Zero variance across plant, distribution centers and route

The problem

What takes the most money is not the filler's speed.

Glass bottles and plastic crates are working capital with a real replacement value, traveling to the customer and sometimes not coming back. Nobody can explain that variance, because by the time it is detected months have passed and there is no one left to ask. Across dozens of nodes, every blind spot is multiplied by the number of sites. The work order defines batch, format and expiry, and the coder prints exactly that, with no intermediate typing. What gets palletised is identified from second zero. It is the foundation the other three rings rest on: if the origin is ambiguous, nothing downstream reconciles.

• It is the gap between what leaves the plant, what the system says and what is physically in each depot and each truck.
• Product that runs out in one depot while it sits spare in the next. Expiry that shows up in the count.
• Glass bottles and plastic crates are the company's working capital, with a real replacement value, that travel to the customer and sometimes do not come back.
• Nobody can explain that variance, because by the time it is detected months have passed and there is nobody left to ask.
• Across dozens of nodes, every blind spot multiplies by the number of depots.

How it fits the IRIS system

Four rings — no new pieces, it is all of the above orchestrated.

Ring 2 — On the dock, JIDOKA AI will not release a mismatched load. During loading, vision counts cases of product and empty containers by format and cross-checks them against the load sheet and the ERP order. If they disagree, the dispatch is not released: the tablet shows both figures and the photo, and the supervisor decides. Trucks stop leaving with a variance nobody has seen. Ring 3 — On the return, the difference has an owner. On the way back, the same vision counts what comes off: unsold product and recovered returnable containers. It cross-checks against the route's sales, and the container difference is assigned to route, customer and day instead of piling up until month end. The dock supervisor adds context by voice without stopping work. Ring 4 — Agents sees what no single site can see alone. Every day, Agents reconciles the four rings across all sites and detects patterns invisible from any single node: a route that systematically loses containers, a format that always expires at the same depots, one site overstocked while another runs dry. And it proposes the concrete action: a transfer, forced rotation, or a visit to the customer holding containers. SMED AI speeds up plant format changes so this discipline never costs availability.

It is the only case in the matrix whose return is not argued with an estimate: it is measured with your own ratio of returnable packaging that does not come back, which you already know and which today nobody can explain.

See the full IRIS architecture →

Before and after

The four rings, and what each one closes

Ring	Which cases it draws on	What it closes
1 · The order is the truth	Shift start (1) + coder (8)	Batch, format and expiry, with no keystroke in between
2 · The dock does not release a mismatch	Vision (9) + dock validation (5)	A dispatch that does not match what was loaded
3 · The network finds out at once	Depot alerts (3) + dock voice (4)	Duplicated decisions and blind spots between depots
4 · The return settles with evidence	Validation (5) + evidence pack (11)	Packaging that leaves and nobody can explain

Impact estimate

Estimated impact — to validate with your own 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.

• Estimated payback 6-12 months.
• Recovering a fraction of the returnable packaging lost today is, across dozens of depots, the largest return of the whole system.
• Reduced write-off from expiry, which today shows up in the count with no assignable cause.
• And variance goes from being detected months later to being assigned the same day.

Recovering a share of the returnable containers lost today and cutting expiry write-offs is, across dozens of distribution centers, the biggest return in the entire system. And it is the only one argued not with a productivity estimate but with a line in the profit and loss account. Estimated payback between six and twelve months. Estimate to be validated against your real container shrinkage and inventory accuracy figures.

And the fair question from the production manager

«Do we have to deploy all twelve cases to get this?» — no, and framing it that way would be the way never to start. The rings are built in layers and each case pays on its own from the first month. With shift start-up and the coder you already have all of ring 1, which is what orders everything else.

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