The knowledge that used to die at shift end

The problem

The person who knows most about the floor is the one who never gets asked.

The shop foreman at a drum factory spends most of the day walking between the winder, the closure press and the assembly area, hands busy adjusting paper tension, checking a closure, or moving a pallet. Information like "winder 2 has been rejecting more since the paper supplier changed" dies at shift end because no one can ask him anything — he can't type.

• The shop foreman spends most of the shift walking between the tube winder, the closure press and assembly, hands busy adjusting paper tension, checking a closure or moving a pallet.
• He sees things no system records: "winder 2 has been rejecting more since the paper supplier changed", an adhesive that sets slower on humid days, a hoop batch that seats badly. None of it fits in a form, and all of it explains rejects.
• He cannot type while he works, and nobody can ask him anything without stopping him. Walking back to an office terminal to record a hunch never happens on a busy shift.
• So that knowledge dies at shift end, and the same problem is rediscovered weeks later by someone else. The next foreman starts from zero, and so does the one after him.

How it fits the IRIS system

Connect in voice mode — the foreman dictates, iLEAN structures and answers in fiber drum making.

With a Bluetooth headset and Connect's voice mode, he dictates without stopping: "note that line 2 is rejecting more since the paper supplier changed". iLEAN listens, structures the dictation as an incident linked to the active batch and line, and replies back in seconds by voice or by a push to a tablet.

It only records what the foreman chooses to dictate. There is no continuous listening and no tracking of the team, which is what keeps a voice tool welcome on the floor months after it is installed. The foreman gains a memory he did not have; the team loses nothing.

See the full IRIS architecture →

Before and after

Today's floor knowledge versus knowledge dictated by voice

Aspect	Today	With iLEAN Connect
An issue spotted at the winder	Remembered, maybe	Logged against the batch and line
Paper supplier change and rejects	A hunch in one head	A pattern with dates
Asking the system a question	Walk to an office terminal	Ask through the headset
Shift handover	A few words in the corridor	A structured summary
Recurring closure or hoop problems	Rediscovered each time	Visible as recurring
Conversations on the drum floor	—	Never recorded

Before: critical operational knowledge lived in the foreman's head and never reached the system. After: it enters in real time and is cross-referenced with live data.

Impact estimate

No standalone payback — and we don't force one.

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 payback range of its own: it is strategic value, hard to monetize upfront, and we would rather say so than invent a figure.
• It is an enabling piece: it feeds all of the plant's continuous improvement work with what the foreman sees. The winder, the closure press and assembly all benefit from what he notices in passing.
• Critical operational knowledge moves from one person's head into the system in real time, cross-referenced with live batch data. A hunch about a paper supplier becomes a record that can be checked against reel batches and test results.
• The value is to be validated with the production lead, case by case, once the first patterns appear.

Strategic value, hard to monetize upfront — it enables all of the plant's continuous improvement work. Estimate to be validated with the production lead.

And the fair question from the production manager

“Will it understand him with the noise of the closure press?” — the headset is built for industrial noise and the vocabulary is configured with your terms: winder, hoop, liner, packing group. Turning a short dictation into an incident with line, batch and symptom is an anchored task, where the best models drop below 1.5% error [1], and the summary can be confirmed on the line tablet before it is stored. Nothing he says becomes a record without him seeing it first.

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