Paid media
Know how many days a creative test needs before you can trust the result.
- Category
- Paid media
- Signup
- None
- Your data
- Stays in your browser
- Also in
- Claude, via MCP
How long should an ad creative test run?
Long enough for every variant to collect its target number of conversions. Days = (variants × conversions needed per variant) ÷ (daily spend ÷ CPA). Three variants needing 100 conversions each, at ₹3,000 a day and a ₹400 CPA, take 300 ÷ 7.5 = 40 days and cost about ₹1.2 lakh.
This test needs ~40 days, long enough that fatigue and seasonality creep in. Cut variants, raise budget, or lower the per-variant bar.
Rule of thumb: ~100 conversions per variant for a directional read, more for small lifts. Ending early is the most common testing mistake.
About this calculator
Calling a creative test winner after three days is the single most common way teams crown noise as a result, because a stable read needs a real sample, not a schedule that felt long enough. This calculator forecasts exactly how many days a test needs, from daily spend, expected CPA, variant count and your target conversions per variant, so you have a concrete end date before you start instead of an impulse to stop early.
How to use it
- Enter your daily test spend and expected CPA per variant.
- Enter the number of variants you're testing and the conversions needed per variant for a stable read.
- Read the forecasted test duration, the date it finishes, and the estimated total test cost.
- Don't call a winner before the forecasted duration completes, even if one variant looks ahead early.
Methodology
Total conversions needed is variants × conversions needed per variant, the aggregate sample size across the whole test.
Conversions per day is daily spend ÷ expected CPA, and test duration in days is total conversions needed ÷ conversions per day.
Estimated test cost is duration × daily spend, and the finish date is simply today plus the calculated duration.
The default target of roughly 100 conversions per variant is a common rule of thumb for a directional read, not a statistically rigorous significance calculation; smaller expected lifts between variants need a larger sample than this to be trustworthy, while dramatic differences can sometimes be read with less.
Test length and cost from spend, CPA and sample size
Conversions needed = Variants × Conversions per variant Conversions per day = Daily spend ÷ Expected CPA Test days = Conversions needed ÷ Conversions per day Test cost = Test days × Daily spend
- Variants
- Number of creatives being compared, including the current control.
- Conversions per variant
- Sample each variant needs before you read the result. The tool defaults to 100.
- Daily spend
- Total daily budget across all variants in the test.
- Expected CPA
- Cost per conversion you expect during the test, from recent results.
100 conversions per variant is a rule of thumb for a directional read, not a significance test. Small differences between variants need much larger samples; large differences can show sooner. The model also assumes spend splits evenly across variants, which platforms often do not do on their own; use a split-test or experiment feature if you need an even split.
Worked example: testing three new creatives
A brand wants to test two new video ads against its current best ad. It can spend ₹3,000 a day on the test, and recent campaigns convert at about ₹400 per purchase. It wants 100 purchases per variant.
- Conversions needed = 3 × 100 = 300.
- Conversions per day = ₹3,000 ÷ ₹400 = 7.5.
- Test days = 300 ÷ 7.5 = 40 days.
- Test cost = 40 × ₹3,000 = ₹1,20,000.
- Testing two variants instead of three cuts it to 200 ÷ 7.5 = 26.7 days and ₹80,000.
- Testing on a cheaper, earlier event such as add-to-cart at ₹80 needs only 300 ÷ 37.5 = 8 days, if that event predicts purchases.
Forty days is longer than most creatives stay fresh, so the test design needs to change: fewer variants, a cheaper conversion event that tracks purchases, or a bigger daily budget for a short period. To check whether the difference you see at the end is real, run the counts through the A/B test significance calculator.
What makes a good creative test?
A good test answers one question with enough data:
- One variable at a time: change the hook, the format or the offer, not all three, so you know what caused the difference.
- A fixed end point: decide the sample before starting and do not stop when one variant looks ahead. Stopping early on a lead inflates false winners, as Evan Miller explains in How Not to Run an A/B Test.
- Under about a month: the tool flags tests over 30 days, because audiences, seasons and creative freshness change over longer periods.
- A business metric: judge on cost per purchase or qualified lead, not click-through rate alone.
Creative testing in India vs the US
Lower CPAs in India let tests reach their sample faster at the same budget, so a test that takes six weeks in the US may take two in India. Language and regional variants are worth testing as separate variables, not mixed into the same test as visual changes. Festive periods distort results; avoid starting a test in the weeks before Diwali if you want a normal baseline.
Higher US CPAs make 100 conversions per variant expensive. Many US advertisers test on an earlier funnel event, or use the platform's built-in experiments that split audiences cleanly. Black Friday and the holiday season change CPMs and conversion rates sharply; see the seasonal ad CPM predictor before planning tests in Q4.
Common creative testing mistakes
- Calling a winner after a few days because one variant is ahead.
- Testing too many variants for the budget, so none gets enough conversions.
- Letting the platform shift spend to an early leader, so other variants never get a fair sample.
- Judging on click-through rate when the goal is purchases or qualified leads.
- Changing budget or targeting during the test, which changes the conditions mid-way.
Sources
- How Not to Run an A/B Test, Evan Miller. Accessed 2026-10-19.
FAQ
Long tests are more exposed to confounding factors, seasonality shifts, creative fatigue setting in mid-test, or audience composition drifting, any of which can contaminate the read. A test stretching past a month is often a sign to cut variants or raise budget rather than let external noise creep in.
It's a reasonable floor for a directional signal, especially when the difference between variants is large and obvious. For subtle lifts (a 5-10% CTR difference, for example) you generally need a considerably larger sample to be confident the result isn't noise.
Raise daily spend to shorten the timeline, reduce the number of variants being tested simultaneously so each gets more of the traffic, or lower the per-variant conversion bar if you can accept a less statistically rigorous, more directional read.
Early results are disproportionately driven by whichever variant randomly gets a lucky run of conversions first. That early lead frequently narrows or reverses as the sample grows, which is exactly what the forecasted duration here is meant to protect against.
Yes, by testing fewer variants, using a cheaper conversion event that predicts the real goal, or accepting a directional read on fewer conversions. Avoid declaring firm winners on very small samples.
Where available, yes. Platform experiments split the audience so each person sees only one variant, which gives a cleaner comparison than running ads side by side in one ad set.
Pause them and note what was different, so the next test builds on what you learned. A losing hook in one audience can still work in another, so keep the assets.
Numbers looking off?
If the model says one thing and the pipeline says another, that gap is usually where the revenue is. A 30-minute call finds it.
It starts with a 30-minute call. Pick a time below.