From a paper blending order to a live batch at second zero

The problem

The batch lives on a clipboard for eight hours.

Between the order being printed and the log being closed lie the eight hours of the shift. During that time nobody outside the line knows how many drums have been filled or whether average weight is drifting off target. Planning works from yesterday's picture and quality cannot anticipate anything. When the log reaches the office it is either typed in a hurry or typed wrong. And there is a silent loss nobody accounts for: the margin notes — "source tank changed", "second batch came out denser" — are exactly the ones that explain a deviation, and they never get transcribed. The cost is not just the clerk's time. It is that the plant loses the ability to explain itself: when someone asks weeks later why that batch came out the way it did, the answer is on a filed sheet nobody is going to look for.

• The blending order comes down from technical office with product, formula, source tank and target batch, and the operator hangs it on the clipboard at the filling station. From that moment until the shift closes, nobody outside the line can see it.
• The packaging log is filled in by hand as drums, jugs and IBCs come off the scale: container count, weight per container, start and end time. Average weight drifting off target shows up when the log is typed, not while it is happening.
• The margin notes are the most valuable part and the first to be lost: "source tank changed to T-4", "second blend came out denser". They explain the deviation, and they never reach the system.
• When the log gets to the office it is typed in a hurry, and often typed wrong. A month later a customer complaint lands and the answer is on a filed sheet nobody will find in time.

How it fits the IRIS system

Connect in photo mode — the operator keeps the clipboard, and adds two photos.

Connect in "photo of the analog" mode.

Nothing changes on the line: the order is still printed, the log is still filled in by hand. What changes is that the batch exists in central memory from the first minute, with its handwritten notes searchable, instead of from the morning after. And because the two photos are taken at start-up and at close, the plant sees the batch opened and closed on the same screen, with container count and average weight against target while there is still time to correct the fill. That is why it is usually the first case a plant approves.

• The operator photographs the blending order at start-up and the packaging log at close, with a phone or the line tablet.
• A model grounded in the document layout extracts the structured fields: product, formula, source tank, batch, pack size, container count, weight per container and handwritten notes.
• The data is stored in central memory and linked to the batch.
• What is captured goes through early human verification (see case 5) before crossing into the management systems.

Zero habit change: the operator learns no new system and types nothing.

See the full IRIS architecture →

Before and after

Today's blending order versus the captured blending order

Aspect	Today, on paper	With iLEAN Connect
Latency line → system	4-8 h, and a full weekend	Seconds
Container count and weights	Typed the next morning	Live, per batch
Margin notes	Never transcribed	Extracted and searchable by batch
Batch progress	A phone call to the line	On screen
Transcription per shift	One log, one clerk	Zero
Operator's habit	—	Unchanged: two photos

from 4-8 hours of latency between line and system to seconds · from lost handwritten notes to notes searchable by batch · from one log per shift to be transcribed to zero transcription · from batch progress known by calling the line to progress available on screen.

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 3-8 months, depending on the number of blending batches per day.
• Transcription time of the plant clerk and the shift supervisor recovered, batch after batch, which in a plant running six to ten blending orders a day is the better part of one person's morning.
• From 4-8 hours of latency between line and system down to seconds, and a weekend batch exists on Saturday, not on Monday.
• Planning and quality stop working from yesterday's picture: average weight drifting off target is visible while the batch is still open.

recovery of the transcription time of the plant clerk and the supervisor, plus visibility of the running batch. Estimated payback of 3 to 8 months depending on blending batches per day. *Estimate to validate.*

And the fair question from the production manager

“What if it misreads a weight or a batch number on the log?” — reading fields from a document with a known layout is an anchored task, where the best models drop below 1.5% error [1]. And there is a second net: every batch goes through early human verification on the line tablet (case 5) before it crosses into the management systems, so a low-confidence value is confirmed or corrected by a person who has the log in front of them.

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