Understanding Attribution Models in Stellar

Last Updated: 2025-05-30

Stellar uses different attribution models for different purposes to provide both statistically accurate A/B test results and comprehensive user journey insights. This guide explains why we use different models and how to interpret the data from each.

1. What Are Attribution Models?

Attribution models determine how credit for conversions is assigned when users interact with your experiments across multiple sessions, devices, or with different characteristics (like UTM parameters). The key question is: when a user converts, which of their previous interactions should get credit?

2. Session-Based Attribution (Experiment Statistics)

When viewing experiment statistics and applying segment filters, Stellar uses session-based attribution. This means conversions are only counted if they occur in a session that matches your filter criteria.

How It Works:

• A conversion is attributed to a segment only if the conversion happened in a session where that segment's criteria were met.
• Each conversion is counted exactly once, ensuring statistical validity.
• This prevents double-counting and maintains the integrity of A/B test results.

Example:

If you filter by utm_campaign=summer_sale:

• ✅ Counted: User arrives with the campaign parameter and converts in the same session
• ❌ Not counted: User arrives with the campaign parameter, leaves, then returns directly (no UTM) and converts

3. Visitor-Based Attribution (Conversion Insights)

The Conversion Insights dashboard uses visitor-based attribution to show the complete user journey. If a visitor ever matched a segment criteria, all their conversions are attributed to that segment.

How It Works:

• Once a visitor is associated with a segment (e.g., mobile device, UTM campaign), all their conversions are attributed to that segment.
• This provides a complete picture of how different user segments perform across their entire journey.
• A single conversion may be attributed to multiple segments if the user matched multiple criteria across different sessions.

Example:

A user's journey:

1. Visits on mobile with utm_campaign=summer_sale
2. Later visits on desktop with no UTM parameters
3. Converts on desktop

In Conversion Insights, this conversion is attributed to:

• ✅ Mobile devices (user used mobile at some point)
• ✅ Desktop devices (conversion happened on desktop)
• ✅ Summer sale campaign (user arrived via this campaign)

4. Why Different Models?

Statistical Accuracy vs. Complete Picture

5. Practical Implications

When Filtering Experiments by UTM Campaigns:

• Experiment Stats: Shows conversions that happened in sessions with that UTM campaign
• Insights: Shows all conversions from users who ever arrived via that campaign
• The insights numbers will typically be higher because they capture the full user journey

When Filtering by Device Type:

• Experiment Stats: Shows conversions that happened on that device type
• Insights: Shows conversions from users who used that device type at any point
• Users who switch devices will appear in multiple device segments in insights

6. Best Practices

• For A/B test decisions: Rely on the experiment statistics with session-based attribution
• For understanding user behavior: Use the insights dashboard to see the complete journey
• Be aware of the differences: Don't be alarmed if numbers differ between the two views—they're measuring different things
• Document your approach: When sharing results, specify which attribution model was used

7. Industry Context

This dual-model approach aligns with industry standards:

• Google Analytics: Uses different attribution models for different reports
• Mixpanel: Offers both user-level and event-level attribution
• Optimizely: Maintains strict session-based attribution for A/B tests while offering broader analytics