Nothing enters the system without a human seeing it

The problem

Nobody wants AI writing serials into the ERP, and nobody wants to type them either.

The big fear is that an AI writes data into the ERP on its own and someone then has to clean the master of serial numbers and configurations. Yet nobody wants to type either.

• The fear is reasonable: an AI that writes generator serials, frequencies and configurations into the ERP on its own, and someone then has to clean the master by hand.
• In an equipment maker that master is the basis of warranty, spare parts and service: a wrong frequency on a serial sends the field technician with the wrong replacement.
• Yet nobody on the floor wants to type either, so data is entered late, partially or not at all, usually by someone who did not see the machine and copies from a traveler or a nameplate photo sent by chat.
• The result is a machine master half captured and half distrusted, which is exactly what blocks approval of every other automation. Nobody signs off a camera or an agent that writes into a master already in doubt.

How it fits the IRIS system

Connect with the station tablet — a review, not a form.

With iLEAN Connect, everything captured, by photo, voice or email, first appears on a tablet at the station as a two-second visual summary. The technician checks the generator serial, frequency and configuration, and confirms or corrects with two taps. Only then does the data cross into the central system. The screen is not a form: it is a review. No training needed and no change to the station's routine.

Everything captured by photo, voice or email first appears on the station tablet as a short summary. The technician looks, confirms or corrects with two taps, and only then does the data cross into the central system. The person stays in command of the master, and the master stops needing a cleanup project. Service, spare parts and warranty finally work from the same serial-number data the floor confirmed.

See the full IRIS architecture →

Before and after

Data typed late versus data reviewed at the station

Aspect	Today	With iLEAN Connect
Generator serial and frequency	Typed later, sometimes wrong	Reviewed at the station
Stack configuration per machine	Partly captured	Confirmed before it is stored
What the technician does	Fills a form or skips it	Reviews a two-second summary
Who signed which serial	Nobody can say	Named person and time
Machine master for service	Distrusted	Clean
Training on the station screen	—	None needed

Before: unvalidated or uncaptured data. After: data validated in seconds and 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: it is an enabling piece with strategic value, and we do not attach a figure to it on its own.
• It is the piece that lets every other piece be signed off: without early human review, no quality or IT lead lets a system write into the serial-number master.
• Its measurable effect: data validated within seconds instead of typed hours later, and a clean master of serials and configurations that service can rely on when it ships a spare part.
• And traceability of who confirmed what, at which station and when, which is what a warranty claim or a customer audit asks for when a machine fails years after delivery.

Enabling piece · data validated in seconds, clean master (estimate to be validated). It is the piece that lets every other piece be signed off: data validated within seconds and a clean master. Strategic value; an estimate to be validated.

And the fair question from the production manager

“If a technician confirms everything anyway, what does the AI save?” — the assembling, not the deciding. Reading a generator nameplate, a traveler or a dictated note is an anchored task where the best models drop below 1.5% error [1], so the summary usually arrives right and the technician only confirms. When it is wrong, the correction is stored with its author and becomes feedback for that exact field.

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