Conversion rate and how to optimise it

Conversion rate is the share of users, sessions, or other properly defined opportunities that end in an important action. That action may be a purchase, a qualified enquiry, registration, or completion of a key task in an application. Conversion-rate optimisation (CRO) means removing friction and checking changes, not mechanically “raising the percentage” at any cost.

The greatest value of CRO is better use of traffic you already have. It does not always cost less than acquiring a new audience — diagnosis, implementation, and a credible test also require work — but it helps improve the offer and experience on the basis of observable behaviour.

How to calculate conversion rate

The formula is simple:

number of conversions ÷ number of conversion opportunities × 100%

If a site received 1,200 sessions and generated 24 qualified enquiries, its session conversion rate is 2%. A meaningful comparison, however, requires a consistent numerator and denominator. A result calculated per user is not the same as one calculated per session, and one person may buy more than once.

Before opening a report, write down:

• which event is the conversion and exactly when you record it;
• whether the denominator is users, sessions, product views, or started processes;
• which test events, duplicates, and internal traffic you exclude;
• whether the result covers everyone or a particular source, device, or stage.

Without this definition, two people can present different, formally correct percentages and draw contradictory conclusions.

Which conversion really matters

Not every action has the same value. Downloading a file, watching a video, or clicking a button may be a micro-conversion — a sign that a user is moving in the right direction. A macro-conversion fulfils a business objective, such as a sale or a valuable conversation.

Situation	Micro-conversion	Macro-conversion
Online store	proceeding to the cart	paid order
B2B service	opening the brief	qualified enquiry
SaaS product	starting a feature	paid activation or renewal

A high percentage of a supporting action does not guarantee an end result. If a simplified form increases the number of submissions but lowers their quality, a local “conversion improvement” may increase service cost. Always observe the target metric and at least one guardrail metric, such as cancellations, returns, or the share of unqualified contacts.

Where to look for the cause of low conversion

Start with the journey, not the button colour. Combine four kinds of information:

1. Quantitative data shows at which step users most often drop off and which sources they come from.
2. Qualitative data — interviews, support requests, and usability testing — helps explain why.
3. Technical checks reveal form errors, mobile problems, slow loading, and broken integrations.
4. Business context checks the fit of the price, promise, scope, and follow-up service.

Typical points of friction include an unclear next step, asking for data too early, a hidden cost, a mismatch between an advert and the page, missing information needed for a decision, or an interface that fails to communicate an error. A session recording can point to the problem location, but it does not explain motivation by itself. That is why numbers should be confronted with direct observation and conversation.

How conversion optimisation works

A useful CRO process has six stages:

1. choose an important conversion and verify that it is measured correctly;
2. find the greatest friction on the journey;
3. formulate a hypothesis describing the audience, change, and mechanism;
4. choose the smallest safe way to test the hypothesis;
5. assess the result together with guardrail metrics and data limitations;
6. record the decision: implement, improve, reject, or gather missing evidence.

An example hypothesis: “If we explain next to the phone field that the number is used only to agree the scope of a conversation, more suitable customers will complete the brief because the fear of unexpected contact disappears.” This is better than “let’s change the form” because it can be verified and separates a cause from accidental correlation.

How to run an A/B test without false certainty

An A/B test compares a control variant with a variant containing a justified change. Users should be assigned to variants randomly and concurrently so that season, traffic source, or implementation order do not impersonate an effect.

There is no universal rule of “at least 7 days and 100 users”. The required sample size depends, among other things, on the baseline rate, the effect size worth detecting, variance, and the chosen analysis method. Before starting, agree on the primary metric, guardrail metrics, user-assignment method, and stopping condition. Do not declare a winner at the first favourable jump.

With low traffic, a classic test can take too long. In that case, usability testing, error analysis, interviews, a gradual rollout with observation, or a before-and-after comparison of stages can be better — with an explicit reservation that the result does not prove causation. Insufficient data is not permission for a stronger conclusion.

When conversion optimisation can cause harm

CRO is not for squeezing every click out of people. The conversion count can rise by hiding information, pre-ticking consent, or making cancellation difficult, but that result shifts the cost to the user and destroys trust.

Do not test everything at once either. A wholesale layout change can improve the result but will not show which mechanism worked. Conversely, isolating a minor detail makes no sense when the offer itself is unclear. The experiment’s scale should match the hypothesis’s scale.

Where to start

Choose one important journey and complete it yourself on a phone and a computer. Check measurement, read recent customer questions, and find the stage where missing information or an error blocks a decision. Only then create a solution variant.

If you are working on a campaign page, also see how to design a landing page. The choice of metric is clarified in the article about KPIs, and the approach to creating further hypotheses in the text on growth hacking. When you need to connect data with product decisions, see analytics for companies.