Paperless batch startup

The problem

The formula that fixes oil, egg and vinegar spends the shift on paper.

the formula changes by product and by customer (own brand or private label), and its paper version can sit undigitized for hours. If an allergen claim comes in, someone has to reconstruct by hand exactly which formula was used that day — slow and error-prone across a catalog with dozens of active SKUs.

• Every startup at the mixing kettle begins with a printed batch order and a formula sheet: batch, weights per ingredient, declared allergens and packaging format, from a squeeze bottle to a foodservice tub.
• The formula changes by product and by customer: the same mayonnaise base runs under your own brand in the morning and as a private label with a different allergen panel in the afternoon.
• Those two sheets can sit undigitized for 4-8 hours, and on a busy shift the formula version actually used is whatever the operator had in hand.
• When an allergen claim comes in, somebody rebuilds by hand which formula ran that day, across dozens of active SKUs, with a binder and a phone call to whoever was on shift.

How it fits the IRIS system

Connect in photo mode — two photos at the kettle and the batch already exists.

Connect, photo mode. The operator photographs the batch order and the formula sheet with their phone at startup. A grounded language model extracts the structured fields — batch, formula, weights, allergens, packaging format — and stores them in iLEAN's central memory, ready to cross-check against the ERP or any later query.

Nothing about the startup changes except two photos. The operator still reads the formula, weighs the oil and the yolk, and signs the sheet; the difference is that the batch is queryable from second zero, with the exact formula version and allergens it ran with.

See the full IRIS architecture →

Before and after

Paper startup versus a startup that exists at second zero

Aspect	Today, on paper	With iLEAN Connect
Time until the batch is in the system	4-8 h, longer at weekends	Zero seconds
Formula version at the kettle	Whatever sheet was in hand	Extracted and stored with the batch
Allergen panel of a private label	Read off the paper from memory	Structured per batch and customer
Weights corrected in pen	Lost unless retyped	Captured as the real dosage
An allergen claim from a retailer	Binder, phone calls, half a day	One web lookup by batch
What the operator does differently on sauce and canning lines	—	Two photos

4-8 hours of latency before digitizing the paper → zero seconds. Manual reconstruction of which formula was used → instant web lookup by batch.

Impact estimate

Impact estimate — to be validated with your daily batch count.

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 batches you start per day.
• Transcription time recovered for the production manager, who today types the batch order and the formula into the system at the end of the shift.
• From 4-8 hours of latency to zero seconds: the batch exists before the first jar is filled.
• And the question of which formula ran that day goes from a manual reconstruction to an instant web lookup by batch.

estimated payback 4-9 months depending on daily batch volume, plus recovered transcription time for the production manager. *Estimate to validate.*

And the fair question from the production manager

"What if it misreads a weight on the formula sheet?" — extracting fields from a sheet whose layout you already use every day is an anchored task, and on anchored tasks the best models stay under 1.5% error [1]. On top of that, every capture comes back as a two-second summary on the line tablet, and the operator confirms or fixes it before it reaches the central system. A value read with low confidence is flagged, never guessed.

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