TEST / MARKETING TOOLKIT

A/B test traffic planner.

Can your traffic answer the question?

Your inputs

Visitor counts and conversion rates. Equal traffic to each variant.

Baseline conversion rate · %Use a stable estimate from the same eligible audience.Minimum relative lift to detect · %20% relative lift takes a 2% conversion rate to 2.4%, not 22%.Eligible unique visitors per dayTotal across both variants; assignment is assumed 50/50.

YOUR SCENARIO / LIVE CALCULATION

Visitors per variant: 21,109
Estimated traffic days: 43
Total eligible visitors: 42,218
Baseline conversion rate: 2%
Rate to detect: 2.4%
Absolute change: 0.4 pp
Full-week planning window: 49 days

Control · expected conversions at planned n422.2

Variant · expected conversions at planned n506.6

Approximate planning only: 95% confidence level, 80% power, 50/50 independent assignment. Do not declare a winner early because a running result looks promising.

A dash means the ratio has no valid denominator or no positive allowable ad budget.

WHAT TO DO WITH THISMake the question answerable.

At this traffic level, a small lift needs a long test. Consider a larger, meaningful change or a higher-traffic surface. Do not loosen the standard just to get a winner.

Example inputs. Edit them to explore your scenario.

Calculated in this browser. Your last valid scenario is saved on this device. A shared link includes your inputs. No ad account connection or automatic campaign changes. Tool interface is in English.

THE METHOD / WHAT IS INCLUDED

Understand the number
before you act on it.

Planning approximation for two independent proportions with equal allocation, two-sided α = 0.05 and 80% power. With p1 = baseline, p2 = p1 × (1 + relative lift), and p̄ = (p1 + p2) ÷ 2, visitors per variant ≈ [(1.959964√(2p̄(1−p̄)) + 0.841621√(p1(1−p1)+p2(1−p2))) ÷ (p2−p1)]², rounded up. Assumes independent visitors, stable traffic, one binary primary outcome, and a fixed-horizon test. It is not Meta’s required sample size or a test-results significance calculator. Rare outcomes, repeated exposure, clustering, multiple comparisons, and sequential stopping need a different analysis.

Defaults are illustrative, not industry benchmarks. Choose inputs from the same reporting scope. A calculated opportunity is a question to test, not a guaranteed result.

R statistics documentation · Two-sample proportions and power ↗

KEEP WORKING

RELATED RESEARCH