Your flat-pack kitchen is not judged in the factory — it is judged by the fitter in the customer's home.

The problem

The customer's site plan never reaches the production queue in time.

In a flat-pack kitchen furniture plant there are two truths that have to line up and almost never do:

1. What the site needs — the order in which the fitter needs the boxes to install the kitchen without stacking 22 cartons in the hallway and coming back three times. That lives in an email from the sales rep, a PDF from the installer, a WhatsApp from the site manager, a call from the end customer.
2. What the factory needs — the order that suits the ERP/MES: batch by edge-band color, group by hinge type, avoid running the edgebander twice for the same color. That lives on another island.

The planning manager tries to reconcile the two by hand every week, in a spreadsheet that ages badly. At the first change (which always comes, usually on a Friday afternoon), the plan stops being valid. Shipping improvises: the kits go out in the order that suits the factory, the fitter on site pays the difference, and so does the brand image. The problem is not a lack of information — it is information that reaches the decision-maker too late.

How it fits the IRIS system

iLEAN does not replace the planner — it puts an agent alongside that never sleeps.

The problem with JIS for flat-pack kitchens is not that a new system is missing: it is that the current system is split between what the factory knows and what the site knows, and reconciling them by hand is a full-time job for one person who can never keep up with the day-to-day. iLEAN acts as the putty between those two truths, and adds a coordinating agent that takes on the boring part.

Connect listens for the site plan whichever way it arrives. The agent cross-references it with the production queue. The person signs — the agent never decides alone.

The iLEAN pieces applied to JIS coordination for flat-pack kitchens:

• Connect — capturing the customer's site plan. It does not matter which channel it comes in on: an email from the sales rep, a PDF from the installer, a WhatsApp from the site manager, the retailer's customer portal. Connect captures it at second zero and turns it into a structured event. "The supplier writes at 8:00 that the truck is not leaving today, and by 8:01 the agents are already replanning."
• Coordinating agent — the brain that cross-references. It lives in Central and works 24/7. It matches the site plan against the ERP/MES queue, the real stock of components and line availability. It proposes the assembly and shipping sequence that minimizes stoppages in the factory while respecting the order the site needs. When the data conflicts, it does not act: it alerts the planning manager with the proposal on the table.
• The three rings — the agent proposes, the person decides and signs. The agent has no hands in SAP or in the MES; it sends a cryptographically signed proposal to the middle ring. The planning manager validates it in one click. That is what makes iLEAN acceptable in a serious operation: the decision that costs money is signed by a person.

See the full IRIS architecture →

Before and after

Manual planning vs. JIS coordinated by an iLEAN agent

Aspect	Weekly manual planning	With an iLEAN agent coordinating JIS
Capturing the site plan	By hand, via email/WhatsApp/PDF	Connect at second zero, nothing to forward
Reaction to a change on site	Meeting, phone call, manual replanning	Agent proposal in minutes, human validation
Shipping sequence	Convenient for the factory	Convenient for the customer's site
Order at the customer's site	Boxes stacked up, two-day installation	Boxes in sequence, installation in one morning
Who decides the final sequence	The person, alone, with partial data	The person, with the agent's proposal and complete data
Cost of last-minute changes	Paid by the factory and the fitter	Absorbed by the agent, with no manual recalculation

Impact estimate

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

The block below is an estimate to be validated with the specific data of your plant. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• Mid-sized plant making flat-pack kitchen furniture, several hundred kitchens a month, multi-SKU with many variants of edge-band color, hardware and handle.
• Connect + coordinating agent pilot: capture of the customer's site plan, an agent that matches it with the production queue and proposes the sequence. First value within a few weeks.
• Indicative payback between 4 and 9 months, depending on kitchens per month, the frequency of last-minute site changes and the current cost of manual rescheduling.
• Hard levers: ≥ 30% reduction in manual planning time, fewer sequence changes on the packing line, lower after-sales cost for installation at the customer's site.

And the planning manager's reasonable doubt

"What if the agent reorders the queue mid-production and makes a mess?" — it cannot, by design. The agent proposes, it does not act. It sends the sequence proposal to the middle ring packaged as signed JSON, and it is only applied once a person validates it. Whatever is critical to the operation is always signed by someone. And because the task is anchored (matching a plan against a queue), not free generation, the best models brought error below 1.5% ^[1]. The feeling in the committee is not "the machine decided on its own", it is "this week planning took one click instead of three meetings".

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