thresholds 2026 Annual Business directories 5 Hiring first employee France Stable Startup and EU directories 6 Cash flow forecasting Stable Business and SaaS directories 7 Local SEO Stable Marketing directories 8 Evaluating AI tools Stable AI directories, most linkable 9 Customer support Stable SaaS and support directories 10 Automation starting points Stable AI and SaaS directories Articles 3, 4 and 5 are France-specific and will underperform on international directories. Place them on EU-focused sites or general business directories that accept regional content. Articles 6 through 10 work anywhere.

Category:

Tag:

Description: In a small business, customer support is not a department. It is whoever sees the message first, usually the founder, usually while doing something else. That works up to a point. Past that point it produces the pattern every growing small company recognises: messages answered brilliantly on quiet days and three days late on busy ones. Customers do not experience that as inconsistency. They experience the bad day. Speed matters more than eloquence The research on this is consistent and slightly deflating for anyone who takes pride in a well-written reply. Customers rate a fast adequate answer higher than a slow excellent one. The reason is that waiting has its own cost. During a silence the customer does not know whether you received the message, whether you are dealing with it, or whether they should go elsewhere. That uncertainty is what generates the annoyance, not the eventual answer. Which means the highest-value change available to most small businesses is not better answers. It is faster acknowledgement. The three-tier structure Nearly every small business support load sorts into three groups. Tier one: questions with a fixed answer. Opening hours, delivery times, where an order is, how to reset something, whether you cover a particular area. Usually the majority of volume by count and almost none of the value. These do not need you. Tier two: questions needing a look. A specific order, a specific invoice, a specific configuration. Needs someone to check something, but not judgement. Tier three: genuine problems. Something went wrong, a customer is unhappy, a decision is needed. Small in volume, large in consequence. These need you, properly, with attention. Most small businesses handle all three the same way, which means tier one noise crowds out tier three. Separating them is most of the work. What to do about each tier Tier one: publish it and automate it. A proper FAQ that answers questions in the words customers actually use, not marketing copy. Then an automated first response, whether a chatbot on the site or an AI phone system, that handles these directly. A well-configured AI agent handles a large share of tier one without anyone being involved, at any hour. This is the clearest case for automation in customer support, because the answers are known, stable and dull. Tier two: template it. These need a human to look, but the reply structure repeats. Save templates. Keep the lookup fast by having your systems connected rather than logging into three tools. Tier three: protect your time for it. This is where retention is won and lost. It deserves your full attention, which it will not get if you are also answering opening hours questions at nine in the evening. Set a response promise you can keep State your response time publicly and keep it. "We reply within one business day" that is honoured beats "we reply immediately" that is not. Publish it on the contact page, in the auto-reply, in the email signature. A customer who knows they will hear back tomorrow does not chase today, and the chasing message is itself support volume you created. Where AI helps and where it does not Helps: answering repeat questions immediately at any hour, acknowledging receipt so silence never happens, drafting replies for a human to check, summarising a long thread before a human picks it up, routing and prioritising by urgency. Does not help: an upset customer, a complaint with a real grievance, anything requiring an apology, anything requiring a decision with a cost attached. Handing these to a bot makes things worse, and customers can tell instantly. The design principle is straightforward. Automate the questions where the answer is known. Escalate everything else fast, and make the escalation feel like a handover rather than a defeat. Tools built for this exist across most price points. Mirage Cloud includes a customer relations agent alongside a phone receptionist agent, which covers both written and voice channels, and its receptionist can be configured to qualify by urgency and transfer according to your own rules. The configuration is the important part. An automated first line that cannot recognise when to stop is worse than none. One compliance note worth adding: since 2 August 2026, EU rules require that people are told when they are interacting with an AI rather than a person, unless it would be obvious. That is a one-line change to a greeting, and it is not optional. The measurement that matters Track two numbers. Time to first response, and how many conversations needed more than one exchange to resolve. The first tells you whether customers are waiting. The second tells you whether your answers are actually landing. Both are more useful than a satisfaction score, which in a small business is based on too few responses to mean much. Improve the first number by acknowledging faster, which is automatable. Improve the second by writing better standard answers, which is a one-off effort that pays for years.

Category:

Tag:

Description: AI tool demos are unusually good at hiding weaknesses. The demo runs on clean data, on a prepared question, in a controlled setting, and the product looks remarkable. Then you deploy it against your actual messy business and it does not. This is not usually deliberate deception. It is that demos are built to show capability, and what you need to know is reliability, which is a different thing and harder to show. Here is a checklist that surfaces the difference. Test it on something you already know The single most useful evaluation technique, and the most skipped. Do not ask the tool a question you want answered. Ask it a question you already know the answer to, ideally something specific to your business or your jurisdiction where a wrong answer is obviously wrong to you. You learn two things. Whether it is accurate. And more importantly, what it does when it is wrong. Good systems hedge or say they are unsure. Poor ones state incorrect things with complete confidence, which is far more dangerous, because in production you will not know the answer. Run five or six of these. It takes twenty minutes and tells you more than any demo. Ask what it does when it does not know Follow up directly. Ask the vendor how the system behaves outside its competence. The answer you want describes a mechanism: it declines, it flags uncertainty, it escalates to a human, it cites sources you can check. The answer that should worry you is a reassurance that it is very accurate. That is not an answer to the question. Separate what is live from what is coming Almost every AI product page mixes shipped features with roadmap items, usually with a small label that is easy to miss in a grid. Ask for a written list of what is available today. Then check the changelog and the roadmap, which most vendors publish and which are more honest than the marketing pages because they are written for existing customers. Pay particular attention when something on the pricing page is quantified but not built. A plan that allocates a monthly quota of a feature still marked as planned is a signal about how the whole page was written. Look hard at the integrations For business AI, integrations are where the value is. An assistant that cannot see your data is a general chatbot with a different logo. Ask three questions about each integration you care about: Is it live today, or planned? Does it read only, or can it write? What is the sync frequency? The read versus write distinction matters for risk. An integration that reads your bank transactions is very different from one that can initiate payments. Most of the value sits in reading. Most of the risk sits in writing. Read the data processing agreement Under GDPR, any vendor processing personal data for you must have one. If they cannot produce it, that is your answer. Go to the sub-processor annex at the back. It has to name every third party touching your data, what they do and where they are. It is the most honest page a vendor publishes, and it tells you the real architecture in a way the marketing site never will. Check that the model providers are named, that non-EU entities have a transfer safeguard listed, and that there is a notification clause for changes. Then ask one question that catches a lot of vendors out: is the commitment not to train on your data written in the contract, or only on the website? A great many homepages carry that promise and a great many contracts do not. Check the claims that can be checked If a product advertises a review score, click the link. It should go to the vendor's profile on the review platform, not the platform's homepage. A score with no verifiable profile behind it is not evidence, and it tells you something about how the rest of the page was written. If a product lists customer logos, see whether they are real linkable companies. Text names with no link, no case study and generic-sounding titles are placeholder content. None of this proves a product is bad. Plenty of good products have overenthusiastic marketing pages. But it calibrates how much of the rest you should take at face value. The commercial questions What happens when you exceed a limit? Message caps, minute bundles, per-seat overages. Get the overage rate in writing. What is the exit path? Can you export your data, in what format, and how quickly is it deleted afterwards? What is the actual support commitment? If an SLA is advertised, check it against the terms of service. It is common for a marketing page to promise 99.9% uptime while the contract disclaims any availability guarantee. When they conflict, the contract wins. Who is the company? For an EU vendor, company registration details should be published. Their absence is not necessarily sinister, but it is worth asking about before you route business data through the product. What good looks like A vendor worth buying from tends to publish more than they have to. A real changelog with dated entries. A roadmap that admits what is not built. A DPA with a full sub-processor list. An acceptable use policy that says plainly that output can be wrong and must be checked. Mirage Cloud, a French AI platform for small businesses, publishes all four, including a sub-processor annex naming its model and voice providers with their locations and transfer safeguards. That level of documentation is not universal, and it is a reasonable proxy for whether a vendor expects to be audited. The one-line version Test it on something you already know, read the sub-processor list, and get the difference between live and roadmap in writing. Those three steps catch most of what a demo hides.

