Selected work

GOOGLE MERCHANT CENTER · EXPERIMENTATION AND MEASUREMENT

Building and measuring layout experiments across three markets

From page variants that could only be compared if every interaction was counted the same way, to winning layouts implemented across three markets.

Problem

The team wanted to know which page layouts and structures would perform best for Google Merchant Center in three English-language markets, part of Google for Retail's 51-market global property. That meant running experiments whose comparisons would hold up even when a variant rearranged the page.

Strategy

Build each variant so it differed only where the test intended, and track every meaningful interaction the same way in every variant, so the analysts would be comparing behaviour rather than differences in how it was counted.

Execution

Across more than one round of tests, I built the page variants for each market, including rounds that changed the page's layout and structure. Because those changes meant the same interaction didn't always live in the same place, or even in the same component, I set up the GA4 interaction tracking so that each meaningful action was captured the same way in every variant, wherever it appeared on the page. I then checked the variants across all three markets before each round launched, working alongside the analysts who ran the comparisons.

Results

The tracking held up well enough for the analytics team to choose winners with confidence, and the winning variants were implemented.

The same interactions tracked the same way in two layouts, an illustrationTwo simplified page layouts. In Layout A the Sign up button sits in the hero and the Learn more link sits below a row of cards. In Layout B the Learn more link sits beside the hero and the Sign up button sits at the bottom of the page. Each interaction is marked with the same number in both layouts and sends the same GA4 event in both, so analysts compare like with like. This is an illustration, not the original pages.Layout ASign up1Learn more2Layout BLearn more2Sign up1Same event in both layouts, wherever it sits1sign_up_clickhero in A, page bottom in B2learn_more_clickbelow the cards in A, beside the hero in BAnalysts compare behaviour, not differences in how it was counted.Illustration, not the original pages
Illustration, not the original pages.