Prediction-to-Execution Cycle Time (PECT)

GMI-AR-013Asset ReliabilityPilot-Ready

Definition

Prediction-to-Execution Cycle Time (PECT) measures the total elapsed time from when an AI system generates a valid maintenance recommendation to when the corresponding work order reaches "work completed" status: decomposed into its constituent organizational phases.

The Measurement Gap

You spent $2 million on a predictive maintenance platform. The vendor's pilot was impressive: the model detected a bearing degradation pattern on a critical gearbox 47 days before failure. Your reliability engineer validated the alert. It was real. The model was right.

The bearing failed 39 days later. Unplanned. The gearbox took the coupling with it. Three shifts of lost production. Emergency parts airfreighted. Weekend overtime for the crew.

What happened? The prediction was accurate. The model performed exactly as advertised. But the organization couldn't move fast enough. The alert sat in a queue for six days before a planner picked it up. Planning took another eight days because the planner was buried in reactive work. The bearing wasn't stocked: procurement lead time was 21 days. By the time the parts arrived and the scheduler found a window, you were at day 41. The asset failed on day 39.

Nobody tracks this. Nobody measures the elapsed time between "the AI told us" and "we finished the work." The predictive platform has its own accuracy dashboard showing a beautiful 87% detection rate. The CMMS shows schedule compliance at 91%. Both systems report success. But the space between them (the organizational response chain) is invisible.

MTBF

doesn't see this. MTBF measures the interval between failures. It can tell you the bearing failed sooner than average, but it can't tell you that the failure was predicted and the prediction was wasted.

Schedule Compliance

doesn't see this either. Schedule Compliance measures whether planned work was executed on time, but the work was never planned in time to BE on the schedule.

PM Completion Rate

, Backlog Age,

Planned Maintenance Percentage: none of them capture the gap between a valid prediction and a completed response.

This gap gets worse as predictive capabilities improve. A model that predicts failures 90 days out is useless to an organization that takes 120 days to respond. A model that predicts 30 days out is valuable to an organization that responds in 14. The prediction lead time is a technology characteristic. The response time is an organizational characteristic. No existing metric measures whether they match.

Here's the measurement gap, stated precisely: there is no metric that tells you the elapsed time between a valid AI-generated maintenance recommendation and the organization's ability to execute that recommendation, which means prediction accuracy is wasted by organizational response latency.

The predictive maintenance industry talks constantly about model accuracy. Almost nobody talks about whether the organization can act within the window the model provides. You can have a perfect model and a broken response chain, and every legacy metric will look fine right up until the failure happens.

Formula

Prediction-to-Execution Cycle Time (PECT) measures the total elapsed time from when an AI system generates a valid maintenance recommendation to when the corresponding work order reaches "work completed" status: decomposed into its constituent organizational phases.

The decomposition is what makes this metric useful. A single total number tells you "we're slow." The phase breakdown tells you WHERE you're slow and, by extension, what to fix.

The formula in plain language:

Start the clock when the AI system generates an alert or recommendation. Track when a planner picks up the resulting work order. Track when the job plan is approved. Track when parts are available. Track when the work is scheduled. Stop the clock when the work is completed. Report each segment separately and as a total.

Notation:

PECT = T_complete − T_alert

Decomposed:
PECT_triage = T_plan_start − T_alert
PECT_planning = T_plan_complete − T_plan_start
PECT_procure = T_parts_available − T_plan_complete
PECT_schedule = T_scheduled_date − T_parts_available
PECT_execute = T_complete − T_scheduled_date

The derivative metric (the one that really matters) is the

Prediction Utilization Ratio (PUR):

PUR = Prediction_Lead_Time / PECT

Where Prediction_Lead_Time = T_predicted_failure − T_alert

PUR tells you whether you're winning or losing the race between prediction and response. PUR ≥ 1.0 means you finished the work before the predicted failure date. PUR < 1.0 means the prediction window closed before you could act.

The variables:

  • T_alert = timestamp when the AI system generated the recommendation (source: predictive platform event log)
  • T_plan_start = timestamp when a planner first claims the resulting work order (source: WO status log)
  • T_plan_complete = timestamp when the job plan is marked planned/approved (source: WO status log)
  • T_parts_available = timestamp when all required materials are staged or confirmed available (source: inventory/procurement system)
  • T_scheduled_date = timestamp of the scheduled execution date (source: WO schedule fields)
  • T_complete = timestamp when the work order status changes to complete (source: WO status log)
  • T_predicted_failure = the AI system's estimated failure date at time of alert (source: predictive platform output)
  • Prediction_Lead_Time = T_predicted_failure − T_alert: how much runway the prediction gave the organization

Worked example:

A vibration analytics platform detects an abnormal pattern on a cooling tower fan motor (Asset CT-FAN-0217) on March 3 and estimates failure within 52 days (predicted failure date: April 24).

  • T_alert = March 3
  • Planner picks up the work order March 11: PECT_triage = 8 days
  • Job plan approved March 18: PECT_planning = 7 days
  • Bearings and seals are not stocked. PO placed March 19. Parts arrive April 3: PECT_procure = 16 days
  • Scheduled for execution during planned outage April 9: PECT_schedule = 6 days
  • Work completed April 11: PECT_execute = 2 days
  • PECT = 39 days
  • PUR = 52 / 39 = 1.33

This organization won the race: barely. They finished 13 days before predicted failure. But look at the decomposition: 16 of the 39 days were spent waiting for parts. Procurement consumed 41% of the response window. If the prediction lead time had been 35 days instead of 52, they would have failed.

Now consider the same scenario with a 28-day prediction lead time. Same organizational response speed: PECT = 39 days. PUR = 28 / 39 = 0.72. The model was right. The bearing would have failed. And the organization would have watched it happen because their response chain is 11 days too slow. PUR < 1.0 converts a predicted failure into an unplanned failure.

Edge cases:

If parts aren't required, set PECT_procure to zero. If the work order is cancelled or superseded, exclude from PECT calculation but track separately: cancelled predictions are a signal about either model quality or organizational decision-making. If multiple alerts are generated for the same asset and condition, use the first alert timestamp to avoid understating response time. If the predictive system doesn't produce a failure date estimate, PECT is still computable but PUR is not: report PECT alone.

Proposed formula, pending validation pilots.

At a Glance

Category
Operational Response and Work Execution
← Prediction Threshold Value Ratio (PTVR)Cohort Homogeneity Score (CHS) →

Assess Your Organization

The AI-GPS Assessment evaluates your AI governance maturity and recommends KPIs like this one for your implementation roadmap.

Take the Free AI-GPS Assessment