Anomaly Detection Alerts

A short watchlist of metrics worth interrupting people for, alerts that arrive with context and an owner, seasonal baselines that cut false alarms, and a labelling and tuning rhythm that keeps alerts trusted.

Free · Mac (Apple silicon) and Windows · v1.0.5 · Other platforms

Then open the Marketplace in the app, search for Anomaly Detection Alerts, and click Install. It goes into your own files, and works offline from then on.

What it installs.

1

Pillar

1

Area

50

Projects

205

Starter tasks

📊 Data & Analytics

Anomaly Detection Alerts

Automatically spotting unusual movements in metrics such as sales, sign-ups or errors and alerting the right people with useful context.

  1. Metric watchlist for anomaly alerts
  2. Alert to action map for each watched metric
  3. Labelled history of past incidents and anomalies
  4. Detection grain and evaluation schedule per metric
  5. Annotation feed for launches, promotions and outages
  6. Choosing an anomaly detection approach or tool
  7. Routing map from metric to owning team
  8. Business impact severity levels for metric alerts
  9. Alert message that carries its own context
  10. Late data guard before anomaly checks run
  11. Metric alert triage rota and weekly handover
  12. Alert outcome labelling log
  13. Monthly detector precision and recall review
  14. Quarterly sensitivity retune of watched metrics
  15. Weekly digest of low-severity anomalies
  16. Watchlist upkeep as metrics change
  17. Snooze and mute register with expiry dates
  18. Missed anomaly review after each incident
  19. Detector heartbeat and silent failure watch
  20. Annual review of the metric alerting programme
  21. Static thresholds and percent change rules
  22. Seasonal baselines for weekly and daily patterns
  23. Outlier-resistant statistics for alert bands
  24. Prediction interval bands as alert limits
  25. Change point detection for level shifts
  26. Alerting on rates with small denominators
  27. Contribution analysis to explain a metric drop
  28. False alarm rates across many metrics
  29. Backtest of candidate detectors on labelled history
  30. Grouping correlated alerts into one incident
  31. Root cause hints attached to every alert
  32. Replacing hard-coded thresholds on the top ten metrics
  33. Choosing alert channels by severity
  34. Minimum volume rules for sparse segments
  35. Self-service alert subscriptions for business teams
  36. Shadow period before detector go-live
  37. Peak trading season alert plan
  38. Launch watch window for a new product or market
  39. Rebaselining after a tracking or definition change
  40. Synthetic anomaly fire drill
  41. Anomaly alerts for a data team of one
  42. Inheriting alerts nobody trusts
  43. Onboarding a teammate to the alert rota
  44. Metric alerts for a business team without analysts
  45. Linking error rate spikes to business metric drops
  46. Anomaly detection for low-volume start-up metrics
  47. Streaming anomaly detection on live events
  48. Multivariate detection across related metrics
  49. Hierarchical detection across thousands of segments
  50. Linking anomalies to releases and feature flags

Where it starts.

Each project arrives with its purpose, milestones and first tasks written out. Here is how the first three read.

Metric watchlist for anomaly alerts

This is a team plan, best installed into a shared workspace so everyone with a role in it can see and own their tasks. Teams that switch detection on for every metric in the warehouse get hundreds of alerts in week one and ignore all of them by week three.

First tasks

  • List the metrics on the five most viewed business dashboards
  • Pull the metrics mentioned in the last twelve months of incident reports
  • Write the action a sudden change would prompt beside each candidate
  • Cut the list to thirty and send it to each metric owner for sign-off

Alert to action map for each watched metric

An alert that only says sales are down leaves the reader to work out whether to check the payment provider, the website or the ad account. Writing, for each watched metric, the first three checks and who can act on them turns an alert from a worry into a short procedure. It is also the fastest way to find metrics that have no real response and should not interrupt anyone.

First tasks

  • Draft the response card for the highest-revenue metric first
  • Ask the owner of each metric what they check first when it moves
  • Link each card to the dashboard and logs the checks need
  • Ask the agent to turn rough owner notes into consistent card wording
  • Refresh the cards for metrics that alerted this month @recurring(monthly:15)

Labelled history of past incidents and anomalies

Without a list of real past problems you cannot tell whether a detector is any good, only whether it is noisy. Reconstructing a year or two of outages, tracking breaks, pricing errors and promotion spikes, with start and end times, gives a labelled set to test every method against. Most teams find twenty to sixty usable events once they search chat history, incident tickets and finance corrections.

First tasks

  • Search incident tickets and chat channels for the past two years of outages
  • Ask finance for any revenue corrections or refund spikes they remember
  • Check each event window against the metric chart and fix the timestamps
  • Load the finished table into the warehouse with a clear name

Make it yours.

Install Anomaly Detection Alerts, then tell Aurum the date you're aiming for. It checks whether the plan fits the weeks you actually have.

Free · Mac (Apple silicon) and Windows · v1.0.5 · Other platforms

Blueprints are general information to help you organise your own work. They are not medical, financial, legal or other professional advice. See the Terms.