Site or carrier alert in seconds

The problem

What is really arriving travels through a channel nobody watches.

Critical alerts about what's really arriving (not what the delivery note says) come through informal channels — email or WhatsApp — to specific people who can't watch the inbox in real time. By the time they're read, the truck is already queued at the scale and the redirect decision comes too late.

• The delivery note says clean concrete. The contractor's WhatsApp to the plant manager says the load now carries gypsum board and wood from an interior strip-out.
• Carriers warn of delays, double loads or a changed source site by email, to whichever personal address they happen to have, competing with everything else in the inbox. Nobody designed that channel; it simply became the one that works.
• Those people cannot watch their phone in real time, so the alert is read two to three hours later — with the truck already queued at the scale, or already tipped.
• Once a mixed load has been processed together with clean material, the redirect decision has no value left: the batch is already compromised. All that is left is to explain it at the next inspection.

How it fits the IRIS system

Connect in chaotic-sources mode — a silent copy on the operational channels, around the clock.

Connect's chaotic-sources mode: a dedicated mailbox or WhatsApp Business number stays silently copied in 24/7. An LLM extracts intent (waste composition change, transport delay), entities (site, carrier, date) and cross-checks the active intake schedule. It returns the manager a concrete action, not just the alert.

Reading the alert sooner is not the value; cross-checking it is. Knowing that a site's load changed matters only if you also know which truck it is, when it arrives and whether your authorization accepts the new mix. That is what comes back, to the plant manager and to the weighbridge, before the truck turns into the gate.

See the full IRIS architecture →

Before and after

The contractor's warning, unread versus cross-checked

Aspect	Today, on someone's phone	With iLEAN Connect
When the alert is read	When its recipient checks the phone	Seconds after it is sent
Change in waste composition	Discovered at tipping	Matched to the scheduled truck
Decision to redirect	After the bad mix is processed	Before the truck reaches the scale
Load outside the authorization	Accepted by mistake	Flagged for refusal
Carrier delay	A hole in the intake plan	Schedule re-sequenced
Final call	Whoever happened to read it	The plant manager, with the action ready

2-3 hours of human latency → seconds. Redirect decision made before the truck arrives, not after a bad mix is processed.

Impact estimate

Impact estimate — to be validated with your alert history.

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 3-7 months, depending on how often contractors and carriers send alerts and how many of them change what the truck carries.
• Unauthorized waste stops being processed by mistake, with everything that means for the environmental authorization.
• From 2-3 hours of human latency down to seconds, nights and weekends included.
• And the decision to redirect or refuse a load is made before the truck arrives, not after a bad mix has reached the crusher.

Avoids incorrect processing of unauthorized waste, estimated payback 3-7 months depending on alert frequency. *Estimate to validate*.

And the fair question from the production manager

“Are you going to read my staff's WhatsApp?” — no. iLEAN sits on a dedicated mailbox and a WhatsApp Business number that contractors and carriers are given for operational alerts; personal chats are never seen. Classifying the intent of an alert and extracting site, carrier and date is an anchored task where the best models drop below 1.5% error [1], and the decision to redirect is always taken by a person. iLEAN proposes the action; the plant manager decides it.

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