Architect an experimentation dashboard for culture

WHAT THIS TESTS This question checks whether you treat experimentation as an organizational capability rather than a statistics problem. The interviewer wants features that change behavior: how teams form hypotheses, share findings, and avoid repeating each other's mistakes. Anyone can build a results table; the signal is whether you design for learning and discovery.

A GOOD ANSWER COVERS A central searchable repository where every experiment lives with a structured record: hypothesis, target metric, segment, duration, and outcome. Templates at creation time that force a falsifiable hypothesis and a primary success metric before anyone can launch, raising baseline quality. Equal first-class treatment of negative and flat results so the organization records what does not work, which is often more valuable. Rich tagging, full-text search, and links between related tests so a team can discover that someone already tried their idea. Automated readouts that summarize the learning in plain language, not just lift and confidence intervals. Notifications and digests that broadcast notable results across teams.

COMMON WRONG ANSWERS Building only a metrics dashboard with p-values and uplift charts. Storing only winning experiments, which biases the institutional memory. Ignoring access and search, so knowledge stays siloed per team. Treating the tool as a reporting layer rather than a workflow that shapes how experiments are designed.

LIKELY FOLLOW-UPS How do you prevent low-quality hypotheses? Gate launch on a filled hypothesis template and peer review. How do you encourage logging failures? Make negative results celebrated in digests and tie them to a metric of learnings captured. How do you handle metric definitions consistently? A shared metrics catalog.

ONE CONCRETE EXAMPLE A marketplace company builds such a dashboard. Before launching, a PM must enter a hypothesis and primary metric from a shared catalog. After the test, an auto-generated readout states the learning in a sentence and tags it by surface. Months later, a new growth team searching checkout finds three prior pricing tests, including two that failed, and skips a redundant experiment, saving weeks of effort.