The coder stitched to the ERP

The problem

Two modern systems, and a person typing between them.

the coder is the last physical step before the customer. One fat-fingered entry on the panel, especially if the container carries a different allergen than the previous shift, can end in a recall.

• The coder that prints batch and best-before date on every jar, bottle or can usually has an API, and so does the ERP. Yet they do not talk.
• So at every batch order start someone types batch, date and, on some lines, the allergen text into the coder panel.
• The coder is the last physical step before the customer: one wrong digit is a jar the plant can no longer trace.
• The real danger is a changeover from a mustard dressing to a sauce without mustard, or the reverse: a run printed with the previous shift's data can end in a recall.

How it fits the IRIS system

Connect links ERP and coder through their APIs — no hardware replaced.

Connect stitches the two modern systems together via API. The moment the ERP flags a batch order start, it automatically publishes batch, date and the allergen list to the coder.

The moment the ERP flags a batch order start, iLEAN publishes batch, date and allergen list to the coder. Nobody types. What gets printed is what the batch order says, and the coder's confirmation is stored with the batch. When the line changes from a mustard dressing to a ketchup, the coder changes with it, with nobody at the panel.

See the full IRIS architecture →

Before and after

Coding by hand today versus coding from the batch order

Aspect	Today	With iLEAN Connect
Entering batch and date	Typed at every changeover	Published from the ERP
Allergen text on the pack	Copied from the formula sheet	Taken from the active recipe
A transposed digit	Periodic, sometimes a recall	Zero human-origin errors
Changeover at the coder panel	Minutes of typing and checking	Automatic
Proof of what was printed	The operator's memory	Stored against the batch
Coder and ERP	Isolated	Both kept, now linked

periodic typing errors, some triggering recalls → zero human-origin coding error.

Impact estimate

Impact estimate — to be validated with your recall and complaint history.

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.

• Estimated payback 3-6 months, one of the fastest in the matrix.
• It comes from a drastic drop in recall risk caused by a mislabeled batch, date or allergen.
• Coding errors of human origin go to zero, because no human types the data anymore.
• And changeovers at the coder stop costing minutes of typing and double-checking on every format change.

fast payback, 3-6 months, from a drastic drop in recall risk caused by mislabeled batch, date or allergen. *Estimate to validate.*

And the fair question from the production manager

"What if the ERP sends the wrong batch order?" — then the problem was upstream and it would have been typed wrong too; the difference is that now it is visible. The data moves by API, with no reading or interpretation, so there is no model error on that path. Where a model does take part, extracting the formula from a photo upstream, it is an anchored task with the best models under 1.5% error [1], and a person confirms it at the line tablet before it ever reaches the coder. The one field nobody types anymore is the one a tired hand used to get wrong.

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