APM with AI — the critical asset stops being measured at month-end and starts anticipating its own failure.

The problem

APM inherits an old problem: asset data lives on islands.

APM (Asset Performance Management) was meant to be the tool that measures and improves the performance of critical assets over time. In practice, in many plants it is a monthly slide deck with five KPIs that nobody uses to decide anything. The cause is not the tool — it is what feeds it:

1. SCADA and PLCs hold the most critical data (cycles, stoppages, temperatures) but without integration they stay "locked inside a machine that tells nobody about it".
2. The CMMS has the maintenance hours — but only the ones that made it into the system. Half of the asset's reality travels by messaging app and word of mouth.
3. The veteran knows the real cost of keeping that asset running, the recurring failure modes nobody documented, and the difference between the SCADA's "running" and actually running. That lives in their head.
4. The ERP has the cost of the spare parts consumed — but it is never cross-referenced with real availability to produce a cost per asset hour.

The result: APM draws an asset delivering 87% of plan on its headline KPI, while the person operating it knows it actually produced 30% less than it should have, burned an extra unplanned spare part, and ran "odd" for three shifts. The plant committee makes investment decisions with the first number; the second one is the real one.

How it fits the IRIS system

iLEAN cross-references the asset's reality: what the sensors see, what the operator says and what the legacy system hides.

A well-built APM with AI is not a prettier dashboard — it is an APM built on complete capture. That means iLEAN acting as the filler between SCADA, CMMS, ERP and the operator's voice, and agents cross-referencing all of it to produce asset metrics that management can defend.

Edge sees the early symptom. Connect joins SCADA, CMMS and voice. The agent produces the real cost per asset hour and proposes the intervention. The maintenance person signs off.

The iLEAN pieces applied to APM:

• Edge — a terminal with computer vision and sensor inputs next to the critical asset. It measures the reality of the cycle on the line (not just the SCADA's version when it reports), detects the early symptom (vibration, thermal halo, leak, noise) and cross-references it with known failure modes. When it triggers, it does so in milliseconds and with no network needed.
• Connect — the filler between SCADA, CMMS, ERP and the operator's voice. It reads from the SCADA even when it is old (directly, or by reading from the PC in the cabinet), from the CMMS even when it is poorly fed, and it captures the operator's voice note and the photo of the panel that used to travel by messaging app.
• Agents (in Central) — they produce the real asset metrics: true availability, MTBF / MTTR cross-referenced with the reports captured by voice, cost per asset hour (energy + spare parts + labor hours + real output), and intervention proposals. The "boring and concrete" part of the agent closes the loop (the asset file for the committee, the dossier for the audit).

See the full IRIS architecture →

Before and after

Classical APM vs. APM with iLEAN

Aspect	Dashboard-based APM	APM with iLEAN Edge + Connect + Agents
Source of the KPIs	Whatever SCADA and the CMMS could export	SCADA + CMMS + Edge + voice + ERP cross-referenced
Reported vs. real availability	Reported matches whatever the system saw	Real, verified on the line by Edge
Cost per asset hour	Estimate from reported spares and hours	Real, with output measured on the line + voice
Failure anticipation	Reactive (the failure already happened)	Early symptom + known failure mode = alert
Comparison between similar assets	Hard — each one reports its own way	Metrics normalized by the agent
Modernization/replacement decision	Gut-feel argument or age-based replacement	Real cost + trend → investment case

Impact estimate

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

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

• Medium or large industrial plant, 5-15 critical assets, APM in use but with low operational impact. Existing SCADA + CMMS + ERP; a parallel spreadsheet kept by the maintenance manager; the operator's voice outside the system.
• Pilot on 3 critical assets selected by Pareto: Connect against SCADA/CMMS/ERP plus voice capture, Edge on at least one of them, and an agent producing metrics and proposing interventions. First value expected within a few weeks.
• Indicative payback between 4 and 9 months. Hard lever: a single unplanned stoppage avoided on a bottleneck asset, or one modernization/replacement decision redirected by the real measured cost.
• Expected reduction in unplanned downtime hours on the pilot assets of the order of ≥30% after the first learning cycle.

And the operations director's fair question

"What if the agent's KPIs don't match the SCADA's?" — good question. The reality is that they often don't match because the SCADA only sees part of the picture. The agent does not invent: it cross-references several sources and, when there is a divergence, it says so with its traceability — this is the SCADA reading, this is the operator's voice note, this is the Edge reading, this is the difference. Hallucination is a problem of free generation, not of anchored tasks; on anchored tasks the best models brought the error below 1.5% ^[1]. The system proposes; the person signs off.

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

Related solutions: RCM with AI · AI-augmented CMMS · Digital operator rounds