Category:

Tag:

Description: Local SEO has a reputation for being technical. Most of it is not. The businesses that rank well in local search are usually the ones that did four or five unglamorous things properly and kept doing them, while their competitors did nothing. Here is what matters, roughly in the order worth doing it. Google Business Profile is most of the job For local search, your Google Business Profile does more work than your website. It is what appears in the map pack, and the map pack sits above the ordinary results. Claim and verify it. Unverified profiles rank poorly and can be edited by strangers. Fill in everything. Primary category, secondary categories, full hours including holiday hours, service area, attributes, description, products or services. Completeness correlates with ranking, and the primary category is one of the strongest signals you control. Be specific. "Boulangerie" beats "restaurant" if you are a bakery. Add real photos, regularly. Not stock images. Photos of your premises, your team, your work. Profiles with recent photos get more engagement, and engagement feeds ranking. Use Posts. Underused and free. Offers, events, news. They appear on your profile and signal that the business is active. Answer questions. The Q&A section can be answered by anyone, including competitors. Seed it with your own common questions and answer them yourself. Reviews, and how to ask for them Reviews affect both ranking and whether anyone clicks. Volume, recency and rating all matter, and a steady trickle beats a burst. Ask everyone, at the right moment. Immediately after the job is done, when the customer is happiest. Not a week later. Make it one click. Google gives you a short review link. Put it in your email signature, on invoices, on a card, in a follow-up message. Never buy reviews. It is detectable, it violates the platform rules, and it costs you the profile. Reply to all of them. Positive and negative. Replying to a bad review is written for the next hundred people who read it, not for the person who left it. Calm, specific, no defensiveness. NAP consistency Name, address, phone number. These should be byte-identical everywhere they appear: your website, Google, directories, social profiles, industry listings. Search engines cross-reference these listings to confirm the business is real and where it says it is. Inconsistency, such as "Rue" in one place and "R." in another, or an old phone number lingering on a directory, dilutes the signal. Audit this once, fix everything, then keep a document recording the exact canonical format so future listings match. The same discipline applies to any directory or citation campaign. Building thirty listings with three variations of your details is worse than building fifteen that all match. Website basics that matter locally A page per location and per main service. One page trying to cover four towns and six services will rank for none of them. Separate pages, each genuinely written, not spun. Local content that is actually local. Mentioning the town name eleven times is not local content. Content about local conditions, local regulations, local events, work you did nearby, is. LocalBusiness structured data. Marks up your name, address, phone, hours and geographic coordinates in a form search engines read directly. A one-off job. Speed and mobile. Most local searches happen on phones, often with intent to act within the hour. A slow site loses those visitors before they see anything. Local citations and directories Listings on directories and industry sites build the consistency signal described above. Quality matters more than quantity. Relevant industry directories, local chambers of commerce and well-known general directories are worth having. Bulk submissions to low-quality link farms are not, and can hurt. For French businesses, the local directories that matter are different from the international ones, so build the list around where your customers actually look. Making it sustainable The reason most small businesses do local SEO badly is not ignorance. It is that it is nobody's job. It requires a small amount of work indefinitely: a post a week, review requests, replies, occasional content. This is where AI tools earn their place. Not by inventing a strategy, but by removing the friction from the recurring tasks. Drafting posts, replying to reviews, producing service page copy, writing the local content that would otherwise never get written. Mirage Cloud includes a marketing agent covering SEO, content and digital presence, with integrations to the social platforms where this ends up published. Two cautions. Review replies should be edited before sending. Generic AI-sounding replies are recognisable and they read worse than no reply. And AI-generated content still needs something specific and true in it, or it will not rank and will not persuade. What to do first If you do only three things: complete your Google Business Profile properly, build a habit of asking every customer for a review, and make your name, address and phone number identical everywhere. That is a weekend of work and it will put you ahead of most local competitors, who have done none of it.

Category:

Tag:

