Photo the production order: the lot exists from minute one

The problem

The order that sets voltage and variant waits in a folder until shift end.

In a commercial refrigerator and merchandiser factory, every line startup begins with a printed production order that sets the model, destination voltage and lot size, plus a model-specific build sheet with its variants. Today both sheets sit in the line lead's folder until someone transcribes them at shift end. A transcription error in voltage or variant only surfaces at final test, once the cabinet is already fully built and rework gets expensive — exactly the kind of friction an export operation serving dozens of countries with different electrical codes can't afford.

• Every startup on a cooler or merchandiser line begins with a printed production order: model, destination voltage, lot size. Next to it sits the build sheet for that model, with its door, shelving, lighting and controller variants.
• Both sheets stay in the line lead's folder for the whole shift. Someone types them in at the end, from handwriting and memory, when the lot has already been built. The handwriting is the line lead's, but the typing is often done by someone who was not at the line when the lot started.
• A wrong voltage or a wrong variant typed at the end of the shift only surfaces at final test, with the cabinet foamed, the refrigeration circuit charged and the doors hung — the most expensive point to rework.
• For a plant exporting to dozens of countries with different electrical codes, a lot that does not exist in any system during its own build is a lot nobody can check against the order while it is still cheap to fix. Coolers for one bottler program can share a cabinet with coolers for another, with only the voltage and the controller setting telling them apart.

How it fits the IRIS system

Connect in photo mode — the line lead adds one photo to the startup they already do.

Connect photo mode: the line lead photographs the production order and the build sheet at startup, with a phone or an industrial tablet. A grounded language model extracts model, voltage, variant, quantity, line and shift, and stores it in iLEAN's central memory. Zero change to the line lead's routine.

It is the case with the least friction of the twelve because it asks the line for nothing new: the order and the build sheet stay on paper, exactly as they are. What changes is that the lot exists from the first cabinet, and every later check has something to compare against. On most plants it is also the first case approved, precisely because nobody on the floor has to be convinced.

See the full IRIS architecture →

Before and after

A cooler line startup on paper versus a captured startup

Aspect	Today, in the folder	With iLEAN Connect
When the lot exists in a system	End of shift, typed by hand	From startup
Destination voltage	Read again at final test	Extracted from the order at minute one
Door, shelf and controller variant	Checked when the unit is complete	Compared from the first cabinet
Wrong-variant rework	Cabinet already foamed and charged	Caught before assembly advances
Lot by model and line	Rebuilt from paper	One query
Line lead's startup routine	Order and build sheet on paper	Unchanged: one extra photo

End-of-shift transcription → lot visible from startup. Variant mismatch caught at final test → caught against the order from the first cabinet.

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 4-9 months, depending on the number of line startups per day.
• Less rework from mismatched variants: the mismatch is caught against the order from the first cabinet, not at final test. A cooler that would have been built to the wrong spec is corrected before it gets its foam and refrigerant charge.
• End-of-shift transcription disappears; the lot is visible to production, quality and planning from startup.
• And every later case in this matrix — final test, serialization, traceability — starts from a lot that already exists in the central memory. That is why it is usually deployed first.

Estimated payback of 4-9 months depending on daily startup volume, with less rework from mismatched variants. Estimate to validate.

And the fair question from the production manager

“What if it reads 115 V where the order says 220 V?” — extracting fields from a production order with a known layout is an anchored task, where the best models drop below 1.5% error [1]. And voltage is a closed list per destination: a value that does not match the order's country is flagged for the line lead to confirm before anything is stored.

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