From gut feel to 100 experiments a year

The problem

Doodle shipped on intuition. There was no reliable event data, no shared definition of activation, and no way to tell whether anything that shipped had worked. That isn't really a tooling problem. It's a problem where every disagreement gets settled by seniority, because there's nothing else available to settle it with.

What we did

• Started with the journeys, not the tools. Sat with engineering and the analysts and defined what success, frustration and drop-off actually looked like inside the product, then instrumented those.
• Got clean events into Amplitude and Google Analytics, accessible company-wide, within weeks.
• Ran the first experiments scrappily on Google Optimize rather than waiting for a platform decision.
• Then ran a proper evaluation with data science, engineering and finance, and landed on GrowthBook.
• Trained every PM and designer on hypothesis design and success metrics, and ran company-wide demos of the results. Including the ones that failed. Especially those.

The call I had to make

I shortlisted two platforms and thought I had alignment. Then an engineer asked why a third wasn't on the list. I'd misread the Principal Engineer's earlier feedback and he hadn't felt heard. We reopened the evaluation, added his suggestion, and picked it. Asking for input isn't the same thing as making space for dissent.

Impact

• More than 100 experiments a year, from zero
• Company-wide analytics access inside a quarter
• PMs and designers writing and running their own tests
• The measurement layer that every other case study on this site was built on top of