Description: Profitable businesses fail. Not often for dramatic reasons. They fail because money arrives later than it leaves, and nobody was watching closely enough to see it coming. The tool that prevents this is a cash flow forecast. Most small businesses do not have one, usually because they picture something complicated. It is not complicated. It is a list of money coming in, a list of money going out, and dates. Why the profit and loss account does not warn you Your P&L records an invoice as revenue when you issue it. Your bank account records it when the client pays, which under French payment terms might be sixty days later. A company can show a strong month on paper and be unable to pay salaries in the same month. The P&L is a scoreboard for a game that already happened. A cash flow forecast is a weather report for the game ahead. The thirteen-week rolling forecast Thirteen weeks is the standard horizon because it is long enough to see a problem coming with time to act, and short enough that your estimates are worth something. Beyond a quarter you are guessing. Rolling means you add a week every week, so you always have thirteen weeks ahead of you rather than watching a fixed window shrink. Build it like this. Start with your actual bank balance today. Not the accounting figure. The number in the account. List every expected receipt by the week you expect it to arrive. Issued invoices with due dates. Recurring subscription revenue. Anything else. Use the date the client will realistically pay, not the due date on the invoice, and if a client is habitually three weeks late then put them three weeks late. List every expected payment the same way. Salaries and contributions. Rent. Suppliers. Loan repayments. VAT. Corporation tax instalments. Insurance. Subscriptions. The annual and quarterly items are the ones people forget, and they are the ones that hurt. Each week, calculate opening balance plus receipts minus payments to get the closing balance, which becomes the next week's opening balance. That is the whole model. A spreadsheet is fine. Reading it The number that matters is the lowest closing balance across the thirteen weeks, and the week it occurs. If that low point is comfortably positive, you have time to think about growth. If it is thin, you know exactly which week is tight and you have weeks of notice to do something about it. If it is negative, you have found the problem early enough to act, which is the entire point. The mistakes that make forecasts useless Optimistic collection dates. The single most common error. If a client has paid at seventy days for a year, they will pay at seventy days again. Forecast the behaviour you observe, not the terms you agreed. Forgetting VAT. For a business collecting VAT, the payment to the tax authority is often one of the largest single outflows, and it is money that passed through your account and was never yours. Treating collected VAT as available cash is how businesses find themselves short on a filing date. Missing the annual items. Insurance renewals, the accountant's fee, the CFE, equipment replacement. They are predictable and they are routinely left out. Building it once. A forecast that is not updated is a historical document. Ten minutes a week is enough. Too much detail. A forecast with two hundred line items does not get maintained. Group your suppliers. Precision you cannot sustain is worse than approximation you can. Where automation genuinely helps The tedious part is not the arithmetic. It is gathering the inputs: what is in the bank, which invoices are outstanding, what is due to go out, what payroll costs next month. That data already exists in your systems. It just sits in three or four different places. Pulling it together by hand is the reason forecasts stop being updated by week three. Tools connected across banking, accounting and payroll can assemble it automatically. Mirage Cloud integrates with Qonto, Pennylane and PayFit, which covers the three sources most French small businesses need, and its finance agent is scoped to cash position, budgets and the kind of question you actually want answered, such as whether a hire in September is affordable given what is due in August. Automation does not replace the judgement. Deciding whether a client will pay late, or whether to delay a purchase, is still yours. What it removes is the data gathering, which is the part that kills the habit. Start smaller than you think If thirteen weeks feels like too much, do four. Four weeks of visibility is enormously better than none, and once the habit exists, extending it is easy. The forecast is not there to make you feel organised. It is there so that the worst week of your next quarter is a week you saw coming.

Category:

Tag:

Description: Hiring your first employee in France is the point where a lot of founders discover that the country's reputation for administrative complexity is not entirely unfair. It is manageable. It is just longer than people expect, and several steps have to happen in a specific order. Here is the sequence. Before you write a contract Identify your collective agreement. This is the step people skip, and it causes the most trouble later. Almost every French business falls under a convention collective determined by its activity, identified by an IDCC number. The agreement sets minimum salaries by classification, trial period lengths, notice periods, holiday rules and often additional benefits. You do not choose it. It applies to you. And it overrides your contract wherever it is more favourable to the employee, so writing a contract before reading it is writing a contract you may have to redo. Find your IDCC before anything else. Decide the contract type. A CDI is the open-ended default. A CDD is fixed-term and only lawful for specific listed reasons, such as replacing an absent employee or a genuine temporary increase in activity. Using a CDD because you are unsure whether the role will last is not a valid reason, and a CDD used improperly can be reclassified as a CDI by a tribunal. Set the salary against the right floor. The applicable minimum is the higher of the SMIC and your collective agreement's minimum for that classification. Budget for employer contributions on top of gross salary, which for a standard role add roughly 40% to 45%, reduced at lower salary levels. The formalities File the DPAE. The déclaration préalable à l'embauche goes to URSSAF and must be filed in the eight days before the employee starts. It is a single declaration that covers several registrations at once. Missing it is a serious matter, not a paperwork slip. Write the contract. A full-time CDI does not legally require a written contract, but working without one is a bad idea and your collective agreement may require it anyway. A CDD must be written and given to the employee within two working days of starting. Set the trial period correctly. For a CDI the statutory maximums are two months for employees and workers, three for supervisors and technicians, and four for cadres. Renewal once is possible only if your collective agreement permits it and the contract provides for it. Get this wrong and you lose the ability to end the contract simply. Open the registre unique du personnel. Every employer must keep one, from the first employee. It records identity, job, contract type and dates. Arrange the medical visit. The visite d'information et de prévention has to happen within three months of hiring, and sooner for certain roles. Set up the mutuelle. Employer-provided health cover is mandatory and you must pay at least half the premium. Cadres also require prévoyance cover. Put up the mandatory notices. Working hours, safety information, the collective agreement reference, the labour inspectorate's contact details and anti-discrimination notices all have to be accessible to staff. What changes as you grow Thresholds trigger new obligations. The most significant early one is 11 employees, which requires a CSE, the staff representative body. Others follow at 20 and 50. Worth knowing the map before you approach a threshold rather than after. Where first-time employers get it wrong Skipping the collective agreement. Discovering six months in that your agreement mandates a higher minimum, a thirteenth month or a shorter trial period is expensive and awkward. Getting the trial period wrong. An incorrectly drafted or improperly renewed trial period is one of the most common defects, and it removes your main flexibility in the first months. Treating a CDD as a low-commitment CDI. Reclassification is a real risk with real consequences. Copying a contract from the internet. Most templates online are either generic or drafted for a different collective agreement. Use one as a structure, not as a document. Forgetting the medical visit. It is easy to miss and it is a compliance failure. Making this less painful The genuinely hard parts here are judgement calls: which contract type fits, how to classify the role, what the trial period should be. Those benefit from a professional. The rest is finding out what applies to you, which is mostly a research problem. AI tools scoped to French employment law can shorten that considerably, because the questions are narrow and the answers are documented. Mirage Cloud includes an agent covering recruitment and French labour law, and integrates with PayFit, which is where the payroll side of this ends up anyway. Use it to understand what applies, draft the first version, and prepare your questions. Then have someone qualified check anything you are going to sign. The AI Act now classifies AI used in recruitment and employment decisions as high-risk, with obligations applying from December 2027, which is another reason to keep a human making the actual decisions. The whole process takes about a week if you work through it in order. It takes months if you discover the collective agreement last.

