Experiments
A/B tests evaluated where your data lives.
Run experiments on the same event stream that powers your analytics. The numbers in the report match the numbers in every chart.
Why split testing is usually noisy
- Most A/B platforms sit on their own event pipeline, so the win-rate they report does not match your dashboard.
- Late-stage tests need stratified analysis. Cheap tools only do single-segment comparisons.
- Statistical significance gets reported, but practical significance (lift size, confidence interval width) usually does not.
How leatmap does it
Variants assigned in the SDK.
Stable, sticky assignment per visitor. No flicker on first paint, no leakage across reloads, no double-bucketing.
Stats, every night.
A nightly job computes p-value, lift, and confidence intervals on every active experiment. Wake up to a clear answer instead of poking at spreadsheets.
Stop on win or stop on harm.
Set thresholds for ship-it and kill-it. The system flags the moment you cross either line, so you do not run a losing variant for two extra weeks.
Segmented analysis, no replumbing.
Filter the experiment view by any event property. The bucketing stays consistent, the conclusions stay honest.
Stop measuring with crossed fingers.
Get a tracking plan you can trust, a collector that enforces it, and a dashboard you actually want to open.