software

AI QMS post-market monitoring (EU AI Act Art. 72)

A post-market monitoring plan template for high-risk AI systems under EU AI Act Article 72 — what to monitor, how often, what to do with the data.

  • EU AI Act Art. 17
  • ISO 9001

Article 72 of the EU AI Act (Regulation (EU) 2024/1689) requires providers of high-risk AI systems to establish and maintain a post-market monitoring system proportionate to the AI technologies and the risks of the high-risk AI system. This article gives you a post-market monitoring plan template, the monitoring metrics that map to Article 72, and the link back to the QMS under Article 17.

What Article 72 actually requires

Three components:

  1. A post-market monitoring system that actively and systematically collects, documents, and analyses relevant data on the performance of high-risk AI systems throughout their lifetime, including post-launch data provided by deployers and other sources.
  2. A post-market monitoring plan part of the technical documentation referred to in Annex IV. The plan documents the methods, the data sources, and the analytical techniques.
  3. Continuous evaluation of compliance with the requirements of Chapter III, particularly Article 9 (risk management), Article 10 (data governance), Article 13 (transparency), Article 14 (human oversight), Article 15 (accuracy, robustness, cybersecurity).

The Article 72 obligations sit alongside Article 73 (serious incident reporting) and feed back into Article 9 risk management.

Plan structure

1. Cover

2. Scope

3. Monitoring objectives

For each of the Chapter III requirements, define the monitoring objective. Examples:

4. Metrics, sources and frequency

A practical metric matrix:

DomainMetricSourceFrequencyTrigger
PerformanceAccuracy on production sampleLogged predictions vs ground truthDailyDrop > 5pp from baseline
PerformanceCalibration errorLogged predictionsWeeklyDrift beyond control limits
RobustnessOut-of-distribution rateDetectorContinuousSpike > threshold
FairnessSubgroup accuracy gapLogged predictions, demographics where lawfully heldMonthlyGap > tolerance
SafetySerious incidentsIncident pipelineContinuousAny
SecurityAdversarial attack indicatorsSecurity stackContinuousSeverity ≥ medium
Human oversightOverride rateLogged operator overridesWeeklyGradual decline (over-reliance)
TransparencyDeployer feedbackChannels per Article 50QuarterlyPattern of misunderstanding
DataInput distribution driftStatistical testsDailyShift beyond control limits

Adapt to your domain. The metrics matrix is the executable part of the plan.

5. Data sources

6. Analytical techniques

7. Decision and action workflow

When a metric breaches a threshold:

  1. Triage. Classify the issue, performance, safety, fairness, security, transparency, oversight, data.
  2. Risk re-evaluation under Article 9. Update the risk register.
  3. Decision. No action / parameter change / model update / deployer notification / market action.
  4. Action implementation with traceability to the build pipeline under Article 17 element 3.
  5. Verification of effectiveness, repeat measurement.
  6. Update to technical documentation per Annex IV.

8. Reporting

9. Records

10. Review and update triggers

Connection back to Article 17

Article 17 element 9 (“setting up, implementation and maintenance of post-market monitoring”) is the Article-72 hook into the QMS. Element 10 (“procedures related to reporting of serious incidents”) covers the Article-73 leg. Together, they make post-market monitoring routine documented information rather than ad-hoc work.

Common pitfalls