Shift knowledge stops dying at the end of the shift

The problem

The freshest knowledge in the plant lives in three or four heads.

The freshest operational knowledge in the plant — the knowledge that would make real continuous improvement possible — lives in the heads of three or four people who walk all day and never enters any system. That line three has been rejecting more crimps since the reel change. That the cartoner is starting to misfold the long leaflet. That the new operator will need backup at tomorrow's changeover. What reaches the shift report is a summary written in a rush at the end, already filtered and stripped of context. Not because nobody wants to record it, but because recording it would mean stopping to type.

• That line three has been throwing more crimp rejects since the reel change. That the cartoner struggles folding the long leaflet. That the new operator will need support at tomorrow's shade changeover.
• None of it reaches any system, because recording it would mean stopping to type and the packaging lead walks the whole shift with their hands full.
• What makes it into the shift report is a summary written in a hurry at the end, already filtered and stripped of context.
• It is not unwillingness to record: it is that the available tool asks for the one thing that person cannot do.

How it fits the IRIS system

Connect in voice mode, with its own activation word.

Connect in voice mode with a bluetooth earpiece and a dedicated wake word, so that ordinary shop-floor conversation is never recorded. The lead dictates without breaking stride. The LLM structures the dictation as an issue tied to the batch, line, shade and shift active at that moment, stores it, and decides whether somebody needs to be notified. The response comes back through the earpiece or as a push to the tablet. Humans in command: iLEAN decides nothing for him, it gives him support.

The activation word is not a technical detail: it is what guarantees that ordinary floor conversation is never recorded. iLEAN listens when spoken to, not all the time, and that distinction is what makes a team adopt it instead of working around it.

See the full IRIS architecture →

Before and after

Today's shift versus the captured shift

Aspect	Today	With iLEAN Connect
Where the observation ends up	In the head of whoever saw it	An issue tied to line, shade and shift
When it is recorded	At the end of the shift, from memory	The moment it is observed
Effort	Stop and type	Speak while walking
Shift report	Written on getting back to the office	Already written by then
Basis for improvement	Impressions	Cross-referenced data
Ordinary floor conversation	—	Not recorded: an activation word is required

Impact estimate

Estimated impact — to validate 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.

• This case carries no payback, and none is invented for it: it is the hardest to monetize directly, and forcing a figure would undermine the credibility of the rest.
• Its value shows up in the ratio of improvement actions that get closed with data rather than with impressions.
• Operational knowledge that stops dying at the end of the shift and becomes cross-referenceable with process data.
• A shift report already written by the time the lead reaches the office, instead of reconstructed from memory.

this is the hardest example to monetise directly and we do not assign it an invented payback. Its value shows up in the share of continuous-improvement actions that actually close on data instead of on impressions. *Estimate to validate with the continuous-improvement lead*.

And the fair question from the production manager

«Is this here to check up on my team?» — no, and framed that way the case would die within a month. An activation word is required for iLEAN to listen: ordinary floor conversation is not recorded, and there is no tracking of where anyone is. The gain runs the other way: the packaging lead no longer has to remember at three in the afternoon what they saw at six in the morning.

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