The Seveso III event happens at 04:12 — and the notification chain depends on somebody remembering who is on call.

The problem

The notification chain is perfect on paper — and depends on a human being awake at 4 in the morning.

In an upper-tier Seveso III plant, the Internal Emergency Plan defines the chain in detail: calling order, maximum response times, escalation, record keeping. In practice, the chain holds together like this:

1. The SCADA or the SIS detect the event and trigger a local alarm. The control room operator phones the on-call manager.
2. The on-call manager does not answer. It is night, they are asleep, the phone is on silent. The operator calls the second name. And the third. Each call costs a minute.
3. When somebody finally answers, the time, the event and who was notified are written down by hand. If it had to be escalated to the authorities (civil protection, the town council), the chain continues.
4. The next day, somebody reconstructs the log for the plant manager. And if the competent authority's inspection arrives, it has to be reconstructed again, with less detail.

The problem is not the plan: it is that executing the plan depends on memory and a diary. And the night the plan fails is the night the inspection discovers the log does not add up.

How it fits into the IRIS system

iLEAN does not write your Emergency Plan — it executes the one you already have, without anyone having to remember it.

The problem is not a lack of protocol: it is that the protocol lives in a PDF on a shared drive, and at 4 in the morning nobody opens it. iLEAN Connect acts as the putty between the event detected by the SCADA/SIS and the human chain, without asking you to change your control system or your emergency plan.

Connect fires the SMS, counts the time, escalates if nobody confirms, and records every signed hop. The plan runs by itself only in the notification part — the person decides in the part that matters.

The iLEAN pieces applied to Seveso III SMS escalation:

• Connect — the outbound voice over SMS. Dual gateway (different carriers) and fallback to an automated voice call. Loaded with the on-call calendar and the notification chain. It is bidirectional: the on-call manager's confirmation (replying OK, clicking the link, picking up the call) enters the system at second zero.
• Agents — the chain with a countdown. The agent classifies the event by Seveso category, locates the active on-call manager according to the calendar, sends the SMS, starts the configurable countdown and, if there is no confirmation, escalates to the next person. It carries the clock and the evidence while the person lives their life — and when the inspection arrives, the log is complete, signed and exportable.
• The three rings. SCADA/SIS events enter ring 3 through a passive mailbox: Connect never opens inbound connections into the OT network. The validation chamber (ring 2) verifies category and threshold. Ring 1 keeps the master record of the chain that was executed. The plant manager can review the log cold without exposing the OT network.

See the full IRIS architecture →

Before and after

Manual call chains vs. SMS with countdown and a signed log

Aspect	Manual call chain	With iLEAN Connect + countdown
Primary notification channel	Manual phone call	SMS with carrier delivery confirmation
Time to first contact	Minutes per call, in series	Second zero, in parallel if the chain requires it
Escalation if nobody answers	The operator remembers and calls the next name	Automatic, with a configurable countdown
Backup if the mobile network fails	«We'll try again»	Dual SMS gateway + automated voice call
Log for the inspection	Reconstructed by hand the next day	Every hop timestamped and signed, exportable
Who decides and signs	The plant manager the next morning	The person at every hop — Connect does not decide

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.

• Upper-tier Seveso III chemical plant, 3-5 on-call managers in rotation, proprietary SCADA/SIS, internal emergency plan approved by the authority. Pilot over the 4-6 most frequent event categories — the Pareto principle from the book applied to alerts.
• Dual SMS gateway + SCADA integration through a passive mailbox + on-call calendar loaded into Connect. First value expected within a few weeks: the first Seveso event with an automatic chain, countdown and signed log.
• Indicative payback between 4 and 9 months. Hard levers: control room operator hours given back (no more chaining phone calls), reduced regulatory risk at inspection (a defensible log is worth more than a reconstructed PDF), and reduced reaction time to a real event — the hard money is here, even if it is difficult to put in a table.
• Expected reduction in time-to-notify the on-call manager of ≥30% in the first quarter, conservatively — the ceiling is set by the on-call calendar and the mobile network.

And the plant manager's reasonable doubt

«What if the AI escalates wrongly and wakes up the person who wasn't on call?» — the escalation is not decided by the AI, it is decided by the on-call calendar loaded into Connect and by the notification chain in the Internal Emergency Plan. The AI executes what your plan says. Even so, the critical part is never decided alone: every SMS has a fast cancellation route for the plant manager, and SCADA false positives are filtered in the validation chamber (ring 2) before the SMS is fired. Reliability on anchored tasks (classifying an event by threshold, locating the contact on the calendar) is high — the best models brought the error below 1.5% on this type of task ^[1].

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