ADR LPG butane tanker: every stop, a signed discharge — product, tank temperature and customer serial.

The problem

Twelve stops, twelve ADR deliveries, twelve serials — and one paper sheet from the driver.

An LPG butane tanker leaves the distributor's depot with its ADR declaration, its written instructions, a delivery route with 8-15 residential or small B2B customers, and a tank whose pressure changes with the sun, with the journey and with every discharge. Each stop generates a different piece of paper in a different system:

1. The driver: a printed delivery note per stop, the meter reading, the customer's signature, sometimes a photo of the tank.
2. The distributor's ERP: waits for the route to close at the end of the day — between the stop and the close, anything can happen.
3. The truck's tank: has its pressure and temperature display, but it is rarely cross-checked against the discharge to a specific customer.
4. The customer tank serial: this is where most things get mixed up — customers with similar serials, one-off replacements, tanks that no longer belong to the distributor.

The delivery manager knows this, but the driver is alone in the cab and the back office receives the reports at the end of the day. The classic system works 99% of the time. That 1% is the discharge into the wrong serial, the tank pressure that goes out of range with nobody recording it, the customer who files a claim with no way to defend it. In LPG, that 1% is heavy: ADR, safety and reputation all at once.

How it fits the IRIS system

iLEAN Tracer does not add another system — it seals the cracks between truck, ERP and customer.

The problem in LPG home delivery is not a lack of information: it is information living on three islands (truck, ERP, customer) which, at the critical moment (the discharge at the next stop, the incident that surfaces at 22:00), does not reach the decision-maker in time. iLEAN Tracer acts as the filler for those gaps, without asking the distributor to change the ERP or renew the fleet.

Edge reads the ADR plate, the customer tank serial and the tank display. Connect captures the meter reading and the driver's messaging. The agent cross-checks product + temperature + serial at every stop. The person validates.

The iLEAN pieces applied to the ADR LPG butane tanker on home delivery:

• Edge — a terminal with computer vision (CNN) on the truck itself. It reads the truck's ADR plate, the customer tank serial at every stop and the tank pressure/temperature display. It works with no network: if the route is rural and coverage drops, the capture-and-sign cycle continues, and syncs when the signal returns.
• Connect — captures everything arriving through loose channels: the customer meter reading, photos of the tank and the valve, the driver's message reporting an incident, truck telemetry if the fleet emits it. Every data point enters tied to the stop and to the ADR batch at second zero.
• Agent — at every stop, it cross-checks the ADR product + the customer serial the ERP expects + the tank pressure + the kg discharged. If the serial is not the one the ERP expected, if pressure moves outside the operating range, or if the customer meter does not match what was loaded, it holds the signature on the discharge and alerts the driver with the concrete next action. The person validates on site or remotely; the discharge does not close by itself. The three safety rings exist precisely for this.

See the full IRIS architecture →

Before and after

Delivery with a paper sheet vs. delivery with iLEAN Tracer

Aspect	Printed delivery note + driver's report	With iLEAN Tracer
ADR product identity	On the truck's paperwork, not on each discharge	Tied to each stop and to the customer serial
Tank temperature/pressure	Visible on the display, never recorded	Captured by Edge and cross-checked with events
Customer tank serial	Checked by eye by the driver	Edge OCR + cross-checked against the ERP
Reconciling kg loaded vs. discharged	At the end of the day, at route close	Per stop, at second zero
Operation with no network	Manual, risk of losing the data	Edge keeps signing locally, syncs on return
File for an ADR auditor / customer claim	Rebuilt by hand, weeks	Dossier per stop, automatic, with photo, serial and human signature

Impact estimate

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

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

• LPG distributor with 10-20 tankers, mixed rural/urban home delivery plus small B2B, logistics ERP and a fleet with partial telemetry.
• Tracer pilot on one or two trucks (Edge for the ADR plate + customer serial + tank display, Connect for the meter reading + the driver's messages, an agent that cross-checks ADR + temperature + serial). First value expected within a few weeks: the first stops already arrive with a signed custody chain.
• Indicative payback between 4 and 9 months, depending on how often incidents occur (discharges into the wrong serial, kg mismatches, customer claims with no defence).
• Reduction of the time spent defending a claim or an ADR inspection of ≥30%, with a defensible floor far below the cost of a single incident or a single ADR penalty.
• The hard lever is a defensible operation: in front of the inspector, the insurer and the customer filing a claim, arriving with the photo, the serial, the reading and the human signature for each stop closes discussions that today drag on for weeks.

And the reasonable doubt of the safety manager

«What if the AI confuses a customer serial or misreads a pressure and we sign something that isn't true?» — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI merely re-contextualises a data point from one system to another (reading a plate, a serial, a display), the best models brought the error below 1.5% ^[1]. And even so, what is critical is never decided alone: iLEAN holds the discharge and the person signs. The three safety rings exist precisely for this.

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