A photo of the work order at run startup

The problem

Every litho run starts on paper and stays on paper for hours.

The shift's work order is printed with ERP data and any last-minute adjustment (corrected Pantone, varnish change, quality note) gets handwritten. The lithography spec sheet is handwritten by habit of the technical office. Both can sit undigitized for 4-8 hours, piling up in a technical office folder until someone transcribes them — when the BRC auditor arrives, someone has to reconstruct which run each coil belongs to.

• The shift work order prints with ERP data, but any last-minute adjustment gets handwritten on it: a corrected Pantone, a change of interior varnish, a quality note carried over from the previous run.
• The lithography spec sheet is handwritten out of the technical office's habit: client, format, ink sequence, varnish, tinplate or TFS gauge.
• Both sit for 4 to 8 hours in a technical office folder until somebody transcribes them, and longer when the run started on the night shift.
• When the BRCGS Packaging auditor asks which run a coil fed, somebody has to rebuild it by hand from folders that were never meant to answer that question.

How it fits the IRIS system

Connect in photo mode — two photos at startup, and the run already exists.

Connect photo mode. The operator takes two phone photos at run startup. An anchored LLM extracts structured fields — client, format, Pantone, varnish, gauge, notes — and stores them in central memory. Zero change of habit for the operator.

The handwritten correction is the most valuable line on the sheet, because it records what was actually printed, not what was planned. Captured at the first sheet, it stops depending on somebody deciphering it a day later.

See the full IRIS architecture →

Before and after

The litho run on paper versus the litho run captured at startup

Aspect	Today, on paper	With iLEAN Connect
Time until the run exists in the system	4-8 h in a folder	Instant, from the first sheet
Pantone corrected by hand	Only on the paper copy	Extracted and tied to the run
Varnish and gauge per run	Typed when somebody has time	Structured fields, queryable by client and format
Coil-to-run link for the auditor	Rebuilt from scattered papers	One web lookup per run
Technical office transcription	Every work order, every shift	Review only, no typing
Operator's routine at the press	—	The same, plus two photos

4-8 h of digitization latency → instant. Auditor facing scattered papers → instant web lookup per run.

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 runs started per day.
• Technical office transcription time recovered, which today goes into copying work orders instead of preparing the next ones.
• Digitization latency drops from 4-8 hours to instant, including runs started on the night shift.
• And the auditor's question about which run used which coil becomes a lookup instead of an afternoon of folders.

Estimated payback of 4 to 9 months depending on run volume per day, recovering the technical office's transcription time. Estimate to validate.

And the fair question from the production manager

“What if it misreads a Pantone reference scribbled in the margin?” — reading fields from a work order and a spec sheet with a known layout is an anchored task, where the best models drop below 1.5% error [1]. And there is a second net: a low-confidence field is flagged, and the line lead confirms or corrects it before it crosses into central memory. Every correction is stored, so the fields that are hardest to read on your sheets improve run after run.

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