Conversion rate is the fraction of users who complete a desired action out of those who had the opportunity to complete it. If 1,000 users visit a signup page and 120 create an account, the conversion rate is 12%. It's one of the most common metrics in product experimentation because it's intuitive, directly tied to business outcomes, and well-behaved statistically.
In experimentation, conversion rate is a binary metric: each user either converts (1) or doesn't (0). This property makes it straightforward to analyze. The variance of a binary metric is determined entirely by the conversion rate itself (p times 1 minus p), which simplifies power calculations and sample size planning.
Why is conversion rate useful in experiments?
Conversion rate works well as an experiment metric for three reasons.
Direct interpretability. A 2 percentage point increase in checkout conversion rate has a clear business meaning. Stakeholders don't need a statistics background to understand the result or its implications. This makes it easier to get organizational buy-in for acting on experiment results.
Reasonable sensitivity. Conversion rates at moderate levels (5% to 50%) have enough variance to be sensitive to real changes without requiring enormous sample sizes. Extreme base rates create problems: a metric with a 0.1% conversion rate requires a very large sample to detect a meaningful relative change, while a 99% conversion rate has almost no room to improve.
Natural pairing with guardrails. When you use conversion rate as a success metric, adjacent conversion rates work naturally as guardrails. An experiment optimizing signup conversion can track purchase conversion as a guardrail to ensure that more signups doesn't mean lower-quality users.
What are common pitfalls with conversion rate metrics?
Wrong denominator. The denominator matters as much as the numerator. "Users who purchased divided by all users" is different from "users who purchased divided by users who reached checkout." Using the wrong denominator dilutes the metric. Confidence supports trigger analysis, which restricts the analysis to users who actually encountered the change, avoiding the dilution that comes from including users who never had the opportunity to convert.
Short-term bias. Conversion rate measures a single action at a single point. It doesn't capture whether the user came back, whether the conversion led to retention, or whether the action was valuable in the long run. A dark pattern that pressures users into converting will increase conversion rate while degrading the user experience. Pair conversion rate with downstream metrics or use it alongside guardrail metrics that capture user satisfaction.
Ignoring per-user rates vs. per-session rates. A user who visits five times and converts once has a per-session conversion rate of 20% but a per-user conversion rate of 100%. The right denominator depends on what question you're asking, and the choice affects both the result and its interpretation.