You read that AI was going to change everything. You may have tried it out without any follow-up, and your teams haven’t made any progress. On October 13, 2026, in front of the companies brought together by the Marne-Ardennes CCI, Mathieu Domon, our president, took the matter on the other hand: he didn’t present a technology, but three tasks that companies currently perform manually. Before founding Optima AI, he spent 21 years at Bureau Veritas, including 15 years in the management of large industrial accounts. This is the field that our method comes from: we start with a task, and the person who answers to it.
In short
- We don’t start from AI; we start from’a task She often returns; a person identified has made it today, and she knows how to describe it, even though nothing is written down.
- Before choosing anything, we measures what it costs this task: who is doing it, how many times, for how long, and what happens when it is delayed.
- The division of roles is decided from the outset: The AI proposes, the human validates.
- On Start small : a process, worked out with those who are doing it, before opening a second one.
- No need for a perfect database: we’re starting from scratch documents that the task really needs, as they are.
Where to start to integrate AI in SMEs?
By a task, not by technology. Choose a job that comes back every week, that mainly consists of reading, sorting, comparing, or writing, and that someone in your family knows by heart. This is the scale at which a small business starts: a well-chosen task, worked on with the people who do it today.
The question «which AI to choose?» comes much later, and it sometimes never arises. When a known rule is sufficient, a simple software program does the job. And when the difficulty comes from organization, no tool will solve it for you.
Which method should be followed for a first AI project in a SME?
A first AI project in SMEs follows five steps: choosing a task, measuring its cost, dividing roles between the AI and the team, starting with a small scope, and then gathering the documents the task requires. None of these steps requires technical expertise: they require knowledge of the business.
- Starting with a task, not with technology. Make a list of tasks that come back without anyone having chosen them: reread a specification sheet, retrieve an old offer, find a reference in a folder, fill out the same questionnaire for a new client. Keep those that someone knows how to describe and judge, even if the instructions are not written anywhere.
- Measuring what it costs. Before talking about a tool, write down in black and white what the task is, how often it is done, and how long it takes each time. Add the cost when it goes wrong: a closed deal the day before the deadline, a dossier that has to be redone several times, a collaborator who no longer does their real job. This list becomes your benchmark: without it, you will never know if the project was useful.
- Deciding what the AI proposes, and what the human validates. An AI can read, sort, compile, reformulate and find information within a pile of documents. It doesn’t know whether a quote is valid, whether an answer suits a client or whether a risk is acceptable: it remains a matter of experience. From the outset, write down who is reading, who is correcting and who is signing.
- Start small. A written outline, real files, and the person who was doing the work next to her: she is the one who identifies what’s not working, and it’s with her that the problem is resolved. It’s only after this first trial in service that the next task is looked at.
- Looking at the data and documents that the task requires. AI works with what is given to it: outdated or contradictory versions, and the result is felt. But waiting for a perfect base is the best reason to never start. List the documents that the task really uses and organize them. Also ask early about privacy: where are your documents stored, who has access to them, are they used to train a model?
What does an AI use case in SMEs look like?
For an ordinary task, which the AI prepares and the team validates. On October 13, 2026, with the CCI Marne Ardennes, our webinar «Integrating AI in your company: myth or reality?» focused on three issues that many SMEs face: time, difficulty, and retirement. It illustrated them through three cases, which can be revisited using the method.
Can AI write the instruction manual for a machine?
She can draft a first draft based on the functional analysis and the risk analysis of the machine; the team that knows the machine reads it, completes it, and validates it. Without these two analyses, there is nothing to draft.
The task checks the method’s boxes: it applies to every new machine, it relies on documents the company already has, and it involves a person who has other urgent tasks. What AI offers: the text of the manual, drawn from these analyses, and a list of what it does not say, reported rather than filled in. What the human keeps: completing what is missing, correcting, and deciding whether the manual can be released.
That is what Optima Doc Creation, one of the agents of’Optima Compliance, Today in a pilot version.
Can AI help prepare a commercial offer or a response to a tender?
Yes, on the preparation; the decision and commitment remain with the sales and business development teams. Responding to a tender or building a commercial offer is four jobs in one: reading the dossier, analyzing it, calculating costs, and writing the proposal. And often, everything is decided the day before the deadline.
What AI offers: reading the file, identifying what is requested, initial drafting based on your previous offers and documents. What the human retains: the decision to respond or not, the price, the commitments made to the client, and the text that is sent. The task often remains the same, its cost is visible, and the roles are clearly defined.
This is the work of’Optima Proposal Manager, Our software for responding to tenders.
How to document the skills of an employee before they retire?
With the person who holds it, and before their departure: it is she who knows what she is doing, and it is she who decides whether what has been written is true. The know-how of a position is often not written down anywhere: it is in the minds of a few people.
This is the case where the criterion «a task that can be described» counts the most: nothing is written down, but the person knows how to describe their work. AI can help to document the skills of a position based on what the person in charge does; the person reads and corrects. A precaution that leaders often keep silent about: writing down what someone does without them is letting them believe that they are preparing to replace them. The approach is done with them, face-to-face.
This is a subject we are working on, with no announced date.
What principle ties together the use cases of AI presented in the CCI Marne Ardennes webinar?
Only one: the AI proposes, the human validates. In the machine’s manual, in the commercial offer as well as in the expertise of a position, the task existed before the AI; a person in the field already performed it, and it is she who retains the decision-making power.
On October 13, Mathieu Domon did not open any of the three cases using technology. He opened them all using a task, and the person who answered it. The replay details the three cases, with a chapter per case and a point on the security and confidentiality of the entrusted data. → Watch the replay
How to choose the first task to entrust to AI in your company?
The right first task often comes back to someone who has identified themselves and described themselves in a few sentences. The safest way is to show it, as it is done today, to someone who can say whether there is a case for a tool, whether a market-ready software is sufficient, or whether the first need is elsewhere.
That’s what we offer: 30 minutes of video conferencing, free of charge and without commitment. You describe the task to us; our team tells you what they think of it, including when the answer is «no tool.».
The transition of a company towards AI is not done alone: it must be accompanied. That is the role we take on, from the first chosen task to the established tool.
Frequently asked questions
Is AI reserved for large companies?
No. A few years ago, a specialized team and months of project planning were required; that’s no longer the ticket to entry. A small business doesn’t need a big «AI» project: it needs a specific task, that costs it time every week, worked with those who do it.
Do you need technical skills in-house to start an AI project?
No. The first steps require knowing your job: choosing the task, stating its cost, deciding who validates it. The technical part falls to the person who designs the tool; the company’s role is to review, correct, and validate what the AI is preparing.
Should we first organize all of our data before using AI?
No. A perfect baseline never arrives, and waiting for it amounts to never starting. It is enough to start with the documents used by the chosen task: old models, procedures, past responses, even imperfect ones, are already a matter of work.
Will AI replace the employees of a small business?
She changes some tasks, not the job itself. The real question is not «who will be replaced?» but «what tasks do we want to entrust, and who will keep the reins?» In a well-managed project, the person who did the task remains the one who reads and decides.
What happens to the documents entrusted for an AI project?
This is a question to ask before starting, regardless of the provider: where are your documents stored, who can access them, do they serve to train a model? At Optima AI, the documents entrusted as part of an Optima Studio project are stored in France, are never accessible to another client, and are not used to train an AI model.
French 
English