Two-Tap Early Validation

The problem

The fear is not AI. It is a wrong allergen in the master data.

The main cultural barrier to digitizing a formula plant is the fear that AI will enter data on its own and contaminate the ERP, especially when the data is a declared allergen.

• The main barrier to digitizing a formula plant is cultural: the fear that the AI will write data into the ERP on its own and contaminate it, and that nobody will notice until a customer does.
• In functional ingredients that fear has a precise shape: a declared allergen that is wrong in the system ends up on the specification sheet your customer uses to label their own product.
• So quality blocks automatic capture, or the plant keeps typing everything by hand out of distrust, and the data that would have helped is simply never captured.
• Either way the master data stays dirty: unvalidated values that slip in through one door, or validated values that never get entered through the other.

How it fits the IRIS system

Connect with early human verification — two taps on the floor tablet before anything is stored.

Connect early human verification mode: every capture goes to a tablet on the floor with a 2-second visual summary, highlighting any allergen or out-of-tolerance value, and the operator confirms or corrects.

This is the case that makes every other Connect case acceptable to quality. Nothing reaches the ERP until a person on the floor has seen it and confirmed it, and allergens and out-of-tolerance values are the first thing on the screen, not something buried on line 23 of a list. Trust is built per capture, not promised in a meeting, and quality can see the correction rate for each type of capture before deciding anything else.

See the full IRIS architecture →

Before and after

Data entry today versus capture with two-tap validation

Aspect	Today	With iLEAN Connect
Who confirms a captured formula value	Nobody, or a typist later	The operator, on the spot
Mustard, soy or milk in the formula	One line among thirty	Highlighted at the top
Out-of-tolerance quantity on a premix line	Seen at the lab, if at all	Flagged before confirming
Time to validate a batch summary	—	About two seconds
Corrections to captured values	Overwritten without a trace	Logged with who and when
Allergen fields in ERP master data	Dirty or incomplete	Clean and validated

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

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: this is an enabling piece for the whole iLEAN matrix. Its value is strategic, and we do not put a month count on it.
• It removes the main reason quality blocks digitization in a formula plant: data reaching the ERP without anyone looking at it.
• Validated data with seconds of latency, instead of either unvalidated data or no data at all, and master data that stays clean as volume grows. That is what lets the weighing sheet, receiving and lab cases go live without quality vetoing them.
• And every confirmation and correction leaves a trace, which the evidence pack and the allergen flagship use directly as proof of human review.

Enabling piece for the whole iLEAN matrix · strategic value, not directly monetizable

And the fair question from the production manager

"Won't operators just tap Confirm without reading?" — that is the right worry, and it is why the screen shows only what matters: allergens and out-of-tolerance values first, everything else collapsed. Extraction itself is an anchored task, where the best models drop below 1.5% error [1], so the operator is checking exceptions, not retyping a sheet, and a pattern of instant confirmations on flagged lines is reported to quality.

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