Quick definition: Conversion rate (CR) is the percentage of a defined eligible population that completes a specified desired action during a stated observation window.
What is conversion rate?
Conversion rate turns a count of completed outcomes into a rate that can be compared across traffic volumes, periods, channels, and experiment variants. The outcome might be a purchase, completed registration, qualified lead, subscription start, account upgrade, or another action with a documented business or customer-value purpose. “Conversion” is not automatically a sale: it is the movement from a defined opportunity to the particular outcome a team has chosen.
A useful conversion-rate name states the action, population, and time frame. “Checkout-start-to-purchase conversion within seven days” is much more interpretable than “conversion.” The label protects against a common reporting failure: two dashboards using the same name while one divides completed purchases by sessions and another divides them by users who started checkout.
CR is related to, but not synonymous with, click-through rate, completion rate, activation rate, or retention. A click-through rate measures a click among impressions or recipients; conversion rate normally measures a later success among eligible units. Completion rate refers to finishing a defined process. Activation rate is an early product-value milestone. These distinctions matter because an apparently improved rate can reflect a changed denominator rather than a better experience.
Conversion-rate formula and denominators
The standard formula is:
conversion rate = unique units completing the conversion / unique eligible units × 100
The numerator and denominator must use the same unit. Purchases divided by sessions is a session conversion rate; users who purchase divided by users who are eligible is a user conversion rate. Neither can be casually described as the other. If one user buys twice, a binary user-conversion numerator normally counts that user once; an order-rate metric can count both orders but should have a different name.
| Denominator | Question answered | Important limitation |
|---|---|---|
| Assigned eligible users | Did the offered experience change conversion? | Includes users who do not engage after assignment. |
| Unique landing-page sessions | How often does an eligible visit convert? | Repeat visitors can contribute multiple sessions. |
| Checkout starters | How efficiently does checkout finish? | Excludes earlier-funnel changes. |
| Email recipients delivered to | How often does a delivered campaign produce the action? | Delivery and identity rules affect the population. |
Eligibility should be defined before the conversion opportunity. Do not divide purchases by “users who viewed the payment page” if the tested design changes who reaches that page; that conditions on a post-treatment event. For an A/B test, all randomized eligible users are usually the most decision-safe denominator because it estimates the impact of assigning the experience.
Conversion rate in A/B testing
CR is a common primary metric for changes designed to influence a binary action: a revised form, value proposition, checkout, sign-up flow, recommendation, or offer. Its advantages are direct interpretation and relatively simple statistical treatment. Before launch, write the exact event, unit, qualification rule, attribution window, baseline rate, minimum detectable effect, and guardrails into the test plan. Guidance on framing the change and expected outcome is available in our A/B test hypothesis guide.
Report both absolute and relative change. If control converts at 4.0% and treatment at 4.6%, the absolute lift is 0.6 percentage points and relative lift is 15%. The absolute number communicates practical scale: at 100,000 eligible users, the point estimate corresponds to about 600 additional conversions. The relative percentage is useful context but can sound large when the baseline is small. See how to interpret confidence intervals before treating a point estimate as a guaranteed outcome.
A positive CR alone is not a rollout decision. A more aggressive discount may raise purchases while reducing margin; a simplified lead form may raise submissions while lowering lead quality; an enrollment flow may improve immediate sign-ups while producing early cancellations. Specify revenue, quality, cancellation, refund, latency, support, and retention guardrails where they apply. This primary-metric discipline is covered in our guide to primary and guardrail metrics.
Worked conversion-rate calculation
An ecommerce team tests clearer delivery-date messaging. The analysis population is every unique visitor assigned to a variant who sees an in-stock product page, and conversion means a completed, paid order within 48 hours. Control has 12,500 eligible users and 625 converters. Treatment has 12,400 eligible users and 682 converters.
control CR = 625 / 12,500 × 100 = 5.00%treatment CR = 682 / 12,400 × 100 = 5.50%absolute difference = 5.50% − 5.00% = +0.50 percentage pointsrelative lift = 0.50 / 5.00 = 10%
The result is not complete until the team estimates uncertainty using the preselected method and checks that assignment, exposure, and event logging work consistently. It also compares average order value, cancellation rate, delivery complaints, and stock availability. If treatment concentrated conversion in products likely to ship late, the initial lift could be offset by a poorer customer outcome.
Data-quality caveats
Conversion events often cross devices and systems. A browser may record a click while the order service records payment success later; a logged-in customer can begin on mobile and finish on desktop. Prefer a server-confirmed success event for the numerator, deduplicate retries and duplicate webhooks, and join identities using a documented stable key. Record event time, timezone, currency or market, and event version.
Audit the denominator separately. A traffic decline in one arm may be an eligibility or exposure bug rather than a performance result. Bots, internal staff, test orders, payment retries, canceled authorization attempts, and migrated users need prespecified handling. A changing attribution window can also make newer cohorts look worse simply because they have had less time to convert.
Common conversion-rate mistakes
- Using an undefined denominator: “visitors” can mean people, sessions, or page views.
- Mixing units: orders divided by people is not a binary user conversion rate.
- Excluding non-exposed users after assignment: this can bias an experiment result.
- Reporting only relative lift: percentage points reveal practical magnitude.
- Optimizing a proxy: a lead, click, or registration may not create downstream value.
- Reading incomplete cohorts: conversion windows must mature before comparison.
Frequently asked questions
What is a good conversion rate?
There is no universal benchmark. Traffic intent, product price, market, funnel step, and the definition all change the rate. Compare equivalent populations and focus on a meaningful improvement.
Should conversion rate use users or sessions?
Use the unit matching the decision and state it explicitly. User-level rates often suit randomized product tests; session-level rates can suit visit-level funnel monitoring.
Is conversion rate a percentage or a decimal?
It is commonly presented as a percentage, but calculations may store it as a proportion. Label the format and avoid confusing 0.05 with 5%.
Can one user have more than one conversion?
Yes for order or event metrics. For a binary user conversion rate, count the user once; otherwise name the measure an order rate or conversions per user.
Why did conversion rate rise while revenue fell?
The variant may have shifted buyers toward lower-value orders, larger discounts, or refundable purchases. Inspect revenue per assigned user, margin, and refunds.
Summary
Conversion rate is the share of a precisely defined eligible population that completes a specified action in a stated window. Its credibility depends on consistent units, a stable denominator, reliable success events, and clear exclusions. In A/B testing, evaluate absolute lift and uncertainty alongside revenue, quality, and customer-experience guardrails.