Skip to content
Faisal Mahmud TusherDigital marketing, SEO & analytics
with Faisal Tusher

Conversion Rate Optimisation

Conversion Rate Optimisation: The Complete Guide

Conversion rate optimisation raises the share of visitors who complete the action a site exists for, by researching why people leave, writing testable hypotheses and running controlled experiments. This hub explains the loop from evidence to decision and introduces the four sections that cover research, testing, funnels and landing pages.
HubbeginnerUpdated

Learn this as a path

Prefer a guided order? These learning paths read pages from this section in sequence, with exercises and checkpoints.

In this section

What CRO is and is not

CRO is often sold as a bag of tricks: change the button colour, add a countdown, shorten the form, put a face in the hero image. Some of those changes help on some sites, and that is exactly the problem. Nobody can say in advance which ones, and a change that lifted one site's rate has lowered another's. What separates the discipline from the tricks is the order of operations. Research finds where people leave and why. A hypothesis states what change should fix that and why anyone believes it will. A controlled experiment (A/B Test) measures whether the change caused the effect or whether the week was simply better. The result is recorded so that the next test starts from more knowledge than the last.

CRO is not persuasion design, growth hacking or user experience work, though it borrows from all three. It is not about tricking people into converting: pressure tactics and hidden costs raise a short-term rate and raise refunds, complaints and churn with it, and consumer-protection rules in most markets take a dim view of them. It is also not confined to the landing page. A site's conversion rate is the product of every step from first visit to completed order or submitted lead, so the discipline reaches into checkout, forms, pricing pages, account creation and the emails that bring people back. The customer journey hub describes the stages people move through before and after they convert; CRO works on the stretch of that journey the site itself controls.

One definition has to exist before any of this starts: what a conversion is. A purchase, a qualified lead, a booking, a trial start, a demo request. It should be written into the measurement plan and recorded as a key event, and the key events and conversions guide explains how Google Analytics records it. Optimising a rate nobody has defined, or one that counts newsletter sign-ups alongside orders, produces confident decisions about the wrong thing.

The CRO loop

Every CRO programme, whatever its size, runs the same loop. The stages are distinct jobs with distinct outputs, and most failures come from skipping one.

StageQuestion it answersOutputSection
ResearchWhere do people leave, and why?A ranked list of problems, each with evidenceResearch
HypothesisWhat change should fix it, and why do we believe that?A written hypothesis with a metric and a predicted directionResearch
PrioritiseWhich hypotheses are worth testing first, given the traffic?A sequenced backlogResearch
TestDid the change cause the effect?A controlled experiment run to its planned sampleExperimentation
LearnWhat do we now know, and what does it change?A documented result, win or lose, feeding the next research roundExperimentation

It is a loop, not a line. A losing test is a research finding: the problem was real but the fix was wrong, or the problem was not what the data suggested. A winning test raises a new question about the next step in the funnel. The speed of the loop is set by traffic. A high-volume retailer can run several tests a month and let the experiments do much of the learning. A business with a few hundred conversions a month spends longer in research, tests fewer and bigger changes, and ships some fixes without a test at all where the evidence is overwhelming, such as a form that fails on one browser. Knowing which kind of programme you can run is the first prioritisation decision, and the prioritising tests guide covers it.

Research

Research is the section most programmes under-invest in, because testing feels like progress and research feels like delay. The conversion research hub argues the opposite: a test built on a guess is a coin toss with a timeline, and the way to raise the share of winning tests is to raise the quality of the evidence behind them. The methods fall into four groups. Quantitative research, from analytics and funnel reports, shows where people leave and how much each leak costs. Behavioural research, from session recordings and heatmaps, shows what they did on the page before leaving. Attitudinal research, from surveys, polls and interviews, is the only way to find out why. Usability testing watches people try to complete the task and catches what none of the others can.

No single method is enough. Analytics is precise about what happened and silent about why; a survey is rich about why and unreliable about how many. The research hub's core guides cover the methods and when to use each, the tutorial for writing a test hypothesis from the evidence, and the method for prioritising the backlog that results. Further tutorials there cover running a conversion research sprint, session recordings and heatmaps, on-page polls and post-purchase surveys, usability testing, expert review and turning customer language into page copy. The measurement that makes quantitative research possible, from events to funnels to segments, is covered in the conversion analytics page of the analytics hub, which is worth reading before the first research sprint if the tracking has never been audited.

Experimentation

A controlled experiment is the only way to know that a change caused a result. Everything else, including before-and-after comparisons, is confounded by seasonality, campaigns, news and the weather. The experimentation hub introduces the practice: the A/B test (A/B Test) as the workhorse, multivariate and split-URL tests as its relatives, and the statistics that decide when a result can be trusted.

The A/B testing guide covers the procedure: choose the unit of randomisation, choose one primary metric and a few guardrails, allocate visitors at random, check the test before launch, and run it to the sample size planned in advance. That last rule is the one most often broken. Evan Miller's essay on how not to run an A/B test, listed in the sources, explains why checking the result repeatedly and stopping when it first looks significant inflates the false-positive rate badly; the sample size and significance guide puts the same point in plain language. The testing pitfalls page in the same section catalogues the other ways a test lies: sample ratio mismatch, flicker, novelty effects, broken tracking and tests that interact. Further guides there cover conversion work without an A/B test for when the traffic or the change does not suit a split, evaluating a website redesign, shipping a winning test variant once a test has won, choosing between client-side and server-side testing and hiring and evaluating a CRO agency.

