Finding $12M of downtime before it happened.
An energy operator was running maintenance on a fixed calendar. We built a predictive layer over existing sensor infrastructure that identified $12M in avoidable downtime in the first operating year.
The situation on arrival.
Maintenance ran on manufacturer-recommended intervals regardless of actual asset condition. This meant simultaneously over-maintaining healthy equipment and under-maintaining assets under unusual stress.
The operator already had extensive sensor telemetry — it was being logged and largely ignored.
How we structured the work.
The constraint that shaped everything was that no new hardware could be installed within the engagement window. We worked entirely with existing telemetry, which forced a focus on feature engineering rather than data acquisition.
We prioritised the twelve highest-consequence asset classes rather than attempting full coverage.
They didn't have a data problem. They had a data-that-nobody-looked-at problem.
Systems now in production.
- A failure-prediction model across twelve asset classes using existing vibration, thermal, and pressure telemetry.
- A maintenance prioritisation queue integrated into the existing CMMS.
- An alert triage workflow so that engineers weren't flooded with low-confidence signals.
What changed for the client.
$12M in avoided downtime identified and validated by the operator's own finance function in year one. Unplanned outages fell 23%.
The programme is now extending to a further eight asset classes.
Considering a similar programme?
We take on a limited number of engagements each quarter. If the shape of this work is familiar, we should talk.
Start a conversation →