The plant manager's headset

The problem

The plant's best sensor walks the floor and never types.

critical operational knowledge — which line looks off, which new operator needs backup, which supplier changed quality — lives in the plant manager's head and never enters the system.

• The manager of a plant with mayonnaise, ketchup and dressing lines spends the shift walking, with hands busy and no keyboard within reach.
• They notice which line looks off, which emulsion is coming out thinner than usual, which new operator needs backup, which supplier's vinegar changed.
• None of it enters a system, because recording means stopping, walking to an office and typing, and the shift does not allow it.
• At shift end that knowledge is gone, and the next time the same viscosity drift shows up on the same line, it is solved again from zero.

How it fits the IRIS system

Connect in voice mode — a headset, a wake word and the manager in charge.

Connect voice mode with a Bluetooth headset and a custom wake word. A language model structures the dictation as an incident tied to the active batch, line and shift, and pushes confirmation or support back by voice or to a team lead's tablet.

The manager decides what gets recorded and when; nothing is captured otherwise. What they get in return is an incident structured by batch, line and shift, and an answer back in their ear or on a team lead's tablet within seconds. Within weeks the plant has something it never had: a dated history of the floor, written by the person who knows it best. Nobody else on the floor wears one or is heard through it.

See the full IRIS architecture →

Before and after

The walk-round today versus the walk-round dictated

Aspect	Today	With iLEAN Connect
What the manager notices	Gone by the end of the shift	Logged, dated and searchable
A thin emulsion on line 2	Mentioned at the next meeting	Tied to the batch and its formula
Backup for a new operator	Asked for in passing	Sent to the team lead's tablet
An ingredient behaving differently	A hunch	Recorded against the supplier lot
Effort to record it	Stop, walk, type	Say it while walking
Conversations on the sauce line	—	Not recorded

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

Impact estimate

Impact estimate — to be validated with whoever leads continuous improvement.

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. Its value is strategic and hard to price upfront, so we do not put months on it.
• It is the enabler behind every continuous-improvement effort in the plant, because those efforts need a written history of what happens on the lines.
• Recurring problems become visible: the third time the same line drifts in viscosity, it shows up as a pattern with dates, not as bad luck.
• And what the manager notices is cross-checked against live batch data instead of staying in one person's head.

strategic value, hard to price upfront, but the enabler behind every continuous-improvement effort in the plant. *Estimate to validate with the continuous-improvement lead.*

And the fair question from the production manager

"Will it understand me next to a running filler?" — dictation against a known vocabulary of lines, products and batches is an anchored task, and the best models stay under 1.5% error [1] on it, with the microphone at the manager's mouth rather than in the room. The structured incident is read back before it is filed, so a misheard line number is corrected on the spot. And only the manager is heard, and only after the wake word.

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