How to Find the Cause of a Metric Change

**Quick Answer:** Find the cause of a metric change by decomposing it across dimensions (time, segment, channel), isolating the driving factors, correlating with potential causes, and validating with additional data. This process typically takes 2-4 hours manually or under 5 minutes with AI tools like OLARI.

The 5-Step Framework for Root Cause Analysis

Step 1: Confirm the Change Is Real

Before investigating, rule out false alarms:

**Data quality issues** — Check for missing data, duplicate records, or tracking errors

**Seasonality** — Compare to same period last year, not just last month

**Statistical significance** — Is the change large enough to matter?

**Example:** A 5% revenue drop might be noise; a 15% drop warrants investigation.

Step 2: Decompose by Dimensions

Break down the metric by every available dimension:

**Time** — When exactly did it start?

**Customer segment** — Enterprise vs SMB vs Consumer

**Geography** — By region, country, or market

**Channel** — Organic vs paid, source/medium

**Product** — By feature, plan, or SKU

**Device** — Desktop vs mobile vs app

**Goal:** Find which segment(s) account for the change.

Example:

Total revenue: -15%

Enterprise: flat

SMB: -8%

Consumer: -42% ← Found it

Step 3: Isolate the Driver

Once you've found the segment, dig deeper:

What changed within that segment?

Is it volume (fewer customers) or value (lower spend)?

When exactly did it start?

Example:

Consumer revenue: -42%

Consumer volume: -38%

Consumer ARPU: -6%

Timeline: Started Tuesday 10am

Step 4: Correlate with Causes

Match the timing and segment with potential causes:

Internal causes:

Product changes or bugs

Pricing updates

Marketing campaign changes

Technical incidents

External causes:

Competitor actions

Market events

Seasonality

Economic factors

Example:

Consumer mobile signup dropped 38% starting Tuesday 10am

iOS app update deployed Tuesday 9am

Correlation: App update broke signup flow on iOS

Step 5: Validate the Hypothesis

Confirm your theory with additional data:

Check error logs for the suspected cause

Interview affected customers

Look for similar patterns in related metrics

Test the fix and monitor recovery

Real-World Example: Conversion Rate Drop

**The Problem:** Conversion rate dropped 20% week-over-week.

Step 1: Confirm

Data quality: ✓ Clean

Seasonality: Not a holiday week

Significance: 20% is material

Step 2: Decompose

By device: Desktop -5%, Mobile -35%

By channel: Organic -8%, Paid -28%

By geography: US -18%, EU -22%, APAC -24%

**Finding:** Mobile + Paid traffic is the primary driver.

Step 3: Isolate

Mobile paid conversion: -35%

Mobile paid traffic: +12%

Timeline: Started Wednesday

**Finding:** Traffic volume up but quality down on mobile paid.

Step 4: Correlate

Wednesday: New ad creative launched

Ad targeting unchanged

Landing page unchanged

**Finding:** New ad creative attracting wrong audience.

Step 5: Validate

Ad A/B test data: New creative has 40% lower intent signals

Solution: Revert to previous creative

How OLARI Automates This Process

The manual process above takes 2-4 hours. OLARI does it in minutes:

1. **Ask:** "Why did conversion drop last week?"

2. **Get:** Complete decomposition, isolation, and correlation

3. **Act:** Specific recommendations based on the analysis

Example OLARI Response:

"Conversion dropped 20% due to mobile paid traffic. The new ad creative launched Wednesday is attracting lower-intent users—mobile paid bounce rate is up 45%. Recommendation: Revert ad creative or adjust targeting to match previous audience quality."

Key Takeaways

1. Always confirm the change is real before investigating

2. Decompose systematically across all dimensions

3. Isolate until you find the specific driver

4. Correlate timing with potential causes

5. Validate before acting

[Start finding metric causes in minutes →](/pricing)

How to Find the Cause of a Metric Change: A Step-by-Step Framework