An engineering team: making AI a shared matter
Seven of eight employees already used AI, each in their own way, and the gains stopped at each desk. In 3.5 days, an assessment grounded in real work: seven interviews, a team workshop, twelve ranked workstreams and an action plan carried by the team.
Context
The firm has eight employees in Réunion: engineers who run projects and administrative staff. Neither the firm nor the people are named here.
After an introductory training session in 2025, everyone had taken up AI in their own way, with their own tools and their own subscriptions. The gains were real, but they stayed individual. The starting principle, set at the scoping stage: most of the value comes from a well-defined need and well-organised data, not from the tool.
So the question was no longer how to use AI. It was how to make it a matter for the whole firm.
The approach
The assessment starts from real work, not from technologies. Each interview, about an hour long, covered what the person actually does with AI in their work, their difficulties and their needs. Each interview summary was reviewed by the person concerned before the synthesis.
The assessment, in three stages
Beforehand
Scoping with the manager
May 2026
Seven interviews and observations
May 2026
Team workshop: findings, priorities, first hands-on use
June 2026
Report delivered to the whole team
3.5 days of work in total.
Each person’s use was placed on six axes: frequency, tools, autonomy, critical reading of answers, integration into the job, and sharing with the team. These estimates are not used to rank people. They are used to calibrate the support, and they are not published here.
What the analysis of the work showed
What the interviews showed.
The assessment did not show a lack of skill. It showed a lack of sharing: what one person discovers, the team does not build on. Documents are structured differently from one project to the next, input data is not prepared, tools are not shared.
Yet three needs came up with almost everyone, without any coordination: making the firm’s documents consistent, structuring the input data for each project, and lightening the analysis of bids received from contractors.
The analysis also showed where AI adds little. On the administrative side, the accounting chain is already handled by management software. For these roles, the expected gains are not in accounting: they are in writing and in preparing meetings. Low use there does not signal a lag: it reflects work that is already well equipped.
Finally, the maturity gap rules out identical training for everyone. It would miss both the beginners and those already well advanced. The assessment proposes two-speed support: individual to respect each person’s pace, collective to build shared practices.
Three questions the tool does not settle
What data can go into AI? Public documents, yes. Bids received from contractors stay confidential until the choice is made. For personal data, the rule set at the workshop is to remove CVs before any analysis. Bid analysis is therefore both the most profitable use and the most constrained: as long as bids are not anonymised, they do not go into the tool as they are.
How far can the answers be trusted? A language model produces coherent answers, not necessarily accurate ones. The rule adopted: AI does the rough work, a person validates. In legal matters, cited articles, figures and decisions are checked.
What happens to the time saved? The question was raised in an interview by a member of the team. Time saved is not a volume to be redeployed mechanically. An hour saved on an unpleasant task can go into doing the rest of the work better. That choice, better quality of work rather than more intensity, decides whether AI is experienced as relief. It is for the team to settle, in its charter.
The action plan
At the workshop, the team set its priorities and adopted a method: practise first, keeping the points of vigilance in mind, then set the rules. The report spells it out: everyone notes in three lines what works and what gets in the way, regular working groups remove the obstacles one by one, and a designated person updates a shared context every month.
The report ranks twelve workstreams by value and effort. Each action has one or more expected owners within the team.
The action plan
0 to 3 months
a shared professional account, the tools inside Word and Excel, a workshop on bid analysis, a first usage charter
3 to 9 months
shared document templates, a reference file for each project, assisted meeting minutes
Later
Excel automations, a pilot agent for email, anonymised bids
Three horizons, expected owners for each action.
For tooling, two team plans were compared against the firm’s real work, mostly writing and working on files. The tools sit inside Word and Excel so people can work in their usual flow without losing document formatting. Indicative cost: about €160 to €200 excl. VAT per month for eight licences. As the report puts it, what matters is not the subscription but the templates and reference files the team puts into it.
Limits
The maturity estimates are indicative. The gain of about 80% on bid analysis is an estimate given in an interview, without a before-and-after measurement. One role was not interviewed; it will be included in the rollout. How assisted bid analysis, planned for the short term, fits with anonymising bids, planned later, remains to be worked out. The charter guidelines are not a GDPR or AI Act compliance review. This case describes the assessment: the rollout of the plan is not measured here.
And in mainland France?
The assessment carries over to an SME or a team in mainland France, on site or remotely. The method does not depend on the place: interviews about real work, reviewed by each person, a team workshop, an action plan carried by the team.
AI assessment: usually 3 to 5 days, €4,500 to €7,500 excl. VAT. See the assessment
Want to know what AI would change in your work?
First conversation: 30 minutes, by video or by phone, free and without commitment. Reply within 48 business hours.