a builder's codex
codex · operators · Elena Verna · ins_experimentation-paralysis

Don't test what won't reach sample size in a month — pre/post is fine

By Elena Verna · Growth advisor · 2026-04-28 · podcast · Elena Verna 3.0 — 10 growth tactics that never work — Lenny's Podcast

Tier A · TL;DR
Don't test what won't reach sample size in a month — pre/post is fine

Claim

A/B testing has a sample-size cost most teams ignore. If the change won't accrue enough sample in 30 days, don't test it — ship it and use a pre/post readout (24h, 7d, 28d, 1yr). Reserve scientific A/B for high-traffic real estate or strategic pivots. Treating every change as an experiment is a velocity-killer dressed as rigor.

Mechanism

Underpowered tests yield indeterminate results — "no significant difference" — which leadership reads as "the change didn't work." Teams then revert good changes because the test couldn't prove them, or freeze in indecision. Pre/post readouts give honest directional signal at low cost, fast enough to keep velocity. Statistical rigor matters when the stakes justify it; it actively harms velocity when it doesn't.

Conditions

Holds when:

Fails when:

Evidence

"If we cannot collect the sample size in a month, we shouldn't test it. Period."

— Elena Verna on Lenny's Podcast, 2026-04-28

The default is pre-vs-post readouts at 24h, 7d, 28d, 1yr. A/B testing is reserved for high-traffic surfaces and strategic decisions.

Signals

Counter-evidence

A/B test purists argue any pre/post comparison is confounded by external variables. They are technically right and operationally wrong: in low-traffic environments, pre/post is the only honest tool. Use the A/B test religion only where data volume genuinely supports it.

Cross-references

Open the interactive view → View original source → Markdown source →