Category:

Tag:

Description: The French VAT threshold has been the subject of two years of noise, three proposals and one full reversal. If you lost track, you are not alone. Here is where things actually landed for 2026. The current thresholds Under the franchise en base regime you do not charge VAT to your clients, and you cannot reclaim VAT on your purchases. The thresholds for 2026 are: Activity Base threshold Tolerance threshold Services and liberal professions €37,500 €41,250 Sale of goods and accommodation €85,000 €93,500 For mixed activities, total turnover must stay under €85,000 and the services portion under €37,500. Nothing changed on 1 January 2026. The thresholds are the same as 2025. The €25,000 proposal that did not happen The 2025 finance bill proposed a single threshold of €25,000 for all activities. That would have made several hundred thousand micro-entrepreneurs liable for VAT overnight. The reaction was loud, the measure was suspended, and a consultation followed. An intermediate version circulated during 2025 keeping €37,500 for services but applying €25,000 to construction work. That was also dropped during parliamentary debate. So the position today is unchanged. The subject may return in a future budget, which is worth knowing if you are near a threshold and planning ahead, but nothing is in force. The trap: two different sets of numbers This is the single most common misunderstanding, and it costs people money. The VAT threshold and the micro-enterprise turnover ceiling are not the same number. The micro-enterprise ceilings are considerably higher, around €83,600 for services and €203,100 for commercial activity. You can therefore be comfortably within the micro regime for income tax and social contributions while being liable for VAT on your sales. People assume that staying a micro-entrepreneur means staying out of VAT. It does not. What happens when you cross The rules changed in 2025 and are stricter than they used to be. Crossing the base threshold (€37,500 or €85,000) in a given year means you leave the franchise on 1 January of the following year. The old rule requiring two consecutive years of overrun was removed. Crossing the tolerance threshold (€41,250 or €93,500) means you become liable for VAT from the first day of the overrun, not from the first day of the month, which was the position until 2024. That is a real change and it is unforgiving. If you cross on 14 October, sales from 14 October carry VAT. First partial year. If you started trading during the year, your threshold is prorated by the number of days the business existed. Starting on 1 July does not give you a full €37,500. Assuming otherwise is a common route to a tax adjustment. What to do if you are approaching a threshold Track turnover monthly, not annually. The tolerance threshold rule makes the exact crossing date matter, so an annual check is too coarse. Work out whether VAT actually hurts you. If you sell to consumers, losing the franchise means either a 20% price rise or a 20% margin cut, which is genuinely painful. If you sell to VAT-registered businesses, they reclaim it, so the change is close to neutral and you gain the right to reclaim VAT on your own purchases. Some businesses with heavy equipment or stock costs are better off registered. Consider voluntary registration. It is allowed, and for a business investing in equipment it can be the better choice. Update your invoice wording. Under the franchise you must carry the exemption mention. From 1 September 2026 the reference changes from Article 293 B of the CGI to the Code des impositions sur les biens et services, with a tolerance until 31 December 2027 during which the old reference stays acceptable. Remember the e-invoicing reform applies to you. Being in the franchise does not exempt you. From 1 September 2026 every VAT-registered business must be able to receive electronic invoices. Getting answers without a monthly retainer None of this is complicated once explained, but it is specific to France, it changes with each finance bill, and general-purpose AI assistants get it wrong regularly because their training data is full of outdated or foreign rules. Tools built for the French market handle this better. Mirage Cloud includes an agent scoped specifically to French accounting, VAT and tax questions, alongside integrations with Pennylane and Qonto. Narrow scope is the point. An assistant that only handles French tax questions is less likely to answer with an American rule. Treat any generated answer as a starting point rather than a filing. Anything with a tax consequence still needs a qualified professional to confirm it, and any vendor claiming otherwise is overselling. But for the question of whether you are close to a threshold and what happens if you cross it, a specialised tool will get you a usable answer in seconds.

Category:

Tag:

Description: One in four business failures in France is linked to late payment. That statistic gets quoted often enough that it has lost its force, so here is the version that lands better: your biggest client paying at ninety days instead of sixty means you are lending them money, at no interest, indefinitely, whether you agreed to it or not. The law is more on your side than most small business owners realise. Very few use it. The rules, briefly Payment terms between businesses are set by Article L441-10 of the Code de commerce. The default term is 30 days from delivery or performance. By written agreement this can extend to 60 days net from the invoice date, or 45 days end of month. That 60-day ceiling is a matter of public order. A contract clause setting 90 days is not a negotiation you lost. It is void, and treated as though it were never written. Your client cannot lawfully impose it, however large they are. What you are automatically owed when payment is late Two things become due from the day after the due date, with no reminder, no formal notice and no negotiation required. Late payment penalties. The rate should be set in your terms of sale, and it can never be lower than three times the legal interest rate. If your terms are silent or non-compliant, the fallback is the European Central Bank refinancing rate plus ten points, which works out at 12.15% for the first half of 2026. Penalties run day by day on the gross amount including VAT. A fixed recovery indemnity of €40. Set by Article D441-5. It is due per late invoice, not per client and not per reminder. Ten late invoices from the same customer means €400, not €40. If your actual recovery costs exceed €40 you can claim more, but then you have to evidence them. Neither of these requires you to ask nicely first. They are due by operation of law. The obligation that runs the other way Here is the detail that catches people out. Your terms of sale and every invoice you issue must state the payment term, the late payment penalty rate, and the €40 indemnity. Those three mentions are not optional. Their absence exposes you to a fine, and it also undermines your own position when you try to claim. A supplier chasing penalties with an invoice that never mentioned them is starting from a weak place. So the first action item is not chasing anyone. It is checking that your invoice template contains all three lines. Enforcement is real now The DGCCRF audits payment terms and publishes the results. Administrative fines reach €75,000 for an individual and €2 million for a company, doubled for a repeat within two years. Sanctions are published by name on the DGCCRF website, and for a large group the naming tends to sting more than the money. Suppliers can report a customer through SignalConso, anonymously if they prefer. For a small supplier facing a large customer who is systematically late, this is genuine leverage. It should be used carefully and usually mentioned before it is used, but it exists. Chasing without wrecking the relationship The legal position is one thing. Keeping the client is another. What works in practice: Send the invoice correctly the first time. A surprising share of late payments are administrative. Wrong reference, wrong recipient, missing purchase order number. From September 2026, structured electronic invoicing removes some of this by design. Chase before the due date. A short message five days out, framed as a check that everything is in order, is not aggressive and it works. It also surfaces problems while there is still time. Escalate on a schedule, not on a mood. Reminder at day one, second at day ten, formal notice at day thirty. Written down in advance so it happens consistently rather than when someone gets annoyed. Make the formal notice specific. "Please regularise this quickly" is not a formal notice. Identify the invoice, the amount, the due date, the penalties claimed, the deadline you are giving, and what happens next. Vagueness reads as an absence of intent to follow through. Decide in advance when you will claim penalties. Many businesses never do, out of fear of the relationship. A reasonable middle position is to state them clearly on every invoice, waive them on a first late payment, and apply them on a repeat. The part nobody has time for All of this depends on somebody noticing an invoice is late. In a small company nobody does, because noticing requires comparing a list of issued invoices against a list of received payments, regularly, forever. This is one of the clearest cases for automation. A tool connected to both your bank and your invoicing can tell you what is overdue and by how many days without anyone comparing anything. Mirage Cloud connects to Qonto and Pennylane among other French tools, which puts banking and accounting data in the same place, and its finance agent is scoped to exactly this kind of monitoring. The tool is not the important part. Noticing on day one instead of day twenty-one is the important part. Most small business cash flow problems are not caused by a lack of money. They are caused by finding out too late.

