PMO: KPIs
Measures for a PMO judged on intervention rather than reporting, including forecast accuracy, warning lead time, and why on-time delivery alone is a misleading number.
Markdown. No sign-up, no email.
A reporting PMO is measured on whether the report went out. A predicting PMO is measured on whether anyone acted in time. These measure the second thing.
The six that matter#
| Measure | Definition | Target | How it gets gamed |
|---|---|---|---|
| Warning lead time | Days between a risk being flagged and it materialising | Above 10 | Flagging everything, so the warning carries no information |
| Forecast accuracy | Predicted completion against actual | Within 10% | Forecasting late and calling every early finish a win |
| Intervention rate | Share of flagged risks that were acted on | Above 70% | Counting "noted in the meeting" as an action |
| On-time delivery | Against the original baseline | Above 80% | Re-baselining, which makes every project on time forever |
| Budget variance | Actual against approved | Within 10% | Moving cost between projects at period end |
| Escaped commitments | Client promises missed without prior warning | Zero | Not recording commitments made verbally |
On-time delivery must be measured against the original baseline. Measured against the current plan it is meaningless: a project re-baselined four times is on time by construction. Track both if you like, but the original is the one that tells you whether the estimate was any good.
Warning lead time is the number that defines this function. A risk flagged the day it materialises is a report. The same risk flagged fifteen days out is a chance.
Two counter-metrics#
| Counter-metric | Catches |
|---|---|
| False-alarm rate on HIGH risks | A prediction engine tuned to flag everything, which is indistinguishable from no engine once people stop reading it |
| Rework as a share of delivered work | Projects hitting dates by shipping problems into the next phase |
Portfolio-level view#
| Measure | Why |
|---|---|
| Projects with a single-person critical path | The most common structural risk and the easiest to fix while it is early |
| Cross-project dependencies with no acknowledged owner | The dependency that fails is almost always the one the other team did not know about |
| Utilisation above 90% | Sustained, this is not efficiency; it is a system with no capacity to absorb anything |
What is deliberately not measured#
- Reports produced. The output of the old model.
- Dashboard freshness. Automatic now, and it was never the constraint.
- Meetings held. A cost, not a result.
- Tasks closed. Trivially inflated by splitting work into smaller tasks.
Cadence#
| Daily | Prediction refresh; anything newly HIGH |
| Weekly | Interventions taken and their effect, commitments at risk |
| Monthly | Forecast accuracy against what actually happened, false-alarm rate |
| Quarterly | Whether the thresholds are still right |
The monthly forecast-accuracy check is the one that keeps the rest honest. A prediction engine nobody scores becomes a confident engine, and a confident wrong engine is worse than a spreadsheet, because people stop bringing their own judgement to it.