AI does not replace your teams. Properly framed, it gives them time back.
AI is already in your organisation, often without a framework. In a 2026 survey, 69% of SMEs say they use tools such as ChatGPT, but 30% do not trust them with confidential data.
Meanwhile, repetitive data entry, searching for information and chasing follow-ups take up a large share of the working day. In our experience, the gains come from two or three well-chosen use cases, not from a mass rollout.
Put AI to work on your repetitive tasks, with clear rules of use and data that stays within your environment.
Start with the assessment to choose the use cases, prove the value on a first case, then extend what works. Each stage is decided separately.
Choose the 2 or 3 use cases worth pursuing.
A concrete, measured result before going further.
Extend what works and support your teams.
| Assessment | First use case | Deployment and adoption | |
|---|---|---|---|
| Workshops by function, prioritisation and tool selection | ✔ | — | For each new case |
| AI acceptable use policy | ✔ | ✔ | ✔ |
| One use case in production, gain measured before and after | — | ✔ | ✔ |
| Deployment by profile, new use cases, CRM and ERP | — | — | ✔ |
| User training | — | Optional | ✔ |
| Quarterly review of gains and risks | — | — | ✔ |
Examples from our assignments, anonymised. Each case is assessed on its impact, effort and return on investment before it is launched, then moves from idea to test, then to a pilot in real conditions.
An assistant answers users’ everyday requests from the knowledge base, escalating to the IT team for anything sensitive.
Multi-jurisdiction legal research (France, Europe, USA) on an external database, with sources always cited. Currently in pilot.
Help with resolution drawing on tickets, manuals and fault history, with hand-over to a human. Currently in pilot.
Sorting and assigning requests to reduce backlog and better meet service commitments, with the possibility to correct.
Help planning installation, commissioning and maintenance work, with scenario simulation and human validation.
Quickly find a drawing, a standard or a project file, with sourced answers and respect for access rights.
We do not sell a tool, we help you choose one. The assessment compares the market’s assistants against your real use cases: Claude, ChatGPT, Mistral, Perplexity, Microsoft Copilot, and others if necessary. The result is a written recommendation: a main tool and, where useful, a second complementary one. We do not push any vendor.
Is the tool good at your line of work: writing, document analysis, information search, code, customer relations?
Which sensitive data is involved, and does the professional plan rule out training on your data?
Where is the data processed, in Europe or outside Europe? Is this compatible with the GDPR and your customers’ requirements?
Does the tool connect to Microsoft 365, SharePoint, your CRM or your ERP?
How much per user per month, for how many people actually concerned?
Will you be able to switch tools without starting over?
We take stock of what is already being done, choose the use cases and the tools, and set the rules.
A first case in production, measured against an indicator chosen with you.
Rollout in waves, training, and regular review of gains and risks.
AI integrated into your existing tools and processes, with the rigour of our cybersecurity practice: we help you choose the most suitable AI tool, with no bias towards any vendor, then deploy it with clear rules: sensitive data stays under your control, access rights are respected and usage is traced.
It depends on the use, not the brand. We compare the tools on your real use cases: quality of answers in your line of work, data protection, hosting location, integration with your tools and cost per user. It is possible to combine two tools, for example a general-purpose assistant for writing and analysis, and a search tool with sources for monitoring.
It depends on the tool and the plan you subscribe to. The professional plans of the main vendors generally state that your data is not used to train the models, which is not the case for all free or consumer versions. We check and document the conditions of each chosen tool before any deployment.
Not necessarily. The assessment often identifies automations that can be done with the tools you already have, before you buy licences for the profiles that will get the most value from them.
Start with a framework, not a ban. The assessment lists the real uses, identifies the data to protect and ends with a simple acceptable use policy that your staff can apply.
Those that combine high volume and low risk: summaries of meetings and emails, an assistant on internal documentation, handling of recurring requests. They demonstrate value within a few weeks.
Before starting, we choose an indicator with you (time per task, number of re-keyed entries, response time) and we measure it before and after go-live.
Its obligations apply in stages, notably awareness for the people who use AI systems. We check with you which ones apply to your use.