Playbook: running a successful AI assessment in an SME

A successful assessment is scoped before it starts. The sponsor is the person who will decide what comes next, often the chief executive: in French SMEs and mid-sized firms, the chief executive leads 73% of AI projects (Bpifrance Le Lab, 2025). Without a decision-maker, the findings remain a report.

The scope is one function, one department or one site, not the whole company: wide enough for the hours at stake to justify the effort, narrow enough for everyone concerned to be interviewed. The Who it is for page gives, by segment, the hours threshold that repays a Diagnostic and a first agent.

• A sponsor who decides and gives the teams time.
• A named scope: who is interviewed, about which tasks.
• A costed stake: hours, lead times or errors that can be counted.
• A timeline without peak activity, works council consultation included.
• A first benchmark, if you wish: the self-assessment.

In French companies with at least 50 employees, the works council (CSE) is informed and consulted on the introduction of new technologies (Labour Code, article L2312-8). Failing an agreement, it gives its opinion within one month, two if it calls on an expert (article R2312-6). That time comes before the first interview: the works council timeline sets it out.

In every case, each employee is informed before being interviewed: no information about them may be collected by a system they have not been told about beforehand (article L1222-4). Say what is observed, tasks rather than people, and what will happen to the time freed up.

The SHIFTLABS Diagnostic takes 2 weeks of data collection; from 50 employees, allow about 1 more month beforehand for the works council information and consultation. An AI agent, which introduces itself as such, conducts 30- to 45-minute interviews; a 5- to 7-minute questionnaire completes them. An expert reviews and validates every summary.

The company’s role: free up time slots, appoint a contact for practical questions, and let the teams talk about the work they actually do, exceptions included. The example interview replays an exchange step by step.

A deliverable is judged by what you can check. Four requirements are enough.

• Use cases costed in hours, with the calculation inputs: volume, time per occurrence, share that can be freed.
• The cases set aside, with the reason: they say as much as the cases kept.
• A first case scoped: its owner, its data, its price, the decision date.
• The status of every figure: an assessment gives an identified potential, not a gain achieved.

The gain is proven after go-live, against a baseline taken beforehand, on a date set in advance. Measuring an AI gain describes the method; how many hours for an AI agent to pay off gives the order of magnitude to aim for.

To build the first agent, the SHIFTLABS price list is published: from €6k to €13k excl. VAT per agent depending on whether it assists, automates or redesigns the process, plus the connections to your systems (up to €3k each), billed once. Nothing obliges you to build it with us.

Six pitfalls, each with the remedy to decide before starting.