Shift-floor judgement stops dying at the end of the shift

The problem

In a business where the raw material changes every day, that judgement is the asset.

So the plant's most valuable knowledge is lost every eight hours. That the washer has been throwing more fines since the new supplier came in. That the pelletiser cutter needs a blade. That the new operator on extruder three is going to need backup. In a business where the raw material changes every single day, that judgement is the asset, and today it never reaches a system.

• The plant's most valuable knowledge is lost every eight hours.
• That the washer is producing more fines since the new supplier came in. That the pelletiser cutter needs a blade.
• That the new operator on extruder three is going to need support.
• And none of it reaches any system, because with hands busy and hearing protection on, writing is not an option.

How it fits the IRIS system

Voice mode, with its own wake word — and it answers back.

With Connect in voice mode, they dictate it while walking. A bluetooth earpiece with industrial noise cancellation and a dedicated wake word. Note that the washing line is throwing more fines since the new supplier's material came in, I want the lab to look at it. iLEAN listens, structures it as an issue and ties it to the active batch, line and shift. And it answers. Seconds later: the lab is free at quarter past twelve, shall I book it. That is humans in command applied to voice: the AI decides nothing, it records and it helps. What used to die at the end of the shift now reaches the handover intact and can be cross-checked against live process data.

He dictates that the washing line is producing more fines since the new supplier and that he wants the lab to look at it. Seconds later: "the lab can do 12:15, shall I book it". The AI does not decide: it prepares and returns.

See the full IRIS architecture →

Before and after

The shift's judgement, before and after

Aspect	Today	With iLEAN Connect
Recording channel	The closing meeting, if there is one	Continuous, during the round
What survives the shift	Whatever somebody remembers	The incident tied to batch, line and shift
Cost of recording an observation	Stop, remove protection, write	Dictate it while walking
Reply to what was dictated	None	Confirmation and booking, in the earpiece
Looking up a figure on the floor	Go back to the office	Ask and hear the answer
Shift handover	Whatever gets mentioned	What was recorded, with its context

Impact estimate

Impact estimate — an enabling piece, and it is worth saying so.

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.

• This case is hard to monetise directly, and the brief says so openly instead of inventing a figure.
• Its value is enabling: every continuous-improvement and predictive-maintenance plan depends on having these signals recorded.
• Without them, maintenance stays reactive and improvement runs on memory.
• It is defended as an enabler of the rest of the matrix, not with a number of its own.

Continuous-improvement enabler · shift-floor judgement stops dying at handover. This example is hard to monetise directly, and it is better said plainly. Its value is enabling: every continuous-improvement plan and every root-cause analysis depends on this information existing at all. In a plant that already works in OEE and total productive maintenance, this is the piece that feeds what the team already knows how to do. To validate with whoever owns continuous improvement. Conservative ranges, an estimate to validate against the plant's real data.

And the fair question from the production manager

"Our shift leads speak in heavy slang." — all the better. Transcribing speech over a closed, known vocabulary — lines, grades, equipment, symptoms — is an anchored task [1]. What gets structured are those entities, not free prose, and anything with low confidence is flagged for review.

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