Lot, date and allergens stop depending on someone typing

The problem

It is the last thing that happens before the door, and it still depends on typing.

Coding combines two features that make it dangerous: it's the last thing that happens before product leaves the plant, and it's one of the few steps still depending on manual typing. An error there isn't caught downstream, because there is no downstream — it goes out of the door. The allergen case is especially delicate in a group making culinary products with and without gluten and canned goods on different recipes. The printed list has to match exactly the recipe running at that moment, not the previous shift's or the next order's. There's no tolerance margin. And the integration has been pending for years at each plant, not through negligence but through prioritisation: there's always something more urgent than connecting a machine that "already works". Which is exactly why the fix has to be cheap and fast, not a multi-year project.

• An error there is not caught downstream because there is no downstream: it goes out of the door.
• One wrong digit means half a production run with the wrong batch code.
• Allergens are the delicate case: the printed list has to match the recipe being produced right now, not the previous shift's and not the next order's. There is no tolerance.
• The integration has been pending for years, not through negligence but through prioritization: there is always something more urgent than connecting a machine that “already works”.

How it fits the IRIS system

Connect stitching two modern systems through the API, replacing neither.

Connect stitching two modern systems via API, replacing neither. The flow:

The pattern replicates plant by plant even when each has a different ERP: solved once and deployed N times, instead of N bespoke projects.

• The ERP marks the start of the production order.
• Connect automatically publishes to the coder the lot, the packing and best-before dates, and the allergen list of the active recipe.
• A lot change becomes a tap in the management system, not typing on a machine keypad.
• What was printed is recorded and available for cross-checking with in-line visual verification and for the audit evidence pack.

The pattern replicates plant by plant identically, even though each has a different ERP: solve once, deploy N times, instead of N bespoke projects.

See the full IRIS architecture →

Before and after

Isolated coder vs. stitched coder

Aspect	Today	With iLEAN Connect
Batch change	Typed into the machine panel	One tap in the management system
Where the data comes from	The memory of whoever types	The active order
Allergen list	Whatever the last shift left	The active recipe's
Typing errors	Several a month per plant	None of human origin
What was printed	Not recorded	Recorded and cross-checkable
Rollout across N plants	N bespoke projects	One replicable pattern

several typing errors a month per plant, some carrying recall risk → zero human-origin coding errors. One bespoke integration per plant → one replicable pattern.

Impact estimate

Impact estimate — to validate against 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.

• Relatively modern coders, with an API, but isolated from the ERP in every plant.
• Indicative payback between 3 and 6 months: among the fastest in the matrix.
• And a sharp reduction in recall risk from mislabeling, which is the argument that carries weight when allergens are involved.
• To be checked against the real history of coding errors of the last few years.

fast payback, estimated at 3 to 6 months, and a drastic reduction in recall risk from mislabelling. *Estimate to validate* against the coding error history.

And the fair question from the production manager

“What if the integration leaves the coder down?” — it is stitched through the API the machine already exposes, without touching its control, and it is tested first on one line during a batch change that was already planned. If something does not fit, it shows on that line and not across the plant. The coder still accepts manual typing exactly as before.

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