Methodology

The science behind UX attribution

We use the same causal inference methods trusted by economists, clinical researchers, and policy makers—adapted specifically for measuring UX impact.

The Challenge

Why traditional analytics fail

Traditional analytics tools show you what happened. Revenue went up. Conversion increased. But they can't tell you why.

When you ship a new checkout flow the same week marketing launches a campaign, how do you know which one drove results? Correlation isn't causation—and “we think it was us” doesn't survive a budget meeting.

A/B tests help, but only for isolated, small changes. Complex UX work—redesigns, new flows, system-wide improvements—can't be cleanly A/B tested.

Our Approach

Causal inference for UX

We apply multiple statistical methods to isolate the true impact of your UX changes from everything else happening in your product.

Synthetic Control Method

We construct a 'synthetic' version of your product that didn't receive the UX change, using weighted combinations of historical data and control metrics. The difference between actual and synthetic outcomes is your causal impact.

Best for: Major redesigns, new feature launches, flow overhauls

Difference-in-Differences

By comparing the change in outcomes for affected users versus unaffected users, before and after your UX change, we isolate impact from time-based trends and external factors.

Best for: Targeted improvements, segment-specific changes

Regression Discontinuity

When UX changes are rolled out based on a threshold (time, user segment, etc.), we analyze the sharp difference in outcomes at that boundary to estimate causal effects.

Best for: Phased rollouts, threshold-based deployments

Bayesian Structural Time Series

We model what metrics would have been without your intervention using historical patterns, seasonality, and related time series. The gap between predicted and actual is your impact.

Best for: Ongoing improvements, continuous deployment

Validation

How we ensure accuracy

Multi-method triangulation

We run multiple analytical methods on each attribution and only report results when they converge.

Confidence intervals

Every attribution comes with statistical confidence levels. We show you the range, not just a point estimate.

External factor controls

We automatically account for marketing campaigns, seasonality, market conditions, and other teams' initiatives.

Continuous calibration

Our models learn from your specific product, users, and business patterns to improve accuracy over time.

The Output

What you get

For each UX change you ship, you receive:

Dollar value of attributed revenue impact
Confidence interval (e.g., 95% CI: $1.2M - $1.8M)
Statistical significance (p-value)
Methodology used and why
Executive-ready summary and visualizations
Foundation

Built on proven research

Our methods are based on peer-reviewed research in causal inference, econometrics, and statistical learning. The same techniques are used by:

Google (CausalImpact)Uber (causal ML)Netflix (experimentation)FDA (drug trials)World Bank (policy evaluation)

See it in action

Start with 10 free attributions and see exactly how we measure your UX impact.

Start free trial →