How Long to Run a Creative Test Before Calling It: Advanced Facebook Ad Creative Testing

Quick answer

Run a creative test for the time it takes to accumulate 50 conversions per variant OR 4 full days, whichever is later. Anything shorter underweights the weekend day-of-week effect. Anything longer wastes spend on dying creatives.

Why fixed durations fail

The "3-day rule" comes from a 2017-era playbook when accounts were smaller and Meta's auction was flatter. Today, CPMs swing 30% inside a single day, and audiences cycle through different demographics on weekends versus weekdays. A 3-day window starting on Wednesday will look very different from one starting on Friday. The fix is to anchor the window to a full week-cycle, not to a fixed clock.

The framework, step by step

Calculate your minimum runtime: (50 conversions per variant) / (expected daily conversions per variant). For most £1k+/month accounts this is 2-4 days.
Add a "weekday completeness" check: the test must include at least one weekend day and one weekday.
The earliest you can call a winner is the later of the two: conversion floor or 4 calendar days.
The latest you should run is 10 days. Beyond that, ad fatigue contaminates the result.
Cut underperformers daily during the test using the £50 threshold rule, but do not declare a winner until the runtime conditions are met.
Lock the result, archive the test, and ship.

Example test matrix

Daily spend	Target CPA	Expected conv/day per variant (5 ads)	Min runtime
£30	£20	0.3	7 days
£100	£30	0.7	7 days
£300	£30	2.0	5 days
£600	£40	3.0	4 days
£1,500	£50	6.0	4 days

Below £100/day you cannot reach 50 conversions per variant in a sane window. Reduce variant count or increase budget.

Pitfalls to avoid

Calling a 24-hour test. The first day is delivery shock, not signal.
Letting tests run 14+ days. Frequency creep contaminates the result.
Mixing weekend and weekday-only tests in the same report. Compare weeks-to-weeks, not days-to-days.
Forgetting the conversion floor. Time is not the same as data.
Cutting ads on day 1 because of CPM. CPM volatility on day 1 is not creative quality.

5 FAQs

What if I have a long sales cycle? Use a proxy event like add-to-cart, then validate purchases over 7-14 days post-test.

Should I test during Black Friday? No. Run the test before or after the spike. CPMs distort everything.

Is 4 days too long for cost-per-lead? No. CPM volatility is identical for lead gen and ecom.

Can I extend a test if it is close? Once. Add 2 days and re-evaluate. Beyond that, declare inconclusive.

Should I report mid-test numbers to stakeholders? Only with confidence intervals. Point estimates mid-test cause panic decisions.

Volume that fits the runtime

Pix-Vu helps you build the variant counts that match your budget — never more, never less. https://pix-vu.com.