The retirement of veteran operators takes with it knowledge that is in no manual. How an AI assistant helps capture it, organise it and put it within reach of newcomers.
Basque industry is facing the mass retirement of operators with decades of experience. With them goes tacit knowledge that is not documented in any manual. It lives in the memory and practice of people who will soon leave their posts.
This generational handover puts quality and delivery times at risk. It is also an opportunity to put in writing what was never written down.
The problem of tacit knowledge
Tacit knowledge is what an operator knows how to do but has never had to explain. It includes things like these:
- How a machine reacts under certain conditions, for example after a long shutdown.
- What to do about a specific fault that does not appear in the manual.
- How to adjust a process that looks standard on paper but in practice needs tweaking.
- Whom to notify, and in what order, when something goes wrong.
It is very valuable and hard to transfer the traditional way. A manual falls short. And shadowing a veteran for months is not always possible, because their time is scarce too.
The solution: internal knowledge chatbots
- The senior operator talks to the system, by voice or text.
- The AI extracts, structures and stores that knowledge in a searchable form.
- New operators access it through an internal chatbot.
- The system is updated with each new contribution.
How the knowledge is captured
The assistant guides the conversation with specific questions about each machine or process. The veteran answers in their own words, as if explaining it to an apprentice. The AI turns those answers into organised records. Then someone at the company reviews them before they are published.
To these records are added the documents that already exist: machine manuals, procedures and technical data sheets. Everything ends up in a single knowledge base that can be queried.
How newcomers use it
The new operator asks in natural language, in Basque or Spanish. The assistant replies with the information the company has validated. If the answer is not there, it says so and refers them to the person in charge.
An example of use
A newly hired operator sees an alarm they do not recognise during the night shift. There is no veteran on the shop floor. They ask the assistant from the workstation computer. They receive the steps left by a colleague who has since retired, together with the relevant section of the manual.
If the problem has not been recorded, the assistant tells them so and indicates whom to notify. The next day, the solution that was applied is added to the base. That way the knowledge grows with use.
What you need and how to measure it
What you need
- Time from the veterans for the conversations, spread over short sessions.
- A person to validate what gets published.
- The documents that already exist, even if they are disorganised.
- A decision on who can use the assistant and what information must not leave the company.
The documents are not used to train AI models. We explain this in private AI for companies.
Common mistakes
- Starting too late, when the veteran has already left.
- Trying to capture the whole plant at once. It is better to start with one machine or one critical process.
- Publishing without validating. A wrong answer on the shop floor is costly.
How to measure it
- Time it takes a new person to work independently.
- Repetitive queries that no longer reach the veteran team.
- Incidents resolved with the assistant's help.
Frequently asked questions
Does it replace in-person training?
No, it complements it: the new operator keeps learning on the shop floor with colleagues. They no longer depend on a veteran being free for every question, because the assistant is available on every shift. It answers with information the company has validated. That way in-person training can focus on the practical side.
What about confidential information?
The company decides which documents go in, who uses the assistant and what it must not talk about. The language model works with zero retention: it does not store what it is sent or use it for training. Before we start, we detail which provider is involved in each component and in which country it is hosted.
Can it be tested before rolling it out across the plant?
Yes, and it is what we recommend. We start with one machine or process and a small group of operators. We check with them that the answers are useful and correct. Once it works, it is extended to other lines and more veterans join the conversations.
Does your company want to take part in a pilot?
There are more use cases in AI for industry. If you prefer to start with something simpler, read 5 processes every industrial SME should automate. You can also see AI automation for SMEs.
The working method is set out in how we work. If your company is interested in taking part in a pilot, write to us from the contact page.