Why every employee has their own AI subscription (and what it's costing your company)

Your employees are using AI. Not the one you chose — the one they found themselves, one evening, trying to finish a deliverable. Each one has their own personal account. Each one pays their own subscription. And each one pastes in, without thinking, client briefs, pricing grids, prospect names. You don't know about it. Nobody flagged it to you. Not out of malice — out of habit. When the company doesn't provide an AI workspace, people make do. That's exactly what they're doing. The problem isn't that they're using AI. The problem is that your company is paying three times over: in scattered subscriptions, in data leaving your perimeter, and in work being redone because nothing is shared. This article lays out the situation, quantifies the real cost, and shows what changes when AI becomes a company structure instead of remaining a collection of personal accounts.

You didn't decide to deploy AI in your SME. It deployed itself. Without validation, without a framework, without anyone asking you anything. Here's what it looks like day to day.

Your sales rep rewrites a follow-up email in her personal AI. She pastes in the prospect's name, the quote amount, the date of the last follow-up. Your accountant asks another AI to rephrase a contractual clause — and attaches the contract. Your project manager types a client brief into it to save ten minutes writing a specification document. Nobody does this out of carelessness. Everyone does it for efficiency. AI is good. AI is fast. And when the company offers nothing, the employee takes what they can find. This is shadow AI — AI used outside the perimeter you control. The term sounds alarming. The reality is mundane. Count the number of people on your team. Ask them, without judgement, whether they use a personal AI for work. The answer will surprise you with how ordinary it is, not with its scale.

Shadow AI doesn't get reported. There's no ticket, no purchase request, no sign-off. The subscription costs between 20 and 25 euros a month on a personal card. It's not an IT issue — your SME probably doesn't have an IT department. It's not an HR issue — nobody considers using an AI to rephrase an email a breach of conduct. The result: you're running a company where a portion of the intellectual work passes through pipes you can't see. Pipes that share no common memory, have no connection to each other, and have no confidentiality obligation towards your clients. The problem isn't insubordination. It's the absence of a framework. When there's no road, people walk through the grass. That's normal. It's your job to build the road.

Banning personal AI doesn't work. It's like banning mobile phones in the office in 2010 — you lose the battle and come across as someone putting the brakes on progress. The question isn't whether your teams should use AI. They already do. The question is whether they're doing it in a space you control or in a space that doesn't belong to you. We've already covered this topic from another angle: comment construire une mémoire d'entreprise avec l'IA quand on n'a pas de DSI. Shadow AI is its direct consequence. No company memory, no shared workspace — so everyone builds their own, elsewhere.

The visible cost is easy to calculate. The real cost is much harder to pin down. Here's a method to measure it yourself, without made-up percentages or studies you can't verify.

Take the number of people on your team who use AI for work. Multiply by 20 euros per month — the standard price of an individual AI subscription. For a team of 8, that's 160 euros per month. Nearly 2,000 euros a year. Spread across 8 different bank cards, 8 different accounts, 8 conversation histories that never intersect. Compare that figure to what a shared company AI workspace would cost. Anakoro Atelier, for example, offers four plans from 19 to 199 euros per month for the whole team — with a shared memory. The maths is simple. What's less simple is admitting you're already paying without realising it. If this topic interests you, we've detailed the logic of the single plan in cet article sur la plateforme IA entreprise au forfait.

The real cost isn't the subscription. It's time. Every employee rebuilds their context with every conversation. They re-explain who the client is, what the project is, what the constraint is. Personal AI has no company memory. It starts from zero every session. Do the maths yourself: estimate the time your team members spend re-giving context to their AI. Two minutes per request, ten requests per day, five days a week. One hundred minutes per person per week spent explaining to a machine what the company already knows. Multiply by your average hourly rate. It's not a spectacular number. It's a steady number, accumulating quietly, month after month. Nous avons écrit un article entier sur ce phénomène : pourquoi votre IA oublie tout ce que vous lui dites.

Your client data is circulating in spaces you don't control. Names, amounts, negotiation histories, contractual terms — everything your employees paste into their personal AI is processed by a third-party service, under terms and conditions nobody in your company has read. This isn't paranoia. It's basic regulatory common sense. If a client asks you where their data is being processed, you need to be able to answer. With individual AI accounts, the honest answer is: "I don't know." That's not a tenable position for a business leader. It's not a position you chose either — it's a position that the absence of a framework has put you in by default.

The leaking of company data into personal AIs isn't an accident. It's the normal functioning of an individual tool used in a collective context. Here's the mechanism, and what changes it.

An employee opens their personal AI. They ask a professional question. For the answer to be useful, they provide context. That context is company data. The name of a prospect. The contents of a quote. A clause from a contract. The margin on a product. They don't do it carelessly — they do it because without that context, the AI gives a generic and useless answer. The problem isn't the employee. The problem is that they have nowhere else to go. When the company doesn't offer an AI workspace with shared memory and protected data, the path of least resistance is the personal account. Every day. Every request. Every piece of data that leaves.

Anakoro is a French-language business suite built natively around AI: a single entry point, a shared company memory, and modules — sales, management, recruitment, visibility — that replace stacked subscriptions. AI isn't a module within Anakoro. AI is the structure of Anakoro. In practice, this means that when a team member uses Atelier, they're working in a space that already knows the company's context. Contacts from Cercle, previous exchanges, shared documents — it's all there. They don't need to re-paste context. They don't need to hand a client's name to a third-party service. The AI works inside the company's perimeter, not alongside it. This is precisely the topic we covered in our guide on la méthode pour remplacer plusieurs logiciels par un seul.

First scenario: your project manager opens her personal AI. She types "rewrite this email for the Durand client, here's the project context" and pastes in a full paragraph with budget, deadlines and names. The AI responds well. The data has left the building. Nobody else on the team benefits from the exchange. Tomorrow, a colleague will ask the same question — and paste in the same data. Second scenario: same project manager, same need. She opens Atelier. The Durand client context is already there — imported from Cercle, enriched by previous exchanges. She asks for the rewrite. The AI responds within the company's perimeter. The exchange remains in the shared memory. Her colleague will find the work already done, not a blank screen. Same need, same speed, different structure. The rest is just plumbing — but it's the plumbing that makes the difference between a leak and a closed circuit.

You can't stop your teams from using AI. You shouldn't want to. What you can do is give them a space where AI works with the company's memory — not against it. Shadow AI isn't a scandal. It's the symptom of an absence. The absence of a company AI workspace, designed for collective use, with data that stays inside. Atelier exists to fill that absence. Four plans, from 19 to 199 euros per month. A shared memory. Multiple models. And one simple rule: it suggests, you decide. Seven days to see what changes. The rest is proven in use.