Industrial tablet validates ECU assembly

The problem

In a certified plant, data that nobody signed is data nobody can defend.

- Without a clear human gate, any capture automation risks contaminating the production master data. - Under IATF 16949 traceability, every ECU serial number must be auditable without ambiguity, a condition that requires a human to see the data before it enters the system.

• Without a clear human gate, any capture automation risks contaminating the production master data, and in an ERP that feeds shipments and traceability, a wrong record propagates.
• Under IATF 16949 traceability, every ECU serial number must be auditable without ambiguity: which housing, which board revision, which firmware was flashed, which torque went on the cover screws.
• That condition requires a human to see the data before it enters the system, not afterwards when someone stumbles on an inconsistency.
• It is also why many AI projects never get approved at a certified plant: quality cannot sign off on a black box writing into the ERP.

How it fits the IRIS system

Connect's early human verification — two taps at the ECU station.

Connect 'early human verification': the tablet shows a visual summary of whatever was captured through any mode (photo, voice, parsed email) and the supervisor confirms with two taps (confirm / fix) before the data crosses into the ERP/MES.

This is what separates iLEAN from an autonomous agent that feeds in data unsigned, and it is why quality can approve everything else. Whatever Connect captures, by photo, voice or parsed email, waits on the tablet until a person confirms it. Two taps are the price of letting everything else be automated.

See the full IRIS architecture →

Before and after

ECU data entry today versus entry through the tablet

Aspect	Today	With iLEAN Connect
Serial number and firmware version	Typed in, sometimes later	Shown captured, confirmed in two taps
Cover screw torque result	Read off the driver and noted	Displayed with the unit, checked by the supervisor
Functional test verdict	Transcribed from the tester screen	Attached to the serial before release
Firmware flashed vs. firmware released	Checked only if someone remembers	Mismatch shown before confirming
A captured field that is wrong	Found downstream, if ever	Fixed at the station before the ERP
AI entering data with no gate	The reason quality says no	Blocked by design

AI-captured, unvalidated data → data captured and signed off by a human in seconds, with full audit trail of who validated what.

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, and we do not put a month range on it because its value is removing risk, not saving hours.
• It removes the black-box AI risk from the production flow, which is usually the blocking condition for approving AI projects at a certified plant.
• Data captured by AI and signed off by a human in seconds, with a full audit trail of who validated what.
• We size its effect together with your quality lead, on the traceability findings it prevents rather than on time recovered, because one ambiguous serial number in an OEM audit costs more than hours of typing.

Removes the 'black box' AI risk in the production flow, usually a blocking condition for approving AI projects at a certified plant. Estimate to be validated with quality.

And the fair question from the production manager

“Doesn't a human check slow the ECU line down?” — the supervisor does not type anything: they review a summary already extracted, and extracting it is an anchored task where the best models stay below 1.5% [1] error. Most confirmations take two taps; the time goes only to the few fields flagged as doubtful, which are exactly the ones worth a human look.

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