Category:

Tag:

Description: If you followed the EU AI Act coverage over the last year you probably came away with two contradictory impressions: that a major deadline landed in August 2026, and that everything got postponed to 2027. Both are true, which is why so much of the commentary has been confusing. Here is the position as it actually stands. What happened The AI Act entered into force on 1 August 2024 with obligations phased in over several years. By late 2025 it was clear that implementation was behind schedule, and on 19 November 2025 the European Commission proposed the Digital Omnibus on AI, a package of amendments whose headline measure was a delay to the heaviest obligations. Negotiations nearly collapsed in April 2026. A political agreement was reached on 7 May, and the Omnibus entered into force on 27 July 2026. What got delayed High-risk obligations under Annex III move from 2 August 2026 to 2 December 2027. These are the use-based high-risk categories, and they include the one most likely to touch an ordinary company: AI used in employment decisions. Recruitment screening, candidate selection, performance evaluation, task allocation, monitoring, promotion and termination decisions all sit in this bucket. High-risk obligations under Annex I, covering AI embedded in regulated products such as medical devices, lifts and radio equipment, move from 2 August 2027 to 2 August 2028. National regulatory sandboxes move from 2 August 2026 to 2 August 2027. If you were racing to complete a conformity assessment for a recruitment tool, you have sixteen additional months. What did not get delayed This is the part that gets lost. On 2 August 2026, three things took effect on schedule. Article 50 transparency obligations. These apply broadly and are the ones most relevant to ordinary businesses. Enforcement powers over general-purpose AI models. The AI Office can request technical documentation, evaluate models, require corrective measures and issue fines. The full penalty regime. Fines are now live across the Act. From 2 August 2026, the AI Office and national authorities in each member state are responsible for implementing, supervising and enforcing the regulation. Before that date the enforcement machinery was partly theoretical. It is not any more. The Omnibus also added a new prohibition to Article 5 covering AI-generated non-consensual intimate imagery and child sexual abuse material. What Article 50 requires of an ordinary business Most small businesses are deployers rather than providers, and the transparency obligations are the practical ones. Tell people when they are talking to a machine. If you run a chatbot or an AI phone system that interacts with customers, those customers must be informed they are dealing with an AI, unless it would be obvious to a reasonable person. For a phone receptionist this matters. A caller does not automatically know. Mark synthetic content. AI-generated or manipulated images, audio and video must be marked in a machine-readable way. There is a grace period to 2 December 2026 for systems already on the market before 2 August 2026. Disclose deepfakes. Content that resembles real people, places or events and could mislead has to be labelled as artificially generated. Disclose emotion recognition and biometric categorisation. If you use these, the people subject to them must be told. The Omnibus also extended the simplification measures for small and medium businesses to small mid-caps, and reworked the AI literacy obligation. The AI literacy point is worth a moment. The Act expects organisations to ensure staff using AI systems have a sufficient level of understanding of them. For a small company that is not a training programme. It is a short internal note about what the tools do, what they get wrong, and when a human has to check. A practical checklist List the AI systems you actually use. Most companies underestimate this. Customer-facing chatbots, phone systems, recruitment screening, content generation, translation, meeting transcription. Write them down. Sort them by risk category. Prohibited, high-risk, limited-risk with transparency obligations, minimal risk. The large majority of small business use falls into the last two. Check anything touching employment decisions. That is your most likely high-risk exposure. December 2027 sounds distant, and conformity assessments take longer than people expect. Fix your disclosures now. Chatbot greeting, phone system opening line, website notice, labels on generated images. This is a week of work at most, and it is enforceable today. Ask vendors where they sit. Any AI vendor selling into the EU should be able to tell you whether they are a provider under the Act, what risk category their system falls into, and what documentation they can give you. Vendors that cannot answer this in August 2026 have not been paying attention. European vendors tend to be readier on this than others, simply because the regulation is their home market. Mirage Cloud, for example, publishes a full Article 28 data processing agreement with a named sub-processor list and a documented acceptable use policy that explicitly requires AI output to be verified before use. Neither of those is exotic. They are just the paperwork you need when a customer or an auditor asks, and having it ready is the difference between a five-minute answer and a five-week project. The realistic summary The delay is genuine and useful for anyone with high-risk exposure. It is not a general reprieve. Transparency rules, general-purpose AI enforcement and the penalty regime are all operative now. For most small businesses the compliance work is modest: know what you use, tell people when they are talking to a machine, label synthetic content, and keep a human in the loop on anything consequential. That is close to what a careful company would do anyway.

Category:

Tag:

