Two-tap human validation

The problem

The cultural brake is not the technology, it is who writes to the master.

The fear that AI will enter data on its own and contaminate the master ERP is the main cultural brake on digitization, in an environment where any mis-crossed data can become a customer nonconformity.

• The fear that an AI will enter data on its own and contaminate the MES master is the main cultural brake on digitizing a structural line, and it is a reasonable fear.
• In this sub-sector a mis-crossed datum is not a typo: a dimension attached to the wrong batch, or a heat number tied to the wrong part number, becomes a customer nonconformity months later.
• So the honest alternative today is to keep typing by hand, which is slow, late and also wrong — just wrong in a way people are used to.
• And when the OEM asks who validated a datum on a crash-relevant part, a system with no named human behind each record has no answer to give — which is a problem long before anybody starts talking about artificial intelligence.

How it fits the IRIS system

Connect with early human verification — two taps at the line.

Connect early human verification mode: every capture, whether photo, voice or email, goes to the tablet with a visual summary before crossing into the master. UI designed for two-second validation.

This is the piece that guarantees the master is never contaminated and that every datum has a responsible person by name. Without it, the other eleven cases do not get approved: nobody in a plant audited under IATF 16949 signs off on AI writing unattended into the system quality depends on.

See the full IRIS architecture →

Before and after

Unvalidated capture versus confirmed capture

Aspect	Today	With iLEAN
Who writes to the MES master	A person, typing late	A person, confirming in seconds
Time it costs the operator	A walk to a terminal	Two taps on the line tablet
What the record proves	That somebody typed it	What summary they saw and confirmed
A mis-crossed datum	Surfaces as a customer claim	Caught on screen before it enters
Photo, voice and email captures	Each with its own route	One single validation gate
Corrections made by the operator	Lost	Stored, and they improve extraction

From unvalidated data contaminating the ERP to validated data with seconds of latency, a clean master.

Impact estimate

Impact estimate — to be validated with your numbers.

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: this is an enabling piece for the whole iLEAN matrix, with strategic value rather than a directly monetizable saving. That is the brief's own framing.
• What it buys is approval: it is the reason the other eleven cases become deployable in a plant under continuous audit.
• The master stays clean, and every record carries the name of whoever confirmed it.
• And unvalidated data stops reaching the MES with hours of latency, which is the state that produces untraceable nonconformities.

Enabling piece for the whole iLEAN matrix; strategic value, not directly monetizable.

And the fair question from the production manager

“If a person confirms in the end, what is the AI actually saving?” — the work is not in deciding, it is in assembling. Today somebody has to walk to a terminal, find the batch and type. Reading and pre-filling the summary is an anchored task, where the best models drop below 1.5% error [1]; the operator only confirms or corrects, and every correction is stored so extraction improves exactly where it failed.

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