Anomaly detection

Analysts could not monitor every client dashboard, so a drop in performance could go unseen until the client raised it. I built a real-time alerting system with the analytics team to monitor performance data and flag anomalies.

1. Each client's web analytics event data is compared with the same day the previous week.
2. The change is scored as a z-score.
3. A score beyond ±2, in either direction, sends an alert to the team's messaging channel.

Comparing against the same weekday means normal weekly patterns, such as quieter weekends, are not flagged as anomalies.

When signups declined on one account, the team received an alert the day after the decline began, before the client had noticed.

Next: Document expiry alerts Back to home