Conversion Rate Optimization
Conversion rate optimization, or CRO, is the practice of raising the percentage of visitors who take the action you want: purchase, sign up, book a call. A typical ecommerce site converts around 2%, which means roughly 98 of every 100 visitors leave without buying. CRO is the work of finding out why and fixing it. Much of the industry runs it as a stream of button and headline tests. Run against profit, it starts with the customer's reasons for leaving and is scored on contribution margin.
How it actually works
The mechanics are structured experiments: a hypothesis about why visitors leave, a variant page that addresses it, a split of live traffic between the two, and enough volume to read the result. The tooling for this is mature. Programs fail on what gets tested, what gets counted, and how the result is read.
How the result is read is where most programs accumulate false wins. The failure is commonly described as calling tests early, but the mechanism is peeking: checking a running test repeatedly and stopping when it shows significance inflates the false-positive rate well past the stated threshold, even when the team acts only on significant readings, because enough looks at random noise will eventually catch it crossing the line. The controls are set before launch: a sample size computed in advance from the minimum detectable effect, with the result read once when that sample is reached, or a sequential testing method designed to be monitored continuously. Significance also has to be practical as well as statistical: a 0.3% lift can be statistically real and still fail to repay the engineering cost of building and maintaining the variant, so the minimum effect worth detecting is set from the economics before the test is sized.
What gets tested has to come from research. The reasons visitors leave are recorded in session recordings, reviews, support tickets, and interviews, in the customer's own language about doubt, confusion, and price. A test built from those sources is grounded in an observed problem. A test copied from a best-practices list is a guess.
What gets counted has to be margin. A sitewide discount banner raises conversion rate and can lower profit on every order that includes it. A shipping threshold, a bundle, or a reordered page can each move the count one way and the economics the other. Judged against contribution margin per visitor, the verdict on a test can reverse: a variant that lowered the conversion count can raise profit, and a variant that raised the count can lose money. The financial case for the discipline is that the traffic is already paid for: media spend is fixed within the period, so a real conversion lift spreads the same acquisition cost over more orders and the incremental revenue drops almost entirely to contribution margin, which makes a durable lift on existing traffic one of the highest-margin dollars available to the business.
In practice
One client came to us with a $52 customer acquisition cost. The work that brought it to $13.50 combined repeatable testing patterns, machine learning to weigh the factors, and roughly 700 tests, each grounded in what customers said and scored against the economics of the order instead of the count of conversions.
Where we come in
We run CRO as part of the customer journey, not as a standalone page program, because the visitor who declines today may buy after the next message if the steps are designed together. Tests are built from qualitative research, prioritized by expected impact on profit, and read against contribution margin. We test small and prove results before we invest, embedded with your team or in fixed-scope sprints.
Start a Revenue Health Pre-Assessment →See it in action
Related terms
- A/B testing
- Splitting live traffic between two versions of a page and comparing outcomes. The method is sound; the results depend on the hypothesis and the metric being compared.
- Statistical significance
- The confidence that a test result reflects a real difference rather than chance. The guarantee holds only when the sample size is fixed in advance; repeatedly checking a running test and stopping on a significant reading, known as peeking, inflates false positives past the stated threshold.
- Minimum detectable effect (MDE)
- The smallest lift a test is sized to distinguish from noise, computed before launch from traffic volume and the required confidence. Setting it from the economics keeps the program from chasing lifts too small to repay the cost of building them.
- Contribution margin
- Revenue minus the variable costs of serving it: goods, shipping, fees, discounts. The metric a test should move, because a conversion can raise the count and lose money at the same time.
- Landing page
- The page a visitor arrives on from an ad or a link. Matching its message to the ad that brought the click is usually worth more than optimizing either in isolation.
- Session recording and heatmaps
- Tools that show where visitors scroll, click, and stall. They locate the problem; the reason for it comes from research in the customer's own words.
- Test velocity
- How many experiments a program runs in a period. Velocity compounds learning only when each test is grounded in a real hypothesis.