Bosch kept the supplier problem inside the AI Academy

A specialist course that finishes before anyone touches a live shortage teaches a skill the plant never sees. The person goes back to the old desk. The late delivery is still someone else's morning.

Bosch's account, published on 16 April 2026 and presented as part of its 2025 annual report, follows the other order. Juliane Lüder, 39, in Abstatt, is building a forecast for parts that may not arrive. "Before, around 40 to 50 percent of deliveries went wrong," she says. The model uses past supplier data to say how many parts will arrive, and how reliable that figure is. It reads historical orders, delivery times and outside factors such as holidays or reported transport bottlenecks.

The page then gives one unit two headcounts. In the story, global supply chain management has some 37,000 associates, supplying 250,000 customers on time every day from around 230 plants. A later line says some 35,000. The two figures are not reconciled, so neither is settled. The plant detail is clearer. The model is already running in Wroclaw, where production-critical parts arrive from hundreds of suppliers, and the planners there refine it. A wider rollout is conditional. The page says "if this works."

She got the skill at the AI Academy, Corporate Research's internal programme for AI specialists. Lüder had trained in business administration and worked in Vehicle Motion. She applied as a controller, with basic self-taught programming and a supplier-risk project already in mind. Eighteen months of training ran beside the regular job: programming, machine learning and putting models to work, with the project phase spent on that forecast. The page says the former controller can now call herself a data scientist. "The AI Academy gave me the opportunity to take my career in a completely new direction."

Martin Thomas, head of the Academy, states the mission as upskilling associates in every business sector to become excellent data scientists and engineers. Domain experts, he says, are irreplaceable. The best solutions appear when AI skills are added to knowledge they already have. Those people are the beacons others call.

Some 100,000 associates have completed Academy training. Lüder is one of them. The offer runs from basic courses for the wider workforce to expert programmes for associates and executives. In Mobility, 120,000 associates are scheduled for basic generative AI training "this year", via an initiative started by the Power Solutions division. The page was modified on 18 August 2026 and does not name the calendar year behind that phrase. Scheduled is not finished. A FAQ says every associate can take part, including expert programmes that put AI onto a real problem in the person's own division.

The cost line is narrower than the headcount. Those 18-month expert programmes require a mid-five-figure investment per participant. No currency is named. An internal analysis says that within the first five years the projects from graduates generate ten times the training cost in value. That is their analysis of the expert track. It is not a finding about the 100,000, and it is not an outside audit.

1. Put a live problem on the application. Lüder brought supplier risk from Vehicle Motion and spent the project phase on that forecast. A completion with no problem attached is a different programme.

2. Keep the 100,000 off her record. The count says how many people finished Academy training. It does not say they can forecast a late delivery, and the page does not claim they can.

3. Call a schedule a schedule. The 120,000 in Mobility are due basic generative AI training. Due is not Lüder's 18 months beside the job, and it is not a model running in Wroclaw.

4. Do not promote an internal multiple into a return. Ten times the cost, on a mid-five-figure spend, is what Bosch says about expert-programme graduates over five years. It does not cover a basic course.

An AI training session is a practical place to put one live shortage in the room with the person who owns it, and to write down what they must be able to do when the practice ends.