Two taps, data verified

The problem

The real brake is the fear of AI writing into the ERP on its own.

The fear that AI will "enter data on its own and someone will have to clean up the ERP master" is the main cultural brake on shop-floor digitization — but the line supervisor doesn't want to type either.

• The most common objection to digitizing the bakery floor is not technical: it is the fear that AI will enter data on its own and someone will have to clean up the ERP master afterwards.
• On the other side, the line supervisor will not type either: there is no time between a pan loaf run and a bun changeover, and the terminal is at the other end of the hall.
• Caught between both, data either enters unverified and contaminates the master, or it does not enter at all because nobody trusts the capture. Either way, the digitization project stalls on the floor, not in IT.
• A wrong piece weight in the system is not a typo: it flows into yield, giveaway and the next batch's settings.

How it fits the IRIS system

Connect with early human verification — a tablet, a summary and two taps at the line.

Connect's early human-verification mode. Every capture — photo, voice, parsed email — reaches the tablet as a visual summary before crossing into the master. The supervisor confirms with a check (send to ERP) or corrects it first, for instance verifying the batch's target weight and format.

This is the piece that makes every other case approvable. The AI assembles the data; a person who looked at the line decides whether it enters. Without that gate, nobody on the floor signs off on anything else in this matrix, and with it the conversation moves from fear to speed.

See the full IRIS architecture →

Before and after

Unverified capture versus capture confirmed at the line

Aspect	Today	With iLEAN
Piece weight entering the ERP	Typed late or not at all	Confirmed in two taps
Format and loaf count per batch	Assumed from the order	Checked against the range on screen
A capture outside its range	Lands in the master	Stopped for correction first
Time asked of the supervisor	A walk to a terminal	A two-second summary
Trust in the item master	Low, so data is held back	Every record has a person behind it
Who writes to the ERP	A person, typing	A person, confirming

Impact estimate

Impact estimate — to be validated with your bakery's numbers.

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 the enabling piece of the whole matrix. Without it, none of the other eleven cases gets signed off on the floor.
• We do not put a month range on it because its value is strategic, not directly monetizable.
• The gain is a clean master: data captured and verified with seconds of latency instead of held back out of distrust, so yield and giveaway figures can be trusted again.
• And every record carries the name of the person who confirmed it, which is what quality and the auditor ask for when a weight complaint arrives from a retail customer.

Enabling piece for the whole iLEAN matrix — without it, none of the other pieces get signed off on the floor. Strategic value, not directly monetizable.

And the fair question from the production manager

“If a person confirms everything anyway, what is the AI saving?” — the work is not in deciding, it is in gathering. Today someone would have to walk to a terminal, find the batch and type weight and format. Here the summary arrives assembled and checked against its range: comparing a value with a known tolerance is an anchored task, where the best models drop below 1.5% error [1], and the person only confirms or corrects.

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