The 8 lean wastes with AI vision — the map no manager could ever build by hand.

The problem

The 8-wastes map ages faster than it can be built.

The continuous improvement manager knows that Ohno's 7 wastes — plus the 8th, unused talent, added later — are bleeding the plant. But quantifying them by hand is a logistical nightmare: timing the waits station by station, drawing a spaghetti diagram of motion, interviewing operators to identify over-processing, counting inventory between stations at a given hour, checking the ERP for overproduction, opening the defect reports, cross-referencing the skills matrix for talent. Three weeks of a technician's time — and next month, all over again.

So the 8-wastes map usually ends the same way: it gets built once a year in a kaizen event, it gets pinned to a wall, and the plant keeps running. The manager's eye catches the big wastes; the small, constant ones — the 90-second wait on a SKU changeover, the extra motion the operator makes because the tool sits two steps away, the over-processing of a redundant visual inspection — are the ones holding the OEE ceiling in place.

How it fits the IRIS system

iLEAN turns the 8-wastes map into a natural by-product of having vision on the line.

If the camera is always there and sees the line the way we would see it, the large and the small wastes surface on their own — the wait between operations, the inventory piling up, the operator walking to search for something, the defect slipping through. Add Connect cross-checking against the customer plan, and an agent cross-checking against the skills matrix, and the 8 wastes are covered without hiring three full-time technicians. It is the filler that closes the gap between aggregate OEE and station-by-station reality.

Edge sees waits, motion, inventory and defects continuously. Connect cross-checks against the plan and captures the operator's voice. The agent quantifies the 8 wastes per shift and per SKU. The manager decides what to attack first.

How the three iLEAN pieces cover the 8 wastes:

• Overproduction — Edge counts pieces + Connect reads plan/customer + the Agent compares: if more is produced than ordered, it is flagged with a timestamp.
• Waiting — Edge measures the time the station spends without useful activity (operator with no part, machine with no material). Per station, shift and SKU.
• Transport — vision over transit zones plus cross-checking with the production plan identifies material movements that add no value.
• Over-processing — Edge compares the operation actually performed against the standard work: extra operations are flags.
• Inventory — vision quantifies the pieces in the buffer zone of each station. History vs. optimal level.
• Motion — vision of the operator at the station detects extra walking, bending, searching. A digital spaghetti diagram without painting the floor.
• Defects — Edge with CNN detects the defect in line, before it moves on (powder coating case, two-stage jidoka).
• Unused talent — Connect captures the operator's voice and initiative, Agents cross-check against the skills matrix. Latent capability nobody has activated gets identified.

See the full IRIS architecture →

Before and after

Manual 8-wastes map vs. the live iLEAN map

Aspect	Stopwatch + interview + spaghetti	With iLEAN Edge + Connect + Agent
Time to build the map	2-4 weeks of a technician	A few weeks of initial setup, then continuous
Update frequency	Annual (kaizen) or quarterly	Per shift, per SKU
Coverage of small wastes	Limited to the manager's eye	Continuous — the camera counts them all
Detection of unused talent	Subjective, depends on the direct supervisor	Objective pattern cross-checking Connect and the skills matrix
Prioritizing what to attack	“Wherever we see the most blood”	By quantified impact on takt and cost
Use of the kaizen event	Diagnosis eats the time	The diagnosis arrives done — the kaizen is decision

Impact estimate

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

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

• Pilot line with 6-10 stations, OEE stable at 65-78%, continuous improvement active but with no traction to climb.
• Edge over critical stations + intermediate inventory zones + Connect against ERP/MES. First quantified 8-wastes map in a few weeks.
• Indicative payback between 4 and 9 months. The hard lever is eliminating 2-3 large wastes that are invisible to OEE today but weigh on cost per unit.
• Expected reduction in total waste ≥ 30% over the waste detected today, depending on the real starting point.

The underlying data point

The plant that detects its wastes continuously eliminates them; the plant that only has a monthly OEE figure does not. The reasonable doubt — “what if the agent miscounts a waste?” — has an answer: on tasks anchored to the source, the best models brought the error rate below 1.5% ^[1]. The agent proposes a quantification; the continuous improvement manager verifies it and decides. Assist and simplify — not replace.

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

Related: Muda/mura/muri detection · Spaghetti diagram with vision