Docs

Anomaly Detection

Fload automatically monitors your metrics for unusual spikes and drops. When something stands out, you get an alert — so you don't have to stare at dashboards waiting for bad news.


How it works

The anomaly detector runs on your historical data daily. It uses a statistical decomposition approach:

  1. Decompose the time series into trend, seasonal, and residual components.
  2. Model the expected range using the trend and seasonal components.
  3. Compare the actual value to the expected range.
  4. Flag values that fall outside the expected bounds with sufficient confidence.

The system compares each data point against multiple time horizons:

  • Yesterday vs. same day last week
  • Yesterday vs. same day last month
  • Yesterday vs. same day last year
  • 7-day rolling average
  • 30-day rolling average

Minimum requirement: 30 days of historical data for reliable detection. New apps may not have anomalies detected until enough history accumulates.


Anomaly types

TypeDescription
SurgeMetric significantly higher than expected
DeclineMetric significantly lower than expected

Severity levels

SeverityMeaning
LowMild deviation — worth watching
MediumNotable deviation — review soon
HighSignificant deviation — investigate today
CriticalExtreme deviation — immediate attention recommended

Which metrics are monitored

The anomaly detector watches all major metrics:

  • Revenue (proceeds, total revenue)
  • Downloads (total downloads, units)
  • Subscriptions (active subs, new trials)
  • Engagement (sessions, active devices)
  • Crashes
  • Ad metrics (spend, installs, CPA) when ad connectors are active

Viewing anomalies

Go to Monitoring in the sidebar to see all detected anomalies. You can filter by:

  • Status (new, viewed, acknowledged, dismissed)
  • Severity (low, medium, high, critical)
  • Type (surge, decline)
  • App (asset)
  • Metric
  • Date range

Anomaly statuses

StatusMeaning
NewJust detected, not yet reviewed
ViewedYou've opened it
AcknowledgedYou've seen it and noted it
DismissedYou've marked it as not worth acting on

Understanding an anomaly

Each anomaly card shows:

  • The metric and date
  • Actual value vs. expected value
  • Deviation percentage
  • Trend and seasonal components
  • Multi-scale comparisons (vs. last week, last month, etc.)
  • AI-generated explanation
  • Suggested actions

AI anomaly chat

You can ask the AI chat about anomalies directly: "Why did my revenue spike last Tuesday?" or "What caused the download drop this week?" The AI has access to your anomaly data and can correlate with other metrics to explain what happened.


Notification channels

Anomaly alerts can be sent to:

  • Email (configured in notification preferences)
  • Slack (when Slack integration is active)
  • Discord (when Discord integration is active)

Configure notification preferences in Settings → Notifications.