Bosca / Experiments

Results & Analysis

The verdict, with the math

Results come straight from the analytics you already collect — turned into a per-variation breakdown, a statistical read on the difference, and a plain recommendation: ship, keep running, or don't.

The scoreboard

Every variation, side by side

Each variation gets its own line — impressions, conversions, conversion rate, the lift over control, and the confidence behind it. The control is picked consistently, so the comparison is always anchored the same way.

  • Lift and confidence are computed against the control variation.
  • A verdict banner turns the numbers into a call you can act on.
  • Guardrail goals can flip a positive result to don't-ship on their own.
results · checkout-cta
variationrateliftconf
control4.8%
treatment5.6%+16.7%98%
SHIP

Goals

Not every metric gets a vote

A goal is either a rate — each converting person counted once — or a count, averaged per person. What a goal does with its result depends on the role you give it.

Primary

The metric the ship decision hangs on. Move it, and the verdict follows.

Secondary

Measured for context and reported alongside — but it doesn't drive the call.

Guardrail

A metric that must not regress. If it slips, it can veto an otherwise-positive result.

The statistics

Rigor you can trust

The numbers are computed properly, not eyeballed. Analysis runs frequentist by default or Bayesian when you want posteriors — and it reads from the same analytics warehouse the rest of the platform queries, so there's no separate metrics pipeline to keep honest.

  • Rates use a proportions test; per-person counts use a means test — each metric matched to its math.
  • CUPED variance reduction tightens count-based goals using pre-experiment data.
  • A sample-ratio check flags bucketing bugs before you trust a result.
  • Optional AI insights add a hypothesis read and follow-up ideas on top of — never instead of — the numbers.
analysis
methodfrequentist · Bayesian optional
varianceCUPED −38%
sample ratiopass · no SRM
sourceanalytics warehouse

Keep exploring

More on Experiments