Photo to the order at mold startup

The problem

Every mold change starts on paper and waits hours to exist.

In a plastic closures factory, every mold change starts with a paper production order and a handwritten startup sheet: mold, cavity, resin color, destination customer. Those two documents can sit for 4 to 8 hours in a folder before anyone transcribes them. If a bottler's auditor shows up, someone has to reconstruct what ran in which shift.

• Each mold change in a closure plant opens with a printed production order and a handwritten startup sheet: mold, cavity count, resin and masterbatch color, destination bottler, plus whatever the technician noted about blocked cavities or hot-runner zones.
• Both documents go into a folder by the press. Transcribing them is nobody's main job, so they wait 4 to 8 hours — longer when the change happened on the night shift or just before a weekend.
• During that window the batch is being molded, bagged and palletized, but in the central system it does not exist yet. Closures pile up in octabins with no record of which mold, cavity set and color made them.
• When a bottler's auditor or a complaint arrives, someone has to rebuild from paper what ran in which shift. It is slow, and a handwritten note such as "cavity 17 blocked" is exactly the detail that gets lost on the way.

How it fits the IRIS system

Connect in photo mode — two phone photos at the press, and the batch is born.

With iLEAN Connect, the technician takes two photos with his phone at startup. An anchored model extracts mold, cavity, resin color, customer and notes. The batch already exists in the central system the moment the photo is taken, with nobody typing anything.

The technician keeps filling in the same startup sheet he has always used; he only adds two photos. That is why this is usually the first case a closure plant approves: it asks the floor for nothing new and closes the gap that bottler auditors find first. Within days, the folder by the press stops being the only copy of what happened.

See the full IRIS architecture →

Before and after

Mold startup on paper versus mold startup photographed

Aspect	Today, on paper	With iLEAN Connect
Time until the batch exists	4-8 h, more on night changes	Zero: at the photo
Mold, cavities and color	Handwritten on the sheet	Extracted fields tied to the batch
Blocked cavities noted by hand	Lost unless retyped	Recorded against the mold
Destination bottler	On the paper order only	Linked to the batch from minute one
Rebuilding a shift for an auditor	Digging through folders	One query by batch or mold
Routine of the setup technician	—	Same sheet, plus two photos

4-8 hours of latency to digitize the order and sheet → zero latency, batch available the moment the photo is taken.

Impact estimate

Impact estimate — to be validated with your daily mold changes.

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 how many batch and mold changes your presses run per day.
• The production supervisor's transcription time is recovered: the hours spent retyping orders and startup sheets go back to the floor, where they are needed during changeovers.
• Capture latency drops from 4-8 hours to zero: the batch is queryable the moment the photo is taken, including night and weekend mold changes.
• A bottler's audit question about what ran in which shift, with which mold and which color, gets answered from the system instead of being reconstructed from a folder.

Payback 4-9 months · from 4-8h latency to zero latency. Payback of 4 to 9 months depending on daily batch-change volume, recovering the production supervisor's transcription time. Estimate to validate.

And the fair question from the production manager

“What if it misreads a cavity number or the masterbatch code on the startup sheet?” — reading fields from a document whose layout is known is an anchored task, where the best models drop below 1.5% error [1]. And there is a second net: the extracted summary goes to the tablet at the press and the technician confirms or fixes it with two taps before anything reaches the ERP. A low-confidence field is flagged, never guessed.

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