Predictive KPIs
Value metrics for predictive maintenance including anomaly-to-WO time, false positive rate, percentage of auto-created WOs, unplanned downtime, and RUL accuracy.
Predictive KPIs
Predictive KPIs quantify the real value that the predictive maintenance engine delivers. While operational dashboards show current status, predictive KPIs answer the most important business question: is the predictive system generating measurable ROI?
Each metric is designed to capture a different aspect of engine performance: detection speed, alert precision, automation level, impact on availability, and accuracy of remaining useful life predictions. The analysis period is configurable between 1 and 365 days, allowing comparison of short periods and annual trends.
These metrics are the bridge between the technical behavior of the algorithm and the operational objectives of the business.
What is it for?
- Demonstrates the ROI of predictive maintenance with objective data.
- Identifies whether the engine is generating too many false positives (alert fatigue).
- Measures the actual automation level of the WO generation process.
- Quantifies the reduction in unplanned downtime attributable to the system.
- Evaluates the accuracy of residual useful life (RUL) predictions against what actually occurred.
How does it work?
Available metrics
| KPI | Description | Unit |
|---|---|---|
| Anomaly-to-WO time | Average time from anomaly detection to WO creation | Hours |
| False positive rate | Percentage of predictive alerts that did not result in a real failure | % |
| % Auto-created WOs | Proportion of WOs automatically generated by the engine vs. manually created | % |
| Unplanned downtime | Hours of unscheduled stops in the analyzed period | Hours |
| RUL accuracy | Mean absolute error between predicted RUL and actual RUL at the time of failure | % |
Anomaly-to-WO time
Measures the system's reaction speed. A low value indicates that the anomaly → alert → WO pipeline is operating efficiently. It is calculated as the mean of all completed transitions in the period.
False positive rate
A high rate indicates the model is being too sensitive or that probability thresholds are too low. It is calculated by dividing alerts without confirmed failure by total alerts issued.
fp_rate = alerts_without_failure / total_alerts * 100Percentage of auto-created WOs
Measures the system's autonomy level. The higher the percentage, the lower the manual workload for the planning team.
RUL accuracy
Compares the residual useful life predicted at the time of intervention with the actual useful life observed. An error below 15% is considered high precision for most industrial assets.
Configurable period
All KPIs are calculated over a selectable period:
| Period | Typical use |
|---|---|
| 7 days | Weekly operational review |
| 30 days | Monthly performance review |
| 90 days | Quarterly evaluation |
| 365 days | Annual ROI report |
Using it from the Dashboard
- Go to Maintenance > Predictive KPIs.
- Select the analysis period in the date picker (1-365 days).
- The dashboard shows the five metrics with their current value and trend vs. the previous period.
- Click any KPI to see the detail and the list of events that compose it.
- Export the report in PDF or CSV from the Export button.
Predictive Configuration
Centralized configuration of predictive maintenance thresholds and weights per tenant. Adjust AHI, failure probability, maturity, and confidence ceiling without code changes.
ISO 13374 — Overview
Introduction to the ISO 13374 condition monitoring standard and how Rela AI implements its 6 levels.