Voice that keeps walking

The problem

The first symptom of a stoppage is heard on the floor and never written down.

Critical operational knowledge — especially early symptoms of mechanical or process failure — lives in the shift leader's head and never enters the system until it becomes a stoppage or a major issue.

• The shift leader spends hours walking between mixers, the divider, the tunnel oven and the slicer and bagger lines, hands full and never near a keyboard.
• They are the first to hear a bearing on the spiral cooler, see dough sticking on the molder or notice the oven belt tracking off center.
• None of that enters the system, because recording it means stopping, walking to the office and typing. It survives as a comment at shift handover, at best, and is gone by the third shift.
• So the symptom only becomes data when it turns into a stoppage, which is the most expensive moment to find out: the line is down, the spare is not in stock and the route trucks are waiting.

How it fits the IRIS system

Connect in headset mode with a custom wake-word — the leader talks, iLEAN structures.

Connect's voice mode with a custom wake-word. The language model structures the dictation as an issue linked to the active line and shift, with priority and destination role, and replies by voice or by push to a supervisor's tablet.

The early warning already exists: it is in the shift leader's ears and eyes. The headset only gives it a way into the system at the moment it happens, tied to the line and the shift, so maintenance sees the pattern before the line stops. It records what the leader chooses to say, nothing more.

See the full IRIS architecture →

Before and after

Today's floor knowledge versus floor knowledge captured by voice

Aspect	Today	With iLEAN Connect
Early failure symptom	Dies at shift end	Enters the system as it is said
Noise on the spiral cooler	Mentioned at handover, maybe	Logged against that asset and line
Who gets the warning	Whoever the leader bumps into	The destination role, with priority
Effort for the shift leader	Stop and type	Speak while walking
The line's history	In one person's memory	Cross-referenced automatically
Operators' conversation at the bakery	—	Never recorded

Impact estimate

Impact estimate — to be validated with your maintenance manager.

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 do not put a month range on it because its return shows up in other cases, not on its own.
• Its value is strategic: it is the precondition for catching failures early and reducing unplanned downtime, as the flagship case (12) shows.
• Knowledge that dies at shift end starts entering the system in real time and gets laid against each line's history, so a symptom heard on Monday and again on Thursday is recognized as the same one.
• The maintenance manager is the right person to size it, because the gain appears as stoppages that never happen and corrective repairs that become planned ones.

Enabling piece for early failure detection — strategic value, a precondition for reducing unplanned downtime (see case 12, flagship). Estimate to validate with the maintenance manager.

And the fair question from the production manager

“Is this a way to keep tabs on my team?” — no. iLEAN listens to the shift leader, not to the operators, and only after the wake-word they choose to say. There is no continuous recording and no location tracking. Structuring a dictated issue into line, asset and priority is an anchored task, where the best models drop below 1.5% error [1], and the leader hears back what was logged.

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