Nothing enters the central system without human validation

The problem

The real blocker is not technology, it is a reasonable fear.

The main cultural blocker to digitalization is the fear that AI enters data on its own and someone later has to clean up the ERP master — especially when the data affects a food-contact spec. But the operator doesn't want to go back to typing either.

• The main cultural brake on digitalization is the fear that AI enters data on its own and somebody later has to clean up the ERP master.
• In metal packaging that fear is well founded: a wrong varnish code or gauge on a food-contact spec is not a typo, it is a potential migration nonconformity.
• At the same time, the operator does not want to go back to typing at a terminal several stations away from the forming line.
• Between the two, data ends up either unvalidated or never captured, and both leave the master less reliable than before.

How it fits the IRIS system

Connect with early human verification — a two-second summary, two taps in metal packaging.

Connect early human verification mode. Every capture (photo, voice, parsed email) shows up on the tablet as a 2-second visual summary. The lead confirms with two taps — enter the ERP or correct before entering.

This is what separates iLEAN from an autonomous AI agent. The AI assembles the record; a person on the line decides whether it enters. Without that gate, none of the other eleven cases would get signed off.

See the full IRIS architecture →

Before and after

Data entry at the forming line, before and after the validation gate

Aspect	Today	With iLEAN
Who writes to the ERP master	Whoever types, often later	A line lead, confirming a summary
Varnish and gauge on a food-contact spec	Typed, sometimes wrong	Shown in the summary and confirmed
Time per record at the line	Minutes at a remote terminal	About two seconds, two taps
A wrong extraction	Cleaned out of the master afterwards	Corrected before it enters
Captures from photo, voice and email	Each with its own path, or none	One validation gate for all of them
Trail of who approved what	Not kept	Name, time and correction stored

Unvalidated data contaminating the ERP, or data not captured due to distrust → data captured and validated in seconds, clean master.

Impact estimate

No standalone payback — it is the piece the rest depends on.

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.

• This case has no standalone payback: it is the enabling piece of the whole matrix and its value is strategic, not directly monetizable.
• Without early human verification, none of the other pieces gets signed off by quality or by management.
• It turns “the AI enters data” into “the AI prepares, a person confirms”, which is a conversation a food-contact plant can actually have.
• Result: data captured and validated in seconds, and an ERP master that stays clean.

The enabling piece of the whole iLEAN matrix: without this early human verification, none of the other pieces get signed off. Strategic value, not directly monetizable.

And the fair question from the production manager

“If a person confirms everything, what is the AI actually doing?” — assembling. Today somebody has to find the run, open the screen and type each field. Here it arrives prepared from the photo, the dictation or the email; extracting those fields is an anchored task where the best models drop below 1.5% error [1], and the lead only confirms or corrects. Every correction is stored and feeds back into extraction. That trail also answers the question every auditor asks: who approved this record, when, and on what.

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