Digital gemba and genchi genbutsu — go to the real place without leaving the place in the dark.

The problem

What reaches the office no longer looks like what happens on the floor.

The operations director goes down to the gemba on the first Monday of every month. He walks, he listens, he sees, he goes back to the office. The rest of the month he manages from the reports the plant managers send up — reports that pass through two or three layers of middle management and get cooked, with no bad faith, so they line up with the plan, with the committee's expectations, with the tone of the last board meeting. By the time it reaches the office, the data is no longer the plant. It is the data the plant believes the director wants to hear.

And when there are several plants in several countries, a weekly physical gemba is impossible. The director ends up managing from aggregated KPIs that hide what matters. The veteran operator still spots deviations before they break something — but his intuition never reaches the committee. Reality and decision are disconnected, and that is precisely what genchi genbutsu exists to prevent.

How it fits the IRIS system

iLEAN scales the gemba — it does not replace it.

A digital gemba done properly is not just a prettier report. It is the materialisation of the first requirement of IRIS: bottom-up reality capture. The R in IRIS means exactly that — Industrial Reality. The system feeds on what actually happens on the floor in real time, not on figures somebody types in, and it lets the executive enter the real place from the office with the same cold eye they would bring in person. The iLEAN filler seals the gap between plant and office that no ERP ever managed to close.

Edge sees the plant as we would see it. Connect lets the operator speak directly to management. The agents cross-check and raise a flag when the ERP says one thing and the plant says another. The director goes down to the real gemba when it is needed — the digital one tells him when.

The three iLEAN pieces applied to digital genchi genbutsu:

• Edge — terminals with vision over critical lines. The camera sees what really happens, not what somebody wants to report. The director can enter the digital gemba and watch the real cycle right now, with the line's own eyes. It works offline: even if the WiFi drops, Edge keeps recording so it can push up to the office once the network returns.
• Connect — the ear and the voice of the plant. Each person carries a phone plus an earpiece (or a tablet where phones are banned), online, full duplex. The veteran operator can say "the dispenser's viscosity is drifting, I'm sending you a video" and that goes up to the system exactly as said. Connect also captures what comes in by customer email or supplier messaging. Nobody filters it.
• Agent — cross-checks what Edge says, what Connect says and what the ERP/MES says. When there is a discrepancy ("the ERP says the batch passed OK, the camera recorded three reworks, the operator said by voice that the part came out borderline") it raises its hand. The director no longer depends on the data surviving the chain of command intact — the agent looks after it.

See the full IRIS architecture →

Before and after

Classical gemba + report vs. physical gemba + iLEAN digital gemba

Aspect	Monthly physical gemba + report	Physical + digital with iLEAN
Frequency of access to the real place	1-2 times a month per plant	Daily, from the office — physical when it is needed
Filters between plant and management	2-3 layers of middle management	Zero — the data rises uncooked
Capture of the veteran's knowledge	Only inside their head	Voice and video captured as it is exercised
Detection of ERP vs. reality discrepancy	Later audit, if it ever comes	The agent raises it in the moment
Multi-plant scale	Limited by travel	As many plants as management needs
Risk of the report that lies	Structural, with no remedy	Broken by bottom-up capture

Impact estimate

Impact estimate for your plant — to be validated with your numbers.

This block is an estimate to be validated against the specific configuration of your group. Order of magnitude, not a commitment.

• Industrial group with 3-6 plants, general management currently doing a physical gemba 1-2 times a month per plant and managing the rest of the time from aggregated KPIs.
• Digital gemba pilot in 1 plant + 1 critical line: Edge over the line, Connect with the shift team, cross-check agent and a management view. First digital gemba running within a few weeks.
• Indicative payback between 4 and 9 months. The hard lever sits in the decisions taken late today because the data arrives cooked — quality deviations, unreported reworks, delays the ERP never picks up.
• Expected reduction of ≥ 30% in the time between something happening on the floor and management knowing about it, depending on your current reporting cycle and the length of the chain of command.

Why this works now

The legitimate question any serious executive asks — "what if the agent just tells me what I programmed it to tell me?" — is answered in the design: the agent does not generate freely, it recontextualises data anchored to captured reality. A model asked to opine in the abstract will fill the gaps; the same model anchored to verifiable evidence — a camera frame, a recorded voice message, an ERP record — stops guessing and starts answering ^[1]. The digital gemba is not a report touched up by AI; it is raw reality, ordered by an agent that pushes it up exactly as it is.

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

Keep pulling the thread: Hoshin Kanri · Multichannel incident capture · All Lean methods