At a glance
- Company
- EISTMaker of custom sheet-metal parts and industrial equipment
- Size
- Micro and small (1–49 employees) The official case page gives the size as five employees and one owner-manager.
- Industry
- Manufacturing
- Kind of work
- Pricing and forecasting
- Where and when
- France. A project of six to eight months, published June 2026.
- Who produced the figures
- The company itselfThe figures are the owner’s own estimates, published by France Num, the French government’s programme for the digital transformation of small firms.
See alsoHerges: an AI model estimated production times five times more accurately than the firm’s experts(Company case) · Rue La La: machine-learning price recommendations raised revenue by about 10% in a test(Company case)
What they did
EIST is a very small firm in Lyon. It makes custom sheet-metal parts and industrial equipment. Every job is different, so every request needs its own quote. Some took several hours to price. The owner, who took the firm over three years earlier, says: “We sometimes missed opportunities because I did not have the time to reply.”
He saw a video of AI estimating the cost of a part from its 3D file, and decided to try it. The project ran for six to eight months. The aim was for the AI to use the firm’s own history to predict, for a new part:
- the manufacturing steps,
- the production time,
- the material cost,
- the margin,
- the final price.
Before any of that could work, the data had to be put in order:
- cleaning the records in the firm’s business software;
- structuring the parts lists;
- linking each 3D file to the record of how the part was actually made;
- tidying up the standard manufacturing routes.
The owner ran the project himself with his assistant and did not hire a consulting firm. The region and a local business-support body gave advice and funding.
The numbers
- About 10 minutes for a quote, where it sometimes took two hours.
- About 90% of quotes are reliable, by the firm’s estimate.
What was hard
The owner says the main difficulty was the data, and the technology came second. “If the data is not clean, the tool will not be reliable.” The firm also had to learn, step by step, how far to trust the results.
What the case shows
- Six people can do this. The owner ran the project himself with his assistant, without a consulting firm.
- The firm’s own records were the raw material. The AI learned from past jobs, so the past jobs had to be recorded properly first.
- One task, chosen because it cost orders. Slow quotes meant lost customers.
- The owner still signs off. The AI prepares the quote, and he checks and adjusts it.
What this says about the practices
Practices the company applied
- Start with one costly process, not a company-wide rollout The owner picked one task: pricing custom parts, which could take hours per request and was costing the firm orders. Do first · Effort: medium (weeks, a few people)
- Keep a person checking anything that goes to a customer or carries legal weight Quotes are still checked and adjusted by the owner before they are approved. Do first · Effort: medium (weeks, a few people)
- Budget for people and process, not just tools A large part of the work was preparing data: cleaning records, structuring parts lists and linking 3D files to production history. Do first · Effort: low (days, one person)
- Put your own records in order before you ask AI to predict from them Before the AI was used, the firm cleaned the records in its business software, structured its parts lists and linked each 3D file to the record of how the part was made. Do next · Effort: medium (weeks, a few people)
What to take from it
If someone cites EIST as proof that AI pays off, keep in mind that the company published the figures itself. Treat the case as an example of what one company did, not as a forecast of what AI will do for you. A practice applied here, to check in your own plans: Start with one costly process, not a company-wide rollout.
Read next
- Herges: an AI model estimated production times five times more accurately than the firm’s expertsA family-owned steel fabricator prices one-off jobs from its experts’ estimates of how long the work will take. A public support centre helped it record real times and train an AI model on them. The model’s estimates were five times closer to the real figure than the experts’ own.
- Rue La La: machine-learning price recommendations raised revenue by about 10% in a testRue La La, an online fashion retailer, had a tool built that predicts demand for products it has never sold and recommends a price for each. In a five-month test, researchers estimated that revenue on the tested products rose by 9.7%. The retailer set firm limits on what the tool could do before it let it near a price.
- Zillow Offers: home-price forecasts missed, and Zillow shut its home-buying businessZillow bought homes to resell, at prices set with its own algorithms and in-person checks. In 2021 its forecasts of resale prices missed by far more than it had planned for. It wrote down $407.9 million of homes, shut the business and cut about a quarter of its staff. Zillow named labour and supply problems as well as the forecasts.
Sources
- France Num (9 June 2026). Comment une TPE industrielle utilise l’IA pour réaliser un devis en 10 minutes au lieu de 2 heures. Case page, in French. francenum.gouv.fr