Description: There is a deadline coming that a lot of French businesses have not registered, partly because the reform has been postponed so many times that people stopped believing in it. On 1 September 2026, every business in France that is registered for VAT must be able to receive electronic invoices. Not send them. Receive them. That obligation applies to everyone, regardless of size, including micro-entrepreneurs operating under the franchise en base who do not charge VAT at all. The obligation to send electronic invoices arrives in two waves. Large companies and mid-caps must issue electronically from 1 September 2026. Small and medium businesses, micro-enterprises and independents follow on 1 September 2027. The reception deadline is the one catching people out, because it sounds passive. It is not. What "receiving an electronic invoice" actually means An electronic invoice in the legal sense is not a PDF attached to an email. That is a paper invoice in a digital envelope, and it does not satisfy the requirement. A compliant electronic invoice is a structured file that machines can read, transmitted through an approved platform. Three formats are recognised: UBL, CII, and Factur-X, which is a hybrid carrying a readable PDF with a structured XML file embedded inside it. Factur-X is the format most small businesses will end up using, because a human can still open it and see a normal invoice. The transmission has to go through a Plateforme Agréée, previously called a Plateforme de Dématérialisation Partenaire or PDP. These are private operators registered by the state. There is also a public portal, but it functions as a directory and routing layer rather than a free invoicing tool, which is a point of confusion worth clearing up early. So from September, when a supplier sends you an invoice, it will arrive in a platform, not in your inbox. If you have not chosen a platform and registered on it, you have nowhere for those invoices to land. Who this affects Every business subject to VAT in France, for domestic business-to-business transactions. That includes companies of every size and micro-entrepreneurs. Being under the franchise en base does not exempt you. Not charging VAT is not the same as not being within the scope of the reform. Transactions with consumers are not covered by the e-invoicing obligation, but they are covered by e-reporting, which also starts on 1 September 2026. E-reporting means transmitting transaction data to the tax administration for sales that do not produce an e-invoice: retail sales to individuals, and cross-border transactions. The invoice mentions that change From 1 September 2026, four new fields become mandatory on invoices: The client's SIREN number The delivery address, where it differs from the billing address Whether the transaction is a supply of goods, a supply of services, or both Where applicable, the option to pay VAT on debits There is also a wording change that affects anyone under the franchise en base. The familiar line "TVA non applicable, art. 293 B du CGI" is replaced by a reference to the Code des impositions sur les biens et services. A tolerance period runs to 31 December 2027, during which the old reference remains acceptable, so there is no need to panic about reprinting anything immediately. What to do before September Choose a platform. This is the one thing that cannot be skipped. Compare on connection to your existing accounting tool, the volume you handle, what the platform does beyond bare transmission, and price. Many accounting packages used by French SMEs are either becoming approved platforms themselves or partnering with one, so start by asking your existing provider what their plan is. It is often the shortest path. Clean your supplier and client data. Every business partner now needs a correct SIREN on file. Invoices will be routed by identifier, so wrong or missing SIRENs cause failures that are tedious to unpick later. Decide who monitors the platform. Invoices arriving in a platform rather than an inbox is a genuine workflow change. If nobody is assigned to check it, invoices will sit unnoticed and get paid late, which brings its own penalties. Talk to your accountant now, not in August. Every accountant in France is having this conversation with every client at the same time. Being early is worth a lot. The part that is actually good news Structured invoices are machine-readable by definition. That means the manual re-keying that consumes a chunk of every small company's admin week has an expiry date. Once invoices arrive as data rather than as PDFs, matching them against payments, spotting duplicates, flagging what is overdue and pushing entries into your accounts stops being typing and starts being automatic. Tools connected across banking and accounting can act on structured invoice data in ways they never could with a scanned PDF. Mirage Cloud is one of the platforms building for this. It connects to Qonto for banking, Pennylane for accounting and Yousign for signatures, with an agent scoped specifically to French accounting and VAT questions. That combination of French tools is the part that matters here, because the reform is French and generic international automation tools are not built around it. The reform will feel like an administrative burden for about six months. After that, for most small businesses, it will quietly remove work that nobody enjoyed doing. The one thing not to do is wait. The reception obligation has no size exemption and no grace period, and 1 September is close.

Category:

Tag:

Description: France quietly built one of the better small business software ecosystems in Europe. Qonto for banking. Pennylane for accounting. PayFit for payroll. Yousign for signatures with proper eIDAS standing. Brevo for email. Each of these is a good product, and several are better than the American equivalent for a French company, because they were built around French rules rather than adapted to them afterwards. The problem is not the tools. It is the gaps between them. Where the manual work actually is Watch how a five-person French company actually operates and the pattern is obvious once you see it. A payment lands in Qonto. Someone opens Qonto, sees it, and mentally matches it to an invoice. Later they open Pennylane and reconcile it properly. If it was a client payment, they cross it off a chase list that lives in a spreadsheet or in someone's head. Payroll runs in PayFit. The figures need to reach the accountant and the cash forecast. Someone exports something and sends it somewhere. A quote goes out, gets signed in Yousign, and now needs to become an invoice. Someone retypes it. A client goes quiet on an invoice. Nobody notices for three weeks, because noticing requires somebody to compare a list of issued invoices against a list of received payments, and nobody's job description includes doing that on a Tuesday. None of these individually takes long. Together they consume an afternoon a week in a small company, and the afternoon is usually the founder's. Why the gaps exist Partly because integration is unglamorous work that nobody wants to pay for. Partly because each of these products reasonably focuses on being excellent at its own job. And partly because the general-purpose automation tools that connect things, Zapier and its competitors, are built around American products first. Their French connector coverage is thinner, and setting them up requires the kind of person a five-person company does not employ. So the integration layer ends up being a human. Usually the founder, usually on a Sunday. What closing the gaps looks like The useful version of AI for small business is not writing marketing copy. It is sitting across these systems and noticing things. A tool connected to both your bank and your invoicing can tell you which invoices are overdue without anyone comparing two lists. Connected to banking and accounting, it can flag a payment that does not match anything. Connected to payroll and cash position, it can tell you in July whether September's hire is affordable. This is unremarkable work. It is also exactly the work that does not get done in small companies, because it requires someone to look at two systems at once, regularly, forever. Mirage Cloud is one of the products building specifically for this stack, with live connections to Qonto, Pennylane, PayFit, Yousign and Brevo alongside the usual Google and Stripe integrations. The French-specific set is the interesting part. Plenty of AI platforms connect to Salesforce and HubSpot. Very few connect to the tools a French company with eight employees actually runs on. What to check before you connect anything Read direction versus write direction. An integration that reads your bank transactions is very different from one that can move money. Know which permissions you are granting. Most of the value is in reading, and most of the risk is in writing. Live versus roadmap. Integration pages routinely list both together with a small label. Confirm in writing which ones work today. Sync frequency. Real time, hourly, daily. For cash flow monitoring this matters. What happens on disconnection. Whether data pulled in stays after you revoke access, and how you delete it if you want it gone. The realistic benefit Nobody should expect the afternoon a week to disappear entirely. Reconciliation still needs judgement, and anything touching your accounts needs a human eye before it is final. What is realistic is turning the afternoon into an hour, and turning the things you find out three weeks late into things you find out the same day. For a small company, the second one is worth more than the first. Most cash flow problems in small businesses are not caused by a lack of money. They are caused by finding out too late.

