An A/B test duration calculator helps you turn traffic, conversion, and business assumptions into a realistic testing timeline. This guide explains the inputs that matter, shows how to estimate experiment length, and provides a planning worksheet and decision checklist for avoiding premature conclusions or unnecessary extensions.
Overview
There is no universal answer to the question, how long should an A/B test run? A useful duration estimate depends on how many eligible visitors your page receives, how often they complete the chosen conversion, how large an improvement you want to detect, and how much uncertainty you are willing to accept.
An experiment that runs for three days may collect enough data for a high-volume page, while a lower-traffic page may need considerably longer. The calendar duration is an output, not the main objective. The objective is to collect enough comparable observations to make a decision without mistaking normal variation for a genuine change.
Use an A/B test duration calculator before launching a test to establish a planning range. It should not be treated as a guarantee of statistical significance or as a substitute for reviewing data quality. A calculator is most useful when paired with a clearly defined primary conversion, consistent tracking, and a decision rule agreed upon before the results are visible.
For measurement foundations, review how to measure button clicks without overtracking. A test cannot be reliable if the conversion event is duplicated, missing, or triggered differently for each variation.
How to estimate
Most duration estimates follow a simple sequence:
- Choose the primary conversion. Select one outcome that directly represents the test objective, such as completed checkout, submitted lead form, account creation, or qualified trial start. Secondary metrics can provide context, but changing the primary metric after launch makes the decision harder to interpret.
- Record the baseline conversion rate. Use a recent period that represents the traffic and page experience you expect during the experiment. If the baseline is 4%, that means approximately four conversions per 100 eligible visitors.
- Set the minimum detectable effect. Decide the smallest improvement that would justify adopting the variation. A 25% relative lift from a 4% baseline would imply a target rate of 5%. The equivalent absolute change is one percentage point. Always state whether your calculator expects a relative or absolute effect.
- Set confidence and statistical power. Confidence reflects the tolerance for false-positive conclusions, while power reflects the likelihood of detecting the chosen effect if it is real. Use the settings required by your experimentation process and keep them consistent across tests.
- Enter eligible traffic and allocation. Provide the number of visitors who can actually encounter the experiment, not total site traffic. If traffic is split evenly, each version receives about half of the eligible visitors. Uneven allocation changes the time needed to collect comparable samples.
- Convert required sample size into days. Divide the estimated total visitors required by the expected daily eligible visitors. Round upward and allow for normal variation in traffic. This produces a planning estimate rather than a fixed end date.
A simplified planning formula is:
Estimated days = required eligible visitors ÷ average eligible visitors per day
The required visitor count comes from the baseline rate, minimum detectable effect, confidence level, power, and traffic split. A proper calculator handles the statistical calculation; your responsibility is to provide defensible inputs and interpret the result in context.
Inputs and assumptions
Baseline conversion rate
Use the conversion rate for the same page, audience, device mix, and conversion definition whenever possible. A broad site-wide rate may not represent the population exposed to the test. If performance varies sharply by device or traffic source, document that limitation rather than hiding it in a single average.
Minimum detectable effect
The smaller the improvement you want to detect, the more data the experiment generally needs. Set this threshold using business value and implementation cost. Testing for a tiny change may be worthwhile on a high-value funnel, but it may not be practical on a low-volume page.
Traffic allocation
A 50/50 split is straightforward and usually gives both variants data at a similar pace. A smaller share for the variation reduces its exposure but also slows evidence collection. Record the intended allocation and check whether the actual allocation matches it.
Primary conversion volume
Visitor volume alone is not enough. If only a small fraction of visitors convert, the experiment may need a longer calendar window than a click-based test. Use the event that represents the decision you are trying to improve, not merely the easiest event to count.
Confidence, power, and practical significance
Statistical significance does not automatically mean business significance. A small change can appear convincing with enough data but have little operational value. Conversely, an important change may remain uncertain when traffic is limited. Define both a statistical decision rule and a practical threshold before launch.
Test quality assumptions
Duration calculations assume that visitors are assigned consistently, tracking works throughout the test, and the variants are exposed to comparable audiences. They also assume that the conversion window is understood. If users may convert several days after the first visit, ending the test immediately can omit late conversions.
Worked examples
Example 1: high-volume landing page
Suppose a landing page has a 5% baseline conversion rate and receives 2,000 eligible visitors per day. The team wants to detect a relative improvement of 20%, which would mean testing for a target rate near 6%. With an even traffic split, both versions receive roughly 1,000 visitors per day. Entering these assumptions into an A/B test duration calculator may produce a required sample size and a calendar estimate. The team should round the estimate upward, confirm that a full weekly traffic pattern is represented, and allow time for conversions that occur after the initial visit.
Example 2: lower-volume pricing page
A pricing page converts at 3% and receives 300 eligible visitors per day. The team wants to identify a 10% relative improvement, equivalent to a change from 3% to 3.3%. Because the desired effect is small and the page has limited traffic, the required sample can be large relative to daily volume. The result may be a long experiment. That is not a calculator error; it reflects the difficulty of distinguishing a small change from ordinary variation. The team may choose a larger, commercially meaningful effect, improve the page before testing, or use a higher-volume proxy metric while treating it as secondary evidence.
Example 3: uneven traffic allocation
If a variation receives only 20% of eligible traffic, it will collect observations more slowly than the control. The test may therefore need more calendar time even if total site traffic is unchanged. Before launch, calculate duration using the smallest allocation and confirm that the reduced exposure is justified by risk, rollout policy, or operational constraints.
These examples illustrate why a duration estimate should be revisited when the baseline, traffic level, allocation, or target effect changes. It is a planning model, not a promise that a result will be decisive on a particular date.
When to recalculate
Recalculate before launch whenever a key input changes. Revisit the estimate during the test if the traffic mix shifts materially, the allocation is incorrect, the conversion event changes, tracking is interrupted, or the expected daily visitor volume differs from the planning assumption.
Do not extend a test simply because the current result is inconvenient or because the variation is temporarily behind. Likewise, do not stop early only because a dashboard shows an attractive lead. Early fluctuations can reverse as more visitors arrive. Follow the pre-set rule unless there is a documented tracking or safety issue.
Use this practical decision checklist before declaring a result:
- Was the primary conversion defined before launch and recorded consistently?
- Did both variants receive the intended audience and traffic allocation?
- Is the sample size at or above the planning target?
- Has the test covered normal traffic cycles relevant to the business?
- Have delayed conversions and the defined attribution window been included?
- Does the observed change meet both the statistical rule and the minimum practical effect?
- Were there outages, releases, campaigns, or tracking changes that could explain the result?
- Is the next action clear: adopt, reject, iterate, or run a follow-up test?
Record the inputs, launch date, allocation, primary metric, decision rule, and final interpretation. When traffic or conversion rates change, return to the calculator with the new assumptions. This keeps experiment duration connected to current conditions rather than an outdated estimate, and it creates a repeatable process for conversion rate optimization.
