The last machine to touch the product should not depend on typing

The problem

The last machine to touch the product is fed by hand on every reference change.

With a catalog of dozens of references, frequent format changes (bulk IQF, one-kilo cases, half-kilo packs), two own brands and private label production, coding changes run into dozens per week. One mistyped digit in a lot, an expiry calculated from the wrong date, or an allergen list belonging to the previous recipe. Each one of those three, on its own, is grounds for a recall. And as soon as exports are involved, that risk multiplies by every market, every labeling language and every template somebody maintains by hand.

• The coder that prints lot, expiry and allergens on every case is usually modern, usually has an API and is usually isolated. So on every reference change somebody types the lot into its panel, with the same data already sitting in the ERP.
• With a catalog of dozens of references, frequent format changes — bulk IQF, one-kilo cases, half-kilo packs — two own brands and private label production, coding changes run into dozens per week.
• One mistyped digit in a lot, an expiry calculated from the wrong date, an allergen list belonging to the previous recipe. Each of those three, on its own, is grounds for a recall.
• And as soon as exports are involved, that risk multiplies by every market, every labeling language and every template somebody maintains by hand.

How it fits the IRIS system

Connect stitches the ERP and the coder together by API, replacing neither.

Connect stitching two modern systems together by API, replacing neither.

The allergen declaration printed on the case comes from the recipe master of the active order, not from whatever was on the coder from the previous run. That single change removes the costliest incident in the sub-sector, and it does so without replacing a machine that already works. For a plant that labels for several export markets, it also means the wording of each market is maintained once, in the master, instead of once per template on the coder.

• When the ERP marks the start of the order, it publishes to the coder the reference, the lot, the dates calculated from that product family's shelf life and the allergen declaration of that recipe.
• Language and format are resolved by destination market, from the recipe master.
• A lot change becomes a tap in the ERP rather than typing on a machine panel.
• Case 12 adds the read-back ring: a camera checks that what is printed matches the active order.

See the full IRIS architecture →

Before and after

The isolated coder versus the stitched coder

Aspect	Today	With iLEAN Connect
Lot and expiry	Typed on the coder panel	Published from the active order
Allergen declaration	Whatever the last run left	The recipe master's, per order
Export market language	A template maintained by hand	Resolved by destination market
A lot change	Typing on a machine panel	A tap in the ERP
Coding errors of human origin	Several a month	Zero
Either system	—	Neither is replaced

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.

• Estimated payback 3-6 months, one of the fastest in the matrix.
• A drastic reduction in mislabeling recall risk — the costliest incident in the sub-sector, because it hits allergen safety and the distributor relationship at the same time.
• Multi-language labeling for export stops depending on templates somebody maintains by hand.
• And the dozens of weekly coding changes stop being dozens of chances to type the wrong digit.

fast payback, 3 to 6 months, and a drastic reduction in mislabeling recall risk — the costliest incident in the sub-sector. *Estimate to be validated*.

And the fair question from the production manager

“Isn't a classic integration cleaner?” — it is the day the project reaches the top of the IT list, and that is the problem: the coder is the classic modern island that stays isolated because there was never time. This stitching leans on the API the coder already has, asks for no shutdown window and replaces nothing. There is no reading model involved: the data travels as data, so the note [1] that applies to captured documents does not even come into play here — and case 12 adds the read-back camera as the second net.

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