Two taps: the human gate that makes everything else signable

The problem

The blocker is never technical. It is the well-founded fear that AI writes on its own.

Because the brake is never technical. It is the entirely reasonable fear that the AI will push data in on its own and someone will then have to clean up the ERP master. But at the same time the operator does not want to type either, and with gloves on, even less. Both things are needed at once: effortless capture and real human control.

• The entirely well-founded fear that AI will enter data by itself and somebody will then have to clean the ERP master.
• But at the same time the operator does not want to type either, and with gloves on even less.
• Both things are needed at once: effortless capture and real human control.
• Any solution that adds administrative work to the day is abandoned within weeks, and the plant goes back to the parallel spreadsheet.

How it fits the IRIS system

An interface designed to validate, not to capture.

It takes the form of an industrial tablet beside the pelletiser and another at goods-in. When iLEAN reads the blend sheet, when it reads the extruder panel, when it parses the broker's message, all of it shows up there as a two-second visual summary with its confidence level. The operator confirms with two taps: accept, or correct before it goes in. The interface is designed to validate, not to capture. That distinction is everything: validating takes two seconds and needs no training, capturing takes minutes and needs an ERP licence. And in a plant where safety is a hard line, two taps on a fixed tablet are two taps nobody takes with their hands where they should not be.

That distinction is everything: validating takes two seconds and needs no training; capturing takes minutes and does. Which is why one is still alive in month six and the other is not.

See the full IRIS architecture →

Before and after

Data entry into the ERP, before and after

Aspect	Today	With iLEAN Connect
Who validates critical data	Whoever types it, if anyone	The operator, in two seconds
What the operator is asked to do	Capture: minutes and training	Validate: two taps, no training
With gloves on	Not viable	Designed for exactly that
Confidence level of what was captured	Not visible	Shown with the summary
What happens to the correction	Corrected and forgotten	Stored, and it tunes the anchored model
Plant trust in the system	Fragile: the parallel spreadsheet appears	It holds, because the signature is theirs

Impact estimate

Impact estimate — an enabling piece, and it is worth saying so.

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 does not monetise directly and we are not going to pretend it does.
• Its value is strategic and it is the largest of the set: without early human verification, the other eleven are not signable.
• It is what turns "the AI captured this" into a record with an author, a timestamp and a signature.
• And it is what stops the plant from keeping the parallel spreadsheet, which is where these projects die.

Enabling piece · without early human verification the other eleven never get signed off. This example does not monetise directly and we are not going to pretend otherwise. Its value is strategic, and it is the largest in the pool: without early human verification the other eleven pieces never get signed off in committee. It is the condition that turns an AI promise into a system an operations director can defend. Conservative ranges, an estimate to validate against the plant's real data.

And the fair question from the production manager

"If a person has to look at it anyway, what do we gain?" — looking is not typing. What they get is a two-second summary with its confidence level, and the anchored extraction work is already done [1]. You go from transcribing to confirming. And the data lands in seconds instead of at shift close.

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