Early human verification: two taps before it hits the ERP

The problem

Nobody wants an AI approving a first-off on a brake part by itself.

the main cultural brake on digitizing safety-part machining is the fear of an AI validating the first-off part on its own and letting an out-of-tolerance lot through. But operators don't want to type either.

• The biggest cultural brake on digitizing safety-part machining is fear: an AI that validates the first-off alone and lets an out-of-tolerance lot of brake pistons through.
• But operators do not want to type either, so today data reaches the ERP late and is entered by someone who never held the part or saw the gauge.
• A lot started without a real first-off confirmation is the hole a containment escapes through, and it rarely shows until the OEM finds it.
• And when the OEM asks who approved it, the answer has to be a name and the evidence they saw, not a tick in a box typed hours later.

How it fits the IRIS system

Connect with early human verification — a visual summary and two taps at the lathe.

Connect's early human verification mode. Every capture — a work order photo, a voice-dictated issue, a supplier alert — reaches the tablet as a visual summary before crossing into the master data; UI designed for two-second validation.

This is what separates iLEAN from an autonomous agent: nothing it captures enters the system unseen. Without this piece the other eleven never get approved, because nobody signs off on AI writing into the master data of a safety part — and nobody should. The two taps are the price of trust, and operators find it a fair one.

See the full IRIS architecture →

Before and after

Data entered later versus data confirmed at the lathe

Aspect	Today	With iLEAN
Who validates the first-off data	Whoever types it	Whoever measured the part
Time to confirm	A trip to a terminal	About two seconds, two taps
Out-of-tolerance dimension	Seen when typed, if at all	Flagged red on the summary
ERP master data	Polluted by unchecked entries	Only confirmed data crosses
When the OEM asks who approved	A tick in a box	A name, a time and the summary shown
Corrections to the AI	Not recorded	Stored and used to improve extraction

unvalidated data contaminating the ERP, or a lot started without first-off confirmation → data captured and validated within seconds, clean master data.

Impact estimate

Strategic value — no payback of its own, and we say so.

The block below is an estimate to be validated against your plant's actual data. We put it forward so the committee has an order of magnitude; we refine it during the assessment.

• No standalone payback. It is the enabling piece of the whole matrix, and we do not dress it up with a number it cannot defend.
• Without it, none of the other cases can be trusted into production on a safety part, so its value is the value of everything it unlocks.
• Unvalidated data stops contaminating the ERP: what crosses has a person's confirmation behind it. Nothing reaches the master data on assumption.
• And a lot never starts without a first-off confirmation that can be shown to an auditor, with name, time and the exact summary that was approved.

the enabling piece of the whole iLEAN matrix — without it, the rest of the cases can't be trusted into production. *Strategic value, not directly monetizable*.

And the fair question from the production manager

“If a person confirms everything, what is the AI saving?” — the assembly, not the decision. Reading dimensions off a first-off sheet is an anchored task, where the best models drop below 1.5% error [1]; the operator only checks a summary that is already built, and every correction is stored so the system learns exactly where it fails.

^[1] OpenAI paper "Why Language Models Hallucinate", 2025 — on the reliability of AI in anchored tasks.