Context
TWE manufactures nonwoven industrial textiles. It is a subsidiary of a German group with operations in Germany, Belgium, China, the United States and India. In France, the company runs two sites, one in Mâcon (Saône-et-Loire) and the other in Crépy-en-Valois (Oise), with about a hundred employees in total, serving the construction, automotive and hygiene markets.
Very early on, the company saw AI coming as a defining issue: for rounding out skills, for supporting future employees and, above all, for keeping control of its know-how in an industrial environment where data (products, manufacturing recipes) is the core of its value and must be protected at all costs.
Challenges
To make the use of AI secure, the group’s IT department had already rolled out Microsoft Copilot, a choice driven by the priority given to security and data protection rather than to performance alone. But rolling out a tool isn’t enough to create use. Several challenges emerged:
- Stemming shadow AI: without a controlled tool for work, employees were turning to their free personal accounts, with the risk of leaking data and know-how.
- Ending the underuse of licenses: Copilot was seen as a “super Google”, used at most to write a few emails, far from the productivity gains expected.
- Making AI concrete and accessible for every job, in an organization where people work in operational roles (maintenance, production, logistics, HR) and where each role is held by only a few people (so very specific needs, to be addressed one by one).
- Supporting the change to ease fears (“AI is going to take my job”, “I won’t have the time”) and bring everyone on board, including the teams least familiar with digital tools.
The solution: why Teamstarter?
TWE chose Teamstarter to go from an installed tool to truly productive use, suited to each workstation, with a method built on hands-on workshops run in groups of about ten people, at both plants, with the management and supervisory teams.
The principles of the rollout:
- Start with the managers and the leadership team, to master the use before gradually opening it up to production, on a voluntary basis.
- Find business use cases and learn to create skills / agents: skills teach the AI to carry out a specific task the way you would do it yourself, without having to explain everything again each time.
- Frame the ROI of each task: for each automation considered, the team estimates how long the task takes and how often it comes up, an eye-opener that motivates people and sets priorities.
- Start simple: increase the difficulty step by step rather than aiming too high from the outset, and support employees day to day to get them unstuck.
- Build it to last: quarterly check-ins over a year, to anchor the uses and adapt the use cases to the (very fast) pace at which the technology evolves.
- Frame the use with an AI policy shared with all employees, and a reminder of the golden rule: AI works for you, not instead of you (proofreading, critical thinking and checking remain essential).
“The more agents I build for myself and for my team, the more time we get back, and that time goes into people instead of mindless data entry and admin.”
— Marylin Malsot, HR Manager, TWE
The results
The first skills built show how varied the use cases are, well beyond support functions:
- HR: automatic extraction of indicators from PDFs into Excel (from about an hour and a half to under 10 minutes for both sites); formatting the questions and answers of the works council (CSE) from handwritten documents; automated replies to candidates and interview reports; automated legal monitoring (a summary of what’s new, with useful links), in one click instead of a tracker painstakingly kept up to date.
- Maintenance & purchasing: comparing quotes to find the best offer according to the company’s criteria.
- Logistics: optimizing truck loading and the number of vehicles to book (pallets, heights, volumes).
Beyond the tools, what stands out is the buy-in: despite a high average age at one of the sites, employees who were reluctant at first (or even opposed) signed up and became users. The “wow” effect of the first workshop, then the snowball effect between colleagues, did the rest.
