Planning an A/B test from a benchmark

📅 2026-06-05

Planning an A/B test from a benchmark

A benchmark answers: “What have others seen?” An A/B test answers: “What happens here?” This guide connects the two.

Step 1 — Translate the insight

From the StatFacts card, write:

If context mismatch is large, widen your expected range or run a discovery test first.

Step 2 — Pick one primary metric

Match the insight’s outcome field when possible. Secondary metrics can guard against mixed effects (e.g. signup up, activation down).

Step 3 — Estimate baseline

You need your current rate to interpret relative lifts. Example:

See Relative vs absolute effects if this math is unfamiliar.

Step 4 — Set success and guardrails

| Element | Example | || | Success | +5 relative points vs control (conservative vs benchmark mid) | | Guardrail | No increase in support tickets; activation ≥ control | | Runtime | 2 weeks or until significance + minimum sample |

Benchmarks set ambition; your power calculation sets feasibility.

Step 5 — Size the sample (simplified)

Use a standard power calculator with:

If required traffic exceeds two weeks of volume, shrink scope or accept a higher MDE.

Step 6 — Document sources in the test brief

Prior: +12–18% relative signup completion (StatFacts, meta-analysis, mobile B2B SaaS) Link: /insight/signup-one-fewer-step_en Our success criterion: +5 relative points in 14 days

Future you (and leadership) will trust results more when the prior is explicit.

After the test