Anomaly Detection

Pro Plan

Anomaly detection automatically identifies unusual patterns in your analytics data. Get alerted to significant changes before they become problems.

What Is Anomaly Detection?

AI monitors your metrics for unexpected changes:

Monitors	Detects
Traffic	Spikes and drops
Conversions	Rate changes
Revenue	Unusual amounts
Errors	Sudden increases
Performance	Slowdowns

How It Works

Statistical Analysis

AI builds an understanding of "normal":

Your typical Tuesday:
- Traffic: 1,000-1,200 visitors
- Conversion: 2.8-3.5%
- Bounce rate: 42-48%

Today (Tuesday):
- Traffic: 580 visitors ← ANOMALY

Learning Period

When you start:

1. AI collects 2+ weeks of data
2. Builds baseline patterns
3. Learns day/time variations
4. Starts detecting anomalies

Continuous Improvement

Over time:

• More accurate detection
• Fewer false positives
• Better seasonality handling
• Adapts to your business

Types of Anomalies

Traffic Anomalies

Type	Example	Significance
Spike	+200% visitors	Campaign success? Bot attack?
Drop	-50% visitors	Server issue? SEO problem?
Pattern change	Weekday traffic on weekend	Audience shift?

Conversion Anomalies

Type	Example	Significance
Rate increase	+40% conversion rate	A/B test working?
Rate decrease	-30% conversion rate	UX issue? Bug?
Value change	2x average order	Big customer? Error?

Error Anomalies

Type	Example	Significance
Error spike	500% increase	Bug deployed?
New error	Never seen before	New feature breaking?
Error pattern	Errors at specific time	External dependency?

Performance Anomalies

Type	Example	Significance
Slowdown	Page load 3x longer	Server issue?
Region-specific	Slow for Europe only	CDN problem?
Page-specific	One page slow	Code issue?

Viewing Anomalies

Anomaly Feed

Access all detected anomalies:

1. Go to AI Insights → Anomalies
2. View chronological list
3. Filter by type and severity
4. Click for details

Anomaly Details

Each anomaly shows:

Element	Information
Metric	What changed
Magnitude	How much
Timing	When detected
Duration	How long
Comparison	vs. expected
Confidence	AI certainty

Sample Anomaly

🔴 CRITICAL ANOMALY

Metric: Conversion Rate
Change: -45% from expected
When: Today, 3:15 PM - ongoing
Duration: 2 hours 30 minutes
Expected: 3.2%
Actual: 1.8%
Confidence: High (97%)

Impact: ~$2,400 estimated lost revenue

Correlated events:
• New deployment at 3:10 PM
• No traffic change
• Mobile affected more

Actions:
→ Review checkout funnel
→ Check deployment logs
→ View affected sessions

Anomaly Settings

Sensitivity

Control detection threshold:

Level	Description	Use Case
Very High	10%+ changes	Every fluctuation
High	20%+ changes	Most changes
Medium	30%+ changes	Significant only
Low	50%+ changes	Major only

Per-Metric Settings

Customize sensitivity by metric:

Metric	Suggested Sensitivity
Revenue	Medium-High
Traffic	Medium
Conversions	High
Errors	Very High
Bounce rate	Low

Quiet Hours

Reduce noise during known variations:

• Maintenance windows
• Campaign launches
• Sale periods

Alert Configuration

Notification Channels

Choose how to be alerted:

Channel	Best For
Email	Non-urgent
SMS	Critical only
Slack	Team awareness
In-app	During work

Alert Rules

Configure when to alert:

If anomaly severity = Critical
AND metric = Conversion OR Revenue
Then: Email + Slack immediately

If anomaly severity = Warning
Then: In-app notification
Daily digest summary

Alert Frequency

Avoid alert fatigue:

• Deduplicate similar alerts
• Set minimum intervals
• Group related anomalies

Investigating Anomalies

Step 1: Verify

Confirm the anomaly is real:

• Check data in dashboard
• Look at related metrics
• Rule out data issues

Step 2: Understand Scope

Determine what's affected:

• Specific pages?
• Certain traffic sources?
• Mobile vs desktop?
• Geographic regions?

Step 3: Find Cause

Look for correlations:

• Recent changes?
• External events?
• System issues?
• Competitive activity?

Step 4: Act

Take appropriate action:

• Fix issues
• Monitor changes
• Update team
• Document findings

Step 5: Resolve

Close the anomaly:

• Acknowledge in system
• Note actions taken
• AI learns from resolution

Common Anomaly Causes

Traffic Drops

Cause	Investigation
Server issues	Check uptime
SEO changes	Search rankings
Ad stops	Campaign status
External events	News, weather

Traffic Spikes

Cause	Investigation
Viral content	Social mentions
Bot traffic	Traffic quality
Campaign launch	UTM parameters
Press coverage	Referrers

Conversion Drops

Cause	Investigation
UX bug	Session recordings
Payment issues	Payment logs
Pricing change	Recent updates
Competitors	Market research

False Positives

Why They Occur

False positives happen when:

• Normal variation triggers alert
• Seasonal patterns not learned
• One-time events
• Data quality issues

Reducing False Positives

• Let AI learn longer
• Adjust sensitivity
• Dismiss and provide feedback
• Set quiet hours appropriately

Feedback Loop

When you dismiss:

1. Click "Not relevant"
2. AI adjusts
3. Similar alerts reduced
4. Better future detection

Best Practices

Response Times

Severity	Response
Critical	Within 1 hour
Warning	Same day
Info	Weekly review

Documentation

Track anomalies:

• What happened
• Root cause
• Actions taken
• Outcome

Team Coordination

• Share critical anomalies
• Assign investigation
• Track resolution
• Learn from patterns

Custom Anomaly Rules

Enterprise Plan

Create Custom Rules

Define your own detection:

Rule: Low-conversion alert
Condition: Conversion rate < 2%
For: 2+ hours
Action: Alert via Slack

Advanced Conditions

Combine multiple factors:

• Traffic + conversion
• Error + performance
• Multiple metrics

Next Steps

• Natural language queries
• Weekly digest setup
• Acknowledging insights