DEV Community: Mirage Cloud IA

How to Write a Quote That Wins the Job (French Legal Requirements Included)

Mirage Cloud IA — Thu, 27 Aug 2026 06:27:15 +0000

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.

Where to Start Automating Admin Work in a Small Business

Mirage Cloud IA — Fri, 21 Aug 2026 05:28:54 +0000

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.

Customer Support When You Do Not Have a Support Team

Mirage Cloud IA — Fri, 21 Aug 2026 04:43:53 +0000

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.

How to Evaluate an AI Tool Before You Buy It: A Practical Checklist

Mirage Cloud IA — Wed, 19 Aug 2026 08:28:58 +0000

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.

Local SEO for Small Businesses With No Marketing Team

Mirage Cloud IA — Tue, 18 Aug 2026 09:05:23 +0000

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.

Cash Flow Forecasting for Small Businesses Without a Finance Team

Mirage Cloud IA — Mon, 17 Aug 2026 10:35:48 +0000

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.

Hiring Your First Employee in France: The Complete Checklist

Mirage Cloud IA — Mon, 17 Aug 2026 08:44:39 +0000

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.

French VAT Thresholds in 2026: What Changed, What Did Not, and What Happens If You Cross One

Mirage Cloud IA — Fri, 14 Aug 2026 04:13:25 +0000

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.

Late Payments in France: Your Legal Rights and How to Actually Get Paid

Mirage Cloud IA — Thu, 13 Aug 2026 04:22:43 +0000

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.

The EU AI Act in August 2026: What Changed, What Got Delayed, and What Small Businesses Must Do

Mirage Cloud IA — Wed, 12 Aug 2026 04:27:20 +0000

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.

E-Invoicing Becomes Mandatory in France on 1 September 2026: What Every Business Must Do

Mirage Cloud IA — Tue, 11 Aug 2026 12:00:45 +0000

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.

Why One AI Assistant Is Not Enough to Run a Business

Mirage Cloud IA — Tue, 28 Jul 2026 11:54:12 +0000

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.