Two Taps Before It Hits the MES

The problem

Nobody signs off AI that writes into the MES on its own.

The fear that AI might 'enter data on its own and then someone has to clean up the MES/ERP master' is the main cultural blocker to digitization in a regulated environment like automotive. At the same time, the operator doesn't want to type either: they need a validation step that doesn't steal cycle time.

• In a regulated environment the first objection to any capture is always the same: if the AI enters a wrong VIN or option, who cleans up the MES master afterwards?
• That fear is reasonable, and it is the main cultural reason digitization projects stall in automotive before they ever reach the line.
• At the same time, the station operator does not want to type either. Every second spent at a keyboard is taken from a cycle measured in takt.
• So data ends up either unvalidated in the MES or not captured at all, and both outcomes show up at the next IATF 16949 process audit.

How it fits the IRIS system

Connect with early human verification — the operator has the last word.

Connect early human verification mode: every capture (photo, voice, parsed email) reaches a station-side tablet as a 2-second visual summary; the operator confirms with 2 taps before the data crosses into the MES/ERP.

This is the piece that makes the other eleven approvable. The AI captures and proposes; a person at the station confirms in two taps, and only then does anything enter the MES or the ERP. The human stays in command, and the master stays clean. It is also what makes the conversation with the floor and with quality a short one.

See the full IRIS architecture →

Before and after

Captures straight into the MES versus captures confirmed at the station

Aspect	Today	With iLEAN Connect
Who confirms a capture	Nobody, or someone days later	The station operator, in seconds
Time taken from the cycle	Typing, or nothing captured	Two taps on a two-second summary
State of the MES master	Cleaned up after the fact	Clean on entry
Captures nobody trusts	Never made	Made and confirmed
Evidence for the process audit	Who typed what, unclear	Who confirmed what, and when
Quality's position on AI captures	Blocked	Signed off with a human gate

Unvalidated data contaminating the MES/ERP, or data not captured out of distrust → captured and validated data with seconds of latency, clean master.

Impact estimate

Impact estimate — the gate every other case passes through, with no payback of its own.

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 put a number on it in isolation.
• Without a human in command, quality does not sign off the other cases, so its value is the value of every case it unblocks.
• Data that used to arrive unvalidated, or not at all, now enters the MES confirmed and with only seconds of latency.
• And the MES and ERP masters stop needing clean-up campaigns before every audit, because nothing enters them without a name and a time.

Strategic value enabling the whole iLEAN matrix; not monetizable on its own. (Enabling value · without the human in command, the rest doesn't get signed off)

And the fair question from the production manager

“Two taps per capture — won't operators just tap OK without looking?” — the summary highlights only the fields that matter, and reading those fields is an anchored task where the best models drop below 1.5% error [1], so most confirmations are genuinely quick. Where confidence is low the field is highlighted and cannot be confirmed blindly, and every tap is logged with the operator and the time.

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