Spaghetti diagram with AI vision — the operator's real route, not the one on the floor plan.

The problem

The spaghetti diagram your plant never manages to keep alive.

The spaghetti diagram is the most revealing Lean tool there is — and, paradoxically, the one almost no plant keeps up to date. The reason is always the same and nobody says it out loud: building it by hand costs too much to maintain.

1. One person, one shift, one workstation. The classical method is someone from the Lean team with a printed floor plan and a stopwatch following an operator for a whole working day, drawing a line every time they move. For a single workstation.
2. And it goes obsolete the moment anything changes. The mix changes, the SKU changes, you move a table: last month's drawing no longer represents the plant. And the plan stays pinned in the improvement room, fossilized.
3. It is the most revealing tool — when it actually gets done. The data that comes out is brutal: the operator walking 800 meters per shift because of a badly placed workstation, the carts crossing the same aisle three times, the tooling cabinet 35 meters away from the workstation that uses it most.
4. Meanwhile, your workforce keeps on walking. Motion waste is one of the seven (or eight) that Toyota taught — and it still goes unmeasured because measuring it by hand does not scale.

It is the syllogism of the two gaps applied to movement: the data exists (people move around the plant every minute), but it lives nowhere (there was no cheap sensor to record it); and even if we recorded positions, there was no agent to cross them with the production order and draw useful conclusions. AI has just delivered both pieces.

How it fits the IRIS system

The cameras already hanging in your shop floor are sensors you never used.

iLEAN's argument here is the filler one: you are not here to fill the plant with new sensors, you are here to get value out of the ones you have left unused for years. The security CCTV covers the flow and is wasted watching empty aisles. iLEAN Edge reads it with convolutional networks that detect people (not faces), reconstructs aggregated trajectories and crosses them with your MES and your ERP.

Edge sees the flow. Connect crosses it with the production order. The agent reconstructs the spaghetti by workstation and keeps it alive. The person decides what moves. iLEAN sees the flow, not the people.

The three iLEAN pieces applied to the spaghetti diagram:

• Edge — terminal with computer vision on existing CCTV cameras (or new ones if the plant has none). The CNNs detect bodies on the floor, not faces, with no facial recognition. The trajectory is reconstructed as aggregated data by zone and by shift. It works without a network: if the plant loses connectivity, Edge keeps aggregating locally and synchronizes when it comes back.
• Connect — captures which production order was running at each moment (MES), which SKU and which family (ERP), which workstations were active (the shift manager's spreadsheet, the cell's old panel). Without that layer, the spaghetti is a pretty drawing with no context.
• Agent — crosses the aggregated route with the production order and reconstructs the spaghetti by workstation, by family, by shift. It tells you which workstation concentrates 80% of the wasted motion and simulates which layout change would give back the most seconds per cycle. It proposes — the person decides and signs off the physical change.

See the full IRIS architecture →

Before and after

Spaghetti diagram with a stopwatch vs. spaghetti diagram with iLEAN

Aspect	Classical method (stopwatch + floor plan)	With iLEAN Edge + Connect + Agent
Time to build it	2-5 days per workstation, one family	Continuous from second zero, every zone
Shelf life of the data	Weeks, until the next mix change	Alive, updated daily by shift
Coverage	One family, one shift, one week	The whole plant, every shift, the whole mix
New sensors	None — just a stopwatch and a floor plan	Software only, on the existing CCTV
Privacy / GDPR	A person physically observing another person	Anonymous body detection, no faces, aggregated data
Link to the layout change	The Lean team proposes; the improvement room decides	The agent simulates and prioritizes; the person signs the change

Impact estimate

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

The block below is an estimate to be validated with the actual data of your plant. We lay it out so the committee has an order of magnitude; we refine it during the immersion.

• Mid-sized plant of 6,000-10,000 m², with 3-5 cells and CCTV already installed in aisles and workstation exits.
• Edge pilot in one zone (reading the existing CCTV + integration with the production order from the MES). First expected value within a few weeks: the first heat map of the real flow, already discussable with the Lean team.
• Travel reduction at the pilot workstation ≥ 30% after the first assisted layout-change cycle — hard lever: cycle seconds recovered × steps per shift × days per year.
• Indicative payback between 4 and 9 months, depending on the documented motion waste and the hourly cost of the workstation.

And the nuance that matters

A live spaghetti diagram is exactly the difference between a plant that can redesign its layout every quarter and one that redesigns nothing: the plant that knows continuously how its people move can act; the one that does not, cannot. The reliability nuance matters too — a general-purpose model asked an open question can make things up, whereas an agent anchored to your camera stream, your MES order and your ERP master data answers over verified data or says it does not know ^[1]. iLEAN sees the flow, not the people — and the counter-model (the lights-out shop floor where there is no flow because there are no people) does not build anything that lasts.

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

Keep pulling the thread: All Lean methods