Shopfloor TV visual management with AI — from the dead board to the board the operator looks at again.

The problem

The TV was bought, hung on the wall, wired to the BI — and nobody looks at it.

Three years ago the operations director decided the plant needed visual management "like everyone else has". Six TVs were bought, hung along the aisles and wired to the BI dashboard. For the first month everyone looked at them; by the third, nobody did. Why? Because the screen was showing last month's OEE and a chart the operator does not understand. The data did not belong to the shift. The message was not actionable. The TV became expensive decoration.

The problem was neither the TV nor the BI. It was that the data reaching the TV took a week to cook — ERP extraction, cleansing, aggregation in the data warehouse, dashboard refresh at 6:00 — and the afternoon shift was making decisions with a mirror of seven days ago. Visual management without reality capture is just a poster.

How it fits the IRIS system

iLEAN does not sell you another TV — it turns the ones you already have into a shift tool.

Live visual management is the visible face of the R in IRIS: bottom-up reality capture. If you capture the data in real time (Edge), if unstructured channels (email, WhatsApp, a photographed analogue panel) arrive at second zero (Connect), and if an agent writes the board in shopfloor language, the TV stops being decoration. iLEAN plugs into the TVs and the BI you already have; it is the filler that seals the crack between "nice data from yesterday" and "useful data right now".

Edge gives you the line's real pace right now. Connect captures what arrives by email or WhatsApp from suppliers and customers at second zero. The agent writes the board in shopfloor language and adapts it to whoever is looking. The TV stops being decoration.

The three iLEAN pieces applied to visual management:

• Edge — vision terminals on every line record pace, stops, defects and events without touching the PLC. What the TV shows is the current cycle, not the monthly average. It works without a network: even if the WiFi drops, the Edge keeps counting and the local board holds up.
• Connect — captures what comes in through human channels (a customer email pulling an order forward, a supplier's WhatsApp delaying material, a photo of the analogue panel on the old machine). The TV reflects those changes at second zero, not at tomorrow's meeting.
• Agent — writes the board the way a good shift leader would: "line 2 at 87% of takt, recovers in 14 min if the changeover lands on time", "SKU changeover on line 3 in 22 minutes, tool X ready". The shift leader's TV is not the line 4 operator's TV; the agent personalizes it. And when there is an andon, it needs nobody to press a button: it detects it from the line.

See the full IRIS architecture →

Before and after

Dead board vs. live iLEAN board

Aspect	TV with a classic BI dashboard	With iLEAN Edge + Connect + Agent
Data freshness	Last 24 h to a week	Second zero, the line's current cycle
Current shift context	Generic, no action	Upcoming stops, SKU changeover, alerts
Andon	Physical button nobody presses	Automatic detection from the Edge
Personalization by role	One screen for everyone	Different view for operator, shift leader, management
Capture from an old machine	Not integrated — never shows up	Phone photo of the panel → onto the board
Operator readership	Fades within weeks; they stop looking	They look again — because it helps them work

Impact estimate

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

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

• Plant with 6-12 TVs already hung on the walls (no screens to buy, the existing ones are reused), 3-6 active lines, BI already in place showing yesterday's data.
• Connecting the TVs to iLEAN's live data + an agent that writes the board per role. First live board within a few weeks.
• Indicative payback between 4 and 9 months. The hard lever is the reaction minutes gained — every stop detected at second zero instead of at the next meeting is worth money.
• Expected reduction of ≥ 30% in reaction time to shopfloor events, depending on your BI's current cycle and how often the shift actually reacts.

Why this works now

The reasonable doubt — "what if the agent gets the board wrong and misinforms the shift?" — has an answer. In tasks anchored to the source (reading a cycle, comparing it with the plan, writing a message), the agent does not opine: it recontextualizes data anchored to captured reality. A model asked to opine in the abstract fills the gaps; the same model anchored to verifiable evidence — a camera frame, an Edge record, a schedule entry — stops guessing and starts answering ^[1]. The agent proposes; the shift leader signs off any plan change that reaches the TV. The warm bed: the shift walks in and finds everything already written.

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

Keep pulling the thread: Andon · Digital gemba · All Lean methods