← All products— Product / 02Prediction Engine

PredictIQ™

Failure prediction, risk scoring, and remaining useful life — trained on your equipment, not a generic library.

PredictIQ ingests vibration, thermal, oil-analysis, process, and work-order data, then trains failure models per asset class against your actual failure history. It outputs a predicted failure date, a confidence interval, a recommended action, and the financial consequence of doing nothing — pushed directly into SAP PM as a work-order recommendation.

Talk to us →See all products

Who it is for

  • Reliability engineers running condition-monitoring programs that produce alerts nobody acts on
  • Maintenance leaders trying to shift from time-based PMs to condition-based and predictive
  • Operations leaders carrying unplanned-downtime risk on critical, single-train equipment

The problem we are solving

Off-the-shelf 'AI for maintenance' tools are trained on someone else's data, fire too many false positives, and give engineers an alert with no recommended action and no business case. They get ignored within a quarter.

How it works

Train per asset class, on your failures

Models are built per equipment class (centrifugal pump, induction motor, reciprocating compressor, gearbox, conveyor, crusher, HVAC) using your historian tags, your CBM data, and your historical work-order failure codes. No generic library transplants.

Predict, score, and price the risk

Each asset receives a health score, a predicted failure window, and a financial impact estimate built from your production rate, downtime cost, and spares lead time. The engineer sees 'Pump P-204: 73% probability of seal failure in 14–21 days, $480k production exposure'.

Push the action into the work order

Recommendations flow back into SAP PM / Maximo as draft notifications with criticality, recommended craft, parts, and estimated duration. The planner approves, schedules, and closes the loop — feedback then retrains the model.

Measurable benefits

Ranges below reflect the low and high end of outcomes observed across pilots and production deployments. Your numbers will depend on your starting baseline.

25–45%
Unplanned downtime reduction
on instrumented critical assets
<8%
False-positive rate
after 90-day tuning window
$1.2–4.8M / yr
Maintenance cost avoidance
per mid-sized plant, typical pilot result

Capabilities

  • Per-asset-class failure models with explainable feature attribution
  • Remaining useful life (RUL) with confidence intervals
  • Anomaly detection on process tags with seasonality awareness
  • Risk-prioritized work-order recommendations with parts and craft pre-filled
  • Feedback loop: closed work orders retrain the model monthly
  • Engineer override and exception capture for model governance

Integrates with

We connect to the systems you already run. No rip-and-replace.

PI System / AVEVA PI, OSIsoft, Aspen IP.21SAP PM / S/4HANA, IBM MaximoVibration platforms: SKF, Emerson AMS, Bently NevadaThermal, oil analysis, ultrasonic inspection appsSnowflake, Databricks, Azure ML, AWS SageMaker

Proof point

An iron-ore mining operator caught three impending gearbox failures on primary conveyors in the first 90 days of a PredictIQ pilot — avoiding an estimated $7.4M in lost production and emergency repair cost.

Frequently asked

The questions buyers in your seat actually ask before committing to PredictIQ.

How is this different from the predictive maintenance module our CMMS vendor already sells?
CMMS-bundled modules typically run a generic anomaly model on a handful of tags. PredictIQ trains per asset class on your failure history, fuses CBM + process + work-order data, and returns a recommended action with a financial impact — not just an alert. The difference shows up in false-positive rate and engineer adoption.
We don't have years of clean failure data. Can you still build a useful model?
Yes. We start with physics-informed and transfer-learned models that perform without dense failure history, then improve as your closed work orders feed the loop. Most asset classes reach a usable accuracy threshold inside 60–90 days.
How do you keep engineers from ignoring alerts after the novelty wears off?
Every recommendation includes a confidence band, the top contributing signals (explainable, not a black box), an estimated production exposure in dollars, and a one-click feedback path. Alerts that get dismissed retrain the model — the system gets quieter and more trusted over time, not louder.
Do recommendations actually create work orders in SAP or Maximo, or just sit on a dashboard?
They write back as draft notifications/work requests with criticality, recommended craft, parts, and estimated duration pre-filled. A planner reviews and releases — the loop closes inside the system your team already uses.
Who owns the models at the end of the engagement?
You do. Code, trained models, feature definitions, and runbooks are handed over and live in your tenant. We can continue to operate them as a managed service if you prefer, but there is no lock-in.

Stop reacting to failures. Predict them.

We will run a 90-day pilot on one critical asset class with a fixed fee and a measured ROI. If the numbers don't land, you don't scale.

Contact us →
← All products↑ HomeNext: OpsMath →