Category:

Tag:

Description: Every small association has the same conversation at some point. Someone looks around the room and asks why four volunteers spent an entire Saturday on paperwork instead of on the thing the association actually exists to do. The paperwork is not optional. Statutes have to be respected. General meetings need convening properly, with notice, and minutes afterwards. Grant applications have deadlines and formats and reporting obligations attached. Accounts need keeping even when the budget is small. In France, an association loi 1901 carries a genuine set of obligations that do not scale down just because the organisation is run by five people in their evenings. What makes this heavier than it needs to be is that almost none of it is hard. It is just repetitive, formatted, and time-consuming, and it lands on people who volunteered to do something else entirely. The three biggest time sinks Grant applications. Most funders want broadly the same information written in slightly different ways. Mission, governance structure, budget, beneficiary numbers, expected impact, evaluation method. Every application means reassembling the same facts into a new template. Associations that apply for six grants a year write essentially the same document six times. Statutory paperwork. Convening notices, attendance sheets, minutes, resolutions, the annual activity report. Each one follows a fixed structure. Getting the structure wrong can make a decision contestable later, which is why people are slow and careful about it, which is why it takes so long. Communication. Newsletters, event announcements, posters, social posts, thank-you letters to donors. Small associations rarely have anyone with a communications background, so this either gets done badly or does not get done. What is genuinely automatable now Grant applications are the clearest case. Once an association's core facts are written down once, generating a first draft tailored to a specific funder's requirements is exactly the kind of work language models are good at. The draft still needs a human to check the figures and add the specific detail that makes an application distinctive, but it removes the blank page, which is the part that eats the Saturday. Meeting documents are the second clear case. Convening notices and minutes follow known structures. Producing a correctly formatted draft from a set of notes takes seconds rather than an evening. Several tools now cover this. Mirage Cloud includes an agent specifically scoped to French association governance, covering loi 1901 obligations, grant files and statutory paperwork, which is a narrower and more useful framing than a general assistant that has to be told what an association is every time. What is not automatable, and should not be The relationship with funders. The judgement about which grants are worth pursuing. The parts of an application where you explain why your organisation specifically should do this work, in your own words, with real examples. Anything with a legal consequence also needs human sign-off. A generated draft of a resolution is a starting point, not a filed document. If a decision could be challenged later, someone who understands the statutes should read it before it goes out. And there is a tone problem worth naming. Donors and funders can tell when they are reading generated text, and it costs you. Use the draft to get past the blank page, then rewrite the parts that matter in a human voice. A sensible starting point Write down your association's core facts once, properly. Legal name, registration details, founding date, mission statement in two lengths, governance structure, current board, annual budget, beneficiary numbers, three concrete examples of work done in the last year. Keep it in one document. That single file is what makes everything else fast, whether you use an AI tool or not. Most of the time volunteers lose to paperwork is spent hunting for facts that already exist somewhere, in someone's email, in last year's application, in a folder nobody can find. Get that document written. Then automate what sits on top of it.

Category:

Tag:

Description: Nearly every AI product sold in Europe now has a flag on its homepage and a line about data residency. Very few buyers read past it, and the line is usually doing more work than it should. Here is the thing that trips people up. "Hosted in France" and "your data stays in France" are different statements, and most AI products can only honestly make the first one. Why the two are not the same A typical AI SaaS product has at least two distinct data paths. The first is storage. Your account, your documents, your conversation history, your uploaded files. This genuinely can sit on servers in one country, and for European vendors it often does. The second is inference. When you type a question and the product generates an answer, that request goes to a language model. Very few European companies run their own models, because training and serving frontier models costs more than most of these companies have raised. So the request goes to OpenAI, Anthropic, Google, Mistral, or some combination. Several of those are American companies with American infrastructure. Storage in Frankfurt or Paris. Inference in Virginia. Both statements true at once. This is not a scandal, and it is not a reason to avoid these tools. It is legal, provided the transfer mechanism is right. But it does mean the flag on the homepage tells you less than you think, and the document you actually need is the one nobody puts on the homepage. The document you want It is called a Data Processing Agreement, or DPA. Under Article 28 of the GDPR, any vendor processing personal data on your behalf has to have one. If a vendor cannot produce one, that is your answer and you can stop there. Inside it, go straight to the sub-processor list. This is usually an annex at the back, and it is the most honest page on any AI vendor's website. It has to name every third party that touches your data, what they do, and where they are. Read that table and you will know more about the product's actual architecture than you will learn from the entire marketing site. What to look for, line by line Who the model providers are. OpenAI, Anthropic, Google, Mistral, Cohere. Their presence is normal. Their absence, on a product that clearly generates text, means the list is incomplete. Where each one sits. United States, EU, or both. This determines what transfer safeguard is needed. The transfer safeguard column. For US sub-processors you want Standard Contractual Clauses, adherence to the EU-US Data Privacy Framework, or both. A US sub-processor with no safeguard listed is a compliance gap. Voice providers, if the product handles calls. Voice synthesis and transcription are frequently outsourced, and call recordings are more sensitive than most other data a small business holds. The change notification clause. A good DPA commits to telling you when sub-processors change, and gives you a window, usually thirty days, to object. The deletion clause. What happens at the end of the contract, in what format data comes back, how long deletion takes. A worked example Mirage Cloud, a French AI platform for small businesses, publishes a full sub-processor annex. Hosting is with IONOS in France. Transactional email is split between Brevo in France and Resend in the United States. Payments go through Stripe. Model inference runs through OpenAI, Anthropic and Google. Voice for the telephone agent runs through ElevenLabs in the United States. Cloudflare handles network security. Each non-EU entry lists SCCs plus Data Privacy Framework as the safeguard. That is a normal architecture for a European AI product in 2026, and publishing it in that detail is more than many competitors do. It also demonstrates the point of this article. A buyer reading only the homepage would come away with a simpler picture than the one in the annex, and the annex is the accurate one. Questions worth asking a vendor Five that reliably produce useful answers. Send me your DPA and your current sub-processor list. Is our data used to train any model, yours or a third party's, and is that commitment written in the contract rather than only on the website? Where is data at rest, and where does inference happen? Please answer separately. What is your retention schedule after account deletion, broken down by category? Have you appointed a DPO, and if not, who is the named contact for data protection questions? That second one catches a lot of vendors out. "We do not train on your data" appears on a great many homepages and in a great many fewer contracts. If it matters to you, get it in the agreement. None of this requires a lawyer to do a first pass. Fifteen minutes with the DPA will tell you more than an hour on the marketing site.

