Zero fruit loss, from scale to pallet

The problem

Fruit loss is rarely one failure: it is a warm hour, a missed defect and a late signature.

Fruit loss from out-of-range temperature or an undetected defect is the sub-sector's biggest economic risk, worsened because the harvest window concentrates most of the annual volume into a few months, leaving little room to react on the fly. An incident in peak season or a defective lot reaching export destroys margin and, worst case, an international distributor's trust.

• Loss from out-of-range temperature or an undetected defect is the sub-sector's biggest economic risk.
• It is made worse by the harvest window, which concentrates most of the year's volume into a few months and leaves almost no room to react on the fly.
• Each link looks acceptable alone: the truck waited a little, the cold room drifted a little, the sorter was set by hand, the defect was subtle.
• Together they produce a lot that ripens too fast or arrives defective at export — and one such lot can cost an international distributor's trust.

How it fits the IRIS system

Four cross-checking rings — the previous eleven cases, orchestrated.

The iLEAN flagship system coordinates 4 rings:

No ring solves the problem alone; each one exists today in some form and fruit is still lost. What gives away the chain of small failures is independent sources ceasing to agree — the lot's temperature, its grading, its images and its sign-off. When they all agree, the pallet ships with its dossier already closed.

• Connect captures orchard-to-lot traceability and harvest alerts in real time;
• the optical sorter applies the size/destination criteria published by the ERP with no manual entry;
• Edge cameras detect external defects in packing and validate critical changeovers, with JIDOKA AI blocking when something doesn't match;
• every lot leaves with an evidence dossier cross-referencing temperature, sorting, visual inspection and quality sign-off. SMED AI speeds up format changeovers so coordination doesn't hurt productivity in peak season.

See the full IRIS architecture →

Before and after

The four rings and what each one blocks

Ring	What it draws on	What it closes
1 · Orchard	Orchard receiving (1, 7) + grower alerts (3)	Orchard-to-lot traceability and a line plan that matches the fruit
2 · Grading and yield	Optical sorter (8) + lab (6) + press panel (2)	Destination from the ERP, dry matter and yield tied to the lot
3 · Packing and pulping	Packing defect (9) + guacamole changeover (10) + supervisor voice (4)	Defects and cross-recipe residue caught before the box
4 · Lot dossier	Evidence pack (11) + verification tablet (5)	A dossier per lot: temperature, sorting, images and sign-off
JIDOKA AI hold	All four rings	Holds the lot when two sources disagree
SMED AI changeovers	Size and recipe changes	Keeps coordination from costing throughput at harvest peak

Temperature or undetected-defect loss incidents, handled case by case and often too late → early detection and blocking before the fruit leaves the plant.

Impact estimate

Impact estimate — to be validated with your quality lead and general management.

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, counted against the first relevant loss avoided.
• Margin protection against perishable-product loss and export returns — a potential impact in the hundreds of thousands of dollars per season.
• Temperature or defect incidents stop being handled case by case, often too late, and are detected and blocked before the fruit leaves the plant.
• We validate these figures with your quality lead and general management, not against a generic benchmark.

Margin protection against perishable-product loss and export returns (potential impact in the hundreds of thousands of dollars per season). Estimated payback 6-12 months against the first relevant loss avoided. *Estimate to be validated with the quality lead and general management.*

And the fair question from the production manager

“Do we need all twelve cases before any of this pays?” — no. The rings are built in layers and most cases pay on their own; receiving and the sorter already close much of the origin and grading rings. On reliability, each cross-check compares anchored readings — a weight, a temperature, a grade — where the best models drop below 1.5% error [1], and JIDOKA AI holds a lot only on a verifiable mismatch.

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