THE LINE SUPERVISOR DICTATES ON THE GO

The problem

The most valuable information of the shift never gets written down.

The line supervisor spends most of the shift walking between the cube press, the sachet filler and cartoning, coordinating several formats and brands alternating on the same line, hands busy adjusting machines. All the knowledge in his head — more rejects since the fat batch changed, a filler that will need backup — dies at the end of the shift.

• The line supervisor spends the shift walking between the cube press, the sachet filler and cartoning, coordinating formats and brands that alternate on the same line. Nothing they do happens sitting down.
• Their hands are busy adjusting machines, and the nearest keyboard is a walk away in an office they reach at the end of the shift, if at all. Anything that requires sitting down will not be written.
• More rejects since the fat lot changed, a filler that will need backup on Friday, a carton magazine that only jams on one format: all of it lives in their head and nowhere else.
• At handover it is compressed into five verbal minutes with the next supervisor already walking, and the rest dies with the shift.

How it fits the IRIS system

Connect in voice mode — a wake word, an earpiece and no walk back to the office.

Connect voice mode: with a Bluetooth earpiece and a custom wake word, the line supervisor dictates on the move. The LLM structures the dictation as an issue linked to the active batch and line, and returns support in seconds by voice or as a push to a tablet.

This case is not bought on savings, and presenting it that way would not survive the first question. It is bought because every continuous-improvement plan on the floor depends on a written history of what actually happens on the line, and today that history does not exist anywhere in the plant.

See the full IRIS architecture →

Before and after

The shift as it is lost versus the shift as it is captured

Aspect	Today	With iLEAN Connect
What the supervisor notices	Lost at handover	Written, dated and searchable
A jam that only one format causes	Solved again from scratch	Surfaces as a pattern with dates
Effort asked of the supervisor	Stop, walk, type	Speak while walking
An answer to a question about the batch	Later, from memory	By voice in seconds
A new supervisor's first month	Depends who trains them	Backed by the line's own history
Ordinary talk on the packing floor	—	Not recorded

Operational knowledge that dies at the end of the shift → enters the system in real time and gets cross-checked against live data.

Impact estimate

Impact estimate — to be validated with your continuous-improvement lead.

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.

• No standalone payback: this is an enabling piece. We call it strategic value and puts no months on it, so neither do we.
• What it unlocks is every improvement plan that needs a written history of the line instead of anecdotes recalled in a Monday meeting.
• The recurring problem stops being invisible: three occurrences across two months read as a pattern with dates attached, not as bad luck told twice.
• And the handover stops depending on what the outgoing supervisor happens to remember in the last five minutes of a nine-hour shift on their feet.

Strategic value, hard to monetize upfront: it enables every continuous-improvement plan on the floor. Estimate to be validated with the continuous-improvement lead.

And the fair question from the production manager

"Will it understand anything with the sachet filler running?" — dictation against a known vocabulary of equipment, formats and brands is an anchored task, and the best models stay under 1.5% error [1] on it even in plant noise, because the microphone sits on the earpiece and not in the room. The second net is that the structured note comes back to the supervisor before it is filed, by voice or on the tablet, so a misheard batch number is corrected in the same breath it was spoken. Nothing enters the system that the person who said it has not heard back.

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