Category:

Tag:

Description: Most small business owners have now tried ChatGPT for something work-related. A fair number gave up. The usual story goes like this. You ask it to help with a supplier email and it does a decent job. You ask it something about VAT treatment and get an answer that sounds authoritative and turns out to be wrong, or right for the wrong country. You ask it to look at last quarter's numbers and remember that it has no idea what last quarter's numbers were. After a few rounds of this you go back to doing things by hand. The tool is not the problem. The shape of the tool is the problem. The two things a general assistant does not have Context. A general chatbot starts every conversation knowing nothing about you. It does not know your VAT scheme, your headcount, your payment terms, which clients are slow payers, or that you operate under French labour law rather than American employment-at-will. You can tell it, every time, in a long preamble. Almost nobody does that consistently, which is why the output quality swings so wildly. Boundaries. A general assistant will answer anything you ask with the same confident tone, whether it is good at that thing or not. It has no concept of being out of its depth. For casual use that is fine. For anything with a compliance consequence it is a liability. What specialisation changes The alternative approach is to build several narrow agents instead of one broad one, each carrying its own instructions, its own reference material, and its own boundaries. An agent scoped to French accounting knows which VAT regime applies, what the filing deadlines are, and what a liasse fiscale is, because that is all it has been set up to handle. An agent scoped to recruitment knows the rules around trial periods and fixed-term contracts under French labour law. Neither of them will confidently improvise an answer to something outside its remit, because they are not built to. You lose flexibility. You gain reliability, which for business work is the better trade. The coordination problem, and how it gets solved Split your assistant into a dozen specialists and you create a new problem. Real questions do not respect departmental boundaries. "Can we afford to hire someone in September" is a finance question, a payroll question, a labour law question and a sales forecast question at the same time. If you have to ask four separate agents and stitch their answers together yourself, you have not saved any time. You have added admin. This is why the better multi-agent products put a coordinator in front. You ask one question, in plain language, and the coordinator works out which specialists to consult and returns a single answer. Mirage Cloud is built this way. An orchestrator agent named Sofia sits in front of eleven specialists covering accounting, finance, HR, marketing, legal, logistics, sales, customer support, nonprofit governance, general assistance and phone reception. Ask a question that spans three domains and the routing happens without you thinking about it. The design goal is that you delegate the way you would to a colleague, rather than assembling a prompt. Where this approach still has limits Worth saying plainly, because most articles on this topic will not. Specialised agents still run on the same underlying language models as general ones. Specialisation improves reliability. It does not eliminate errors. Anything with a legal, tax or financial consequence still needs a qualified human to review it before it goes anywhere. Any vendor telling you otherwise is overselling, and most vendors' own terms of service say exactly this in the small print. Multi-agent systems are also only as good as their connections. An agent that cannot see your bank account or your invoicing system is guessing. The value is in the integrations, so when you evaluate one of these products, look at the integration list before you look at the agent list. And there is a floor on complexity below which this is overkill. If you are a solo freelancer with ten invoices a month, a general assistant and a spreadsheet will serve you fine. How to evaluate one Ask what happens when an agent does not know something. Good systems say so. Bad ones invent. Ask which integrations are live today versus on a roadmap. Most vendors list both in the same grid with a small label, and it is easy to miss. Ask what the fallback is when the automated answer is wrong, and whether there is a human you can reach. Then test it on something you already know the answer to. It is the fastest way to find out whether a product is genuinely specialised or just a general model with a name and an avatar attached.

Category:

Tag:

Description: Ask a plumber, a dental practice or a small law firm how many calls they miss in a week and you will usually get a shrug. Nobody counts. The phone rings while you are with a customer, or driving, or it rings at seven in the evening when the office is shut, and the caller hangs up. There is no record. It never enters a spreadsheet. That is the problem with missed calls. They are invisible. A lost email sits in a folder somewhere. A missed call just disappears. The research that does exist puts the number higher than most owners expect. Studies of inbound call handling in service businesses consistently find that somewhere between a fifth and a third of calls go unanswered, and that most callers who reach voicemail do not leave a message. They call the next business on the list. For a trade business where the average job is worth a few hundred euros, that maths gets uncomfortable quickly. Three missed calls a week, a third of which would have converted, is roughly one lost job every week and a half. Why the usual fixes do not work Voicemail. People under forty largely refuse to use it. If your voicemail box is your safety net, the net has holes in it. Call diversion to a mobile. This just moves the problem. You are still on a roof, or in a meeting, or asleep. A human answering service. These work well and have done for decades. They also cost real money, usually charged per call or per minute, and the person answering does not know your business. They take a name and a number, which is barely more than voicemail with better manners. Hiring someone. Fine if you have the volume to justify a salary. Most small businesses do not. What an AI receptionist actually does The honest version: it picks up, holds a conversation in natural language, and does something useful with the outcome. The useful part varies by product, but the capabilities worth looking for are these. It answers the questions people actually call about. Opening hours, whether you cover a particular postcode, whether you handle a particular type of job, roughly what something costs. A large share of inbound calls to small businesses are information requests that never needed a human. It takes a structured message. Not just a name and number. The reason for the call, the urgency, the address, whatever fields matter in your trade. That arrives as a written summary you can act on. It qualifies. You can set rules. A burst pipe is urgent and should trigger a text to your mobile. A quote request for work six weeks out can wait until morning. A cold sales call can be politely ended. It transfers when it should. Good systems know their limits and hand off to a real person rather than looping. What to check before you buy one Voice quality is the first thing prospects notice and the thing most cheap systems get wrong. Ask to hear it. Ask to hear it handling an interruption, because real callers talk over the system constantly. Setup effort is the second. Some products expect you to write prompts, which most business owners neither want to do nor should have to. Others run a questionnaire about your business and build the configuration for you. Mirage Cloud takes the second approach with its receptionist agent, Xavier, and lets you talk to the finished thing in your browser before you point a phone number at it. That browser test matters more than it sounds. It is the difference between finding a problem yourself and finding it because a customer complained. Language handling is third, and it is where a lot of international products fall down for European businesses. A system trained mostly on American English will mangle French street names and struggle with regional accents. If you operate in France, test it on French addresses before you commit. Then the boring commercial questions. How are minutes billed, and what happens when you exceed the bundle. Whether calls are recorded and transcribed, where those recordings live, and how long they are kept. Whether you can run more than one receptionist if you have separate lines for sales and support. Where this genuinely does not help If your callers need a decision only you can make, an AI receptionist buys you a message and nothing more. If your call volume is genuinely tiny, five calls a week, the payback is thin. And if your business depends on a warm personal relationship from the first ring, some customers will notice and dislike it. The businesses that get the most out of this are the ones with high inbound volume, repetitive questions, and an owner who cannot be at a desk. Trades, clinics, salons, small agencies, property management. If that sounds like you, the honest question is not whether an AI receptionist is as good as a great human one. It is whether it is better than the voicemail you have now.

Category:

Tag: