Metrics·Glossary term

Conversion

Conversion A/B testing Reference guide

Conversion is a concept used in metrics, kpis & business outcomes.

Quick definition: A conversion is a documented action by an eligible person, account, or session that fulfills a defined business or customer-value objective.

What is a conversion?

A conversion is an outcome, not an inherently positive click. A retailer may define an online order as a conversion; a B2B software company may use creation of a qualified workspace; a public service may use successful completion of an application. The action should represent a meaningful step toward value for the customer and the organization, be observable in data, and be specific enough for two analysts to identify the same event.

Teams often call every tracked action a conversion. That makes the term less useful. Viewing a pricing page, clicking a button, submitting a low-intent form, starting checkout, and completing payment are distinct events with different implications. A good metric specification names the exact action, unit, eligibility population, timestamp, window, source of truth, and exclusions. For example: “unique new accounts assigned to the experience that create a verified project within 14 days.”

Conversion differs from conversion rate. A conversion is the counted success; conversion rate is the proportion of eligible units that convert. It also differs from a micro-conversion and macro-conversion. A micro-conversion is an intermediate behavior such as adding an item to cart or starting a trial. A macro-conversion is usually the primary outcome, such as a paid order or an activated account. Intermediate events can diagnose a funnel but should not automatically replace the outcome the change is meant to improve.

Defining a conversion precisely

Start with the decision the measure will inform. If a checkout experiment determines whether to launch a payment option, “payment successfully captured” is stronger than “purchase button clicked.” If onboarding changes are intended to help teams obtain value, a completed, durable setup may be stronger than “checklist opened.” Then define the measurement contract:

  • Unit: person, session, order, device, account, workspace, or household.
  • Eligibility: who had a real opportunity before the outcome and how exclusions are determined.
  • Event: the semantic action and the system that confirms it.
  • Window: the time allowed after eligibility or exposure.
  • Counting rule: whether repeated successes count once or many times.
  • Attribution: how the outcome is linked to a campaign or variant.

Use the same unit in any associated rate. If a conversion means a unique account’s first paid subscription, the numerator is converted accounts and the denominator must be eligible accounts. A person-level numerator over a session denominator can be useful only if labeled as a session rate and interpreted accordingly.

Conversions in A/B testing

In an A/B test, conversions provide the successes for a binary metric. Assign eligible units before exposure, log exposure consistently, and analyze people in the variants to which they were assigned. This intent-to-treat approach estimates the effect of offering the variant. Removing users who did not reach a later page may bias the comparison, because the variant could affect that journey.

Preselect one primary conversion when the test is designed around a central action. Multiple primary outcomes increase the chance of a false discovery unless the analysis plan accounts for multiple comparisons. Secondary conversions can identify mechanism: a new registration flow may improve email verification but not first-use completion. Before launch, state whether the test is seeking a change in conversion, revenue, quality, or a composite decision criterion. This guide to writing an A/B test hypothesis explains how to make that causal claim testable.

Conversion count by itself is rarely enough because variants may receive unequal eligible traffic. Calculate conversion rate, show both counts and denominators, and report the absolute difference in percentage points with uncertainty. A control rate of 2.0% and treatment rate of 2.4% is a 0.4-point lift and 20% relative lift; both labels matter. Estimate the interval and avoid declaring a winner based only on a threshold crossing. See our p-value explanation and confidence-interval guide.

Conversion scenario

A SaaS company tests whether a template gallery helps new workspaces reach initial value. Its macro-conversion is “workspace publishes a report using real data within 21 days.” A micro-conversion is “workspace selects a template.” In control, 1,000 of 5,000 assigned workspaces publish a report. In treatment, 1,125 of 5,000 publish.

control conversion rate = 1,000 / 5,000 = 20.0%
treatment conversion rate = 1,125 / 5,000 = 22.5%
change = +2.5 percentage points, or +12.5% relative

Template selection rose much more sharply, from 30% to 55%. That supports the proposed mechanism but does not replace the macro-conversion result. The team checks 30-day retained use, report errors, support requests, and whether templates lead users to publish useful reports rather than empty demonstrations. It also verifies that both variants wrote the report-published event from the same backend source.

Conversion data-quality caveats

Use an event that reflects the underlying outcome, not merely a client attempt. Browser events can be blocked, duplicated by retries, or emitted before a server transaction fails. Payment, enrollment, and provisioning conversions are often best confirmed from a server-side system of record. Preserve an event identifier, user or account identifier, event timestamp, event version, and status so that deduplication and reconciliation are possible.

Identity is especially important when conversion crosses devices or when several users contribute to an account outcome. Define whether a shared workspace conversion credits every member, its owner, or the workspace itself. Exclude test records, employees, bots, fraud, and canceled transactions with rules fixed before reading the experiment result. A changing event schema, timezone, or refund rule can create an artificial trend.

Allow the conversion window to mature. A customer exposed yesterday has had less opportunity to convert than one exposed last month. Compare cohorts with equal follow-up, or use a preplanned time-to-event method. Do not backdate an outcome to manufacture a lift, and do not silently change the attribution window between variants.

Common conversion mistakes

  • Counting clicks as completed outcomes: attempts and confirmed success are different events.
  • Choosing the easiest event: a superficial micro-conversion may not predict value.
  • Leaving the unit implicit: one account conversion can involve multiple people.
  • Changing the event during a test: the two arms are no longer measured consistently.
  • Counting duplicate payments or retries: use stable transaction identifiers and success states.
  • Ignoring cancellation and quality: a converted record may later be reversed or unqualified.

Frequently asked questions

Is a conversion always a purchase?

No. It is any pre-agreed outcome that represents the objective being measured, such as verified signup, activation, qualified lead, or completed task.

Can a test have more than one conversion?

It can track several, but specify which one is primary and how multiple testing will be handled. Secondary outcomes are useful for diagnosis and guardrails.

What is the difference between a micro-conversion and a macro-conversion?

A micro-conversion is an intermediate action; a macro-conversion is the main desired outcome. The distinction depends on the decision and funnel.

When should a repeat event count as a conversion?

Count repeats when the business question concerns events or orders. Count once when the measure is whether a unique unit achieved a binary milestone.

Why can conversion rise while retention declines?

The experience may persuade poorly matched customers, lower qualification standards, or promise more than the product delivers. Pair conversion with later outcomes.

Summary

A conversion is a precisely defined success event for a stated population and period. It becomes a trustworthy metric only when its unit, eligibility, counting rule, confirmation source, and attribution window are explicit. In experiments, analyze conversion rates with reliable exposure data, uncertainty estimates, and downstream quality guardrails.

Sources