Manufacturing & industrial
Operators overriding AI-driven quality and maintenance calls
The system flags a batch as within tolerance. The operator has seen this pattern fail before. Who wins, and can you show they were trained for it?
01 — The question you cannot currently answer
You have the security answers. You have the data-protection answers. Then comes a question about the AI literacy of the people delivering the contract, and there is nothing to put in the box.
This is how the AI Act reaches companies that are not based in the EU. Not through a regulator. Through your customer. They carry the Article 4 obligation, that obligation covers “other persons dealing with the operation and use of AI systems on their behalf,” and on their contracts that is your team.
A generic AI-awareness certificate does not close this. The Act asks for a sufficient level — calibrated to the person and the context. A certificate evidences neither.
02 — Signal starts from your operations, not a template
Article 4 requires literacy calibrated to “the context the AI systems are to be used in.” A bought-in course cannot satisfy that, because it does not know your context.
Signal ingests what you already have — standard operating procedures, incident and escalation logs, expert frameworks, decision scenarios, SME interviews — and derives which capabilities your people actually need to demonstrate.
The output is a Skill Demand Register: a defensible statement of what good judgment looks like in your operation, sourced from evidence rather than assumed from a syllabus.
03 — The evidence gets sealed, dated and made traceable
Two years from now, someone will ask why you trained these people on these things. “It seemed sensible” is not an answer that survives a review.
Signal seals its findings into an immutable, dated Snapshot. Every capability requirement traces back to the specific document, interview or incident that produced it — so the reasoning is inspectable long after the people who did the reasoning have moved on.
This is the difference between having done training and being able to show why that training was the right training.
05 — One programme, four depths — because the Act says so
Read the operative wording: literacy sufficient “taking into account their technical knowledge, experience, education and training and the context the AI systems are to be used in, and considering the persons or groups of persons on whom the AI systems are to be used.”
That is a specification for differentiation. An organisation that puts 900 people through one identical module has not read it — and has produced a record that reads, to a reviewer, as exactly what it is.
Forge builds the tiers separately. The person who can override an AI output goes deep. The person who supplies inputs to it does not. Both are covered, and the difference is on the record with the reasoning attached.
06 — Structured down to the individual decision
Programme → module → section → interaction → decision practice. Every branch of the structure terminates in a moment where a person has to decide something.
That matters for evidence. A module you can only report on at programme level tells a reviewer that someone finished it. A structure that resolves to individual decisions lets you say which judgments were practised, by whom, and how they went.
It also makes the programme maintainable. When the regulation shifts — and it has already shifted once via the Omnibus — you amend the affected moments rather than rebuilding the course.
08 — Overriding a machine is a skill. It has to be rehearsed.
Article 14(4)(d) requires that an overseer be able to “decide, in any particular situation, not to use the high-risk AI system or to otherwise disregard, override or reverse the output.”
Think about what that asks of a person. Contradicting a system that is right most of the time, in front of colleagues, under time pressure, with no certainty they are correct. Someone who has never done it once has not demonstrated they can.
A simulation is the only place they can practise it before it counts — and the only place you can capture proof that they did.
10 — Build it once. Answer every EU buyer with it.
The work in steps 1 to 9 is not per-deal work. It is a capability system you build once and maintain.
The next EU customer sends their annex and the answer already exists — current, sourced, and specific to how your people actually operate. The deal stops stalling in review.
Companies treating this as a form-filling exercise will do it again for every customer, badly, forever. The ones that build the capability once turn a procurement blocker into something closer to a reason to pick them.