Nothing reaches the ERP without human verification

The problem

The fear that stops digitization is a dirty ERP master.

The fear that AI will "enter data on its own and someone will have to clean up the ERP master" is the main cultural barrier to digitizing continuous production.

• On the glazing line, application weight, glaze density, recipe and batch are recorded by hand or not at all, because entering them in the ERP means leaving the line.
• When AI is proposed to capture them, the objection comes first from quality and IT: “it will enter data on its own and someone will have to clean up the master”. IT has seen it happen with earlier tools.
• It is a reasonable fear. A wrong glaze batch on a finished-goods record ends up in a customer complaint about shade, and tracing it back takes days. Meanwhile, the same batch keeps shipping.
• So the plant ends up in one of two bad places: data entered without validation, or data not captured at all out of distrust. Neither one supports the rest of the digitization. And every other case in this matrix depends on getting past that point.

How it fits the IRIS system

Connect with early human verification — confirm or fix, in two seconds.

Connect's early human verification mode. Every capture reaches the tablet with a visual summary before crossing to the master record. UI designed for a 2-second check.

This is what separates iLEAN from an autonomous agent: nothing it captures crosses into the master until a person on the line has seen it. Without that gate, none of the other eleven cases gets signed off. With it, the conversation with quality and IT changes from whether to how. And every confirmation leaves the name of the person who saw the data, which is what a shade claim needs months later.

See the full IRIS architecture →

Before and after

Glazing data today versus glazing data verified on the tablet

Aspect	Today	With iLEAN Connect
Application weight and glaze density	Noted on paper, typed later	Captured and shown for review
Who checks glazing data before it is posted	Nobody, or nobody in time	The glazing line operator
Time per verification	A walk to a terminal	About two seconds
A wrong glaze batch	Found in a shade complaint	Fixed on the tablet
ERP master quality	Cleaned up afterwards	Clean from the start
Origin of every master record	A person typing late	A person confirming on the line

Unvalidated data contaminating the ERP, or data not captured out of distrust → data captured and validated within seconds, 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: its value is strategic and we do not monetize it directly.
• It is the enabling piece of the whole matrix: without it, none of the other cases gets approved by quality or IT.
• Data captured and validated within seconds, and an ERP master that does not need cleaning afterwards. Quality stops spending its first hour of the day correcting yesterday's entries.
• And every correction is stored with its author, which is the trail an auditor or a shade complaint asks for.

The enabling piece of the whole iLEAN matrix — without it, none of the other pieces get signed off. *Strategic value, not directly monetizable*.

And the fair question from the production manager

“If someone confirms everything, what is the AI doing?” — the work is not in deciding, it is in assembling. Reading weights, densities and batches from the line is an anchored task, where the best models drop below 1.5% error [1], so the operator receives the summary already built and only confirms or fixes. And each fix is stored: the system learns exactly where its extraction fails. And no value reaches the master without a named person behind it.

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