Testing is also a culture question. Most tests do not win, and a programme that only reports winners teaches people to stop running honest tests. A result that says the change made no measurable difference is information that saves the next person's time.

Funnels and checkout

A funnel (Funnel) is the ordered sequence of steps between arriving and converting: product page, basket, checkout, payment, confirmation for a shop; landing page, form, qualification, booking for a lead-generation site. Each step loses some share of the people who entered it. The funnel and checkout optimisation hub is built around one rule: find the step that loses the most people in absolute terms and work there first, because a modest improvement at the biggest leak is worth more than a large one at a small leak.

The funnel analysis tutorial in that section builds the funnel from the events already recorded, compares open and closed funnels, and segments by device and source until the drop-off has a location and a suspect. The checkout optimisation guide then covers the usual suspects in e-commerce: forced account creation, address and form friction, delivery costs revealed late, limited payment options, and error handling that loses the order at the last step. Checkout work needs a second metric alongside rate. Average order value (AOV) can fall when a checkout change removes the moment people added a second item, so a rate win can be a revenue loss. Cart and browse recovery, through email and remarketing, run alongside the fixes rather than instead of them. Further guides there cover the other pages a funnel passes through: pricing pages, category and listing pages, sign-up and trial-start flows, long application forms, booking and appointment flows, upgrade paths for existing customers and the confirmation page that follows the conversion, with a tutorial on checking conversion paths after a release for when the site changes.

Landing pages

A landing page has one job: continue the promise the ad or link made and move the visitor to a single action. The landing pages hub covers that job in its core guides. The anatomy of a landing page guide explains the sections and what each does: headline and subhead, hero and offer, benefits and proof, objections and the call to action, in an order that follows the visitor's intent rather than the organisation chart. Message match covers the agreement between ad, keyword and page that paid traffic depends on. The forms guide treats every field as a cost to be justified. Trust and proof covers reviews, logos, guarantees and security signals and where to place them. Speed and mobile explains why a slow page converts worse before anyone reads it, and the landing page versus homepage comparison settles where campaign traffic should go. Further guides there treat the homepage as a conversion page, accessibility as conversion work and honest persuasion with urgency, scarcity and defaults.

Landing pages are where CRO and paid media meet. A campaign's cost per lead is set as much by the page as by the bid, and the quickest improvement in many accounts is a page that matches the ad instead of a homepage that matches nothing in particular.

A first CRO programme

A first programme does not need a testing tool, a budget line or a dedicated team. It needs a defined conversion, honest measurement, a research sprint, and the discipline to write things down. In order:

  • Define the conversion and instrument it. One primary outcome per site or funnel, recorded as a key event, checked end to end. If the tracking is wrong, every later decision is wrong.
  • Establish the baseline. Conversion rate by device, by source and by landing page, over a period long enough to include a normal week and a quiet one. The conversion rate calculator turns raw counts into rates for comparison.
  • Run a research sprint. A few weeks of funnel analysis, recordings, an on-page poll and a handful of usability sessions, following the methods guide. The output is a ranked list of problems with evidence attached. The research sprint tutorial walks through it step by step.
  • Fix the broken things without a test. Errors, dead links, forms that fail on one device, prices that differ between the ad and the page. These are not hypotheses; they are bugs.
  • Write hypotheses for the rest and prioritise them. The tutorial and the prioritisation guide in the research section cover both. Traffic decides how many tests are realistic; be honest about it. Where it cannot support a test, the guide to conversion work without an A/B test sets out what to do instead.
  • Run the first test properly. One change, one primary metric, a planned sample, no peeking. Document the result whether it wins, loses or does nothing.
  • Set a cadence. Research refreshes quarterly, the backlog is reviewed monthly, and the key performance indicator (KPI) the programme reports on is revenue or qualified leads, not the count of tests run.
Illustrative example — not a real result

A subscription-box retailer's first quarter

A meal-kit retailer with about 60,000 monthly sessions and a 2.1% order rate starts a programme in January. The tracking audit finds the purchase event firing twice on refreshed confirmation pages, which had inflated the reported rate to 2.6%; fixing it is the first job. Funnel analysis shows the biggest absolute loss between the plan-selection page and the start of checkout: 14,000 people reach the plans each month and 3,400 continue. Recordings and a poll on that page point to confusion about whether delivery is included. The team ships a plain delivery line as a bug fix, then runs one A/B test on the plan page that replaces three plan cards with a single recommended plan and an "other plans" link. The test runs for four weeks to its planned 17,000 visitors per arm and reports a 9% relative lift in checkout starts with no fall in average order value, which was the guardrail. The documented result, the fixed tracking and a backlog of 11 ranked hypotheses are the quarter's output.

When a specific symptom has already appeared, such as a rate that fell after a relaunch, forms that are started and never sent, or a test that never reaches a verdict, the conversion rate troubleshooting section diagnoses it cause by cause.

Frequently asked questions

Is conversion rate optimisation only for e-commerce sites?

No. Any site with a defined outcome can run the same loop: a lead-generation form, a demo request, a booking, a trial start or an app download. The research methods, the hypothesis discipline and the testing statistics are identical; what changes is the conversion being counted and the funnel steps that lead to it. Lead-generation sites usually have fewer conversions than shops, which pushes the programme towards research and fixes rather than many small tests.

How much traffic does a site need before testing makes sense?

The useful measure is conversions rather than sessions, because a test has to detect a change in the rate at which visitors convert, and that needs enough conversions in each arm within a period the team will wait. The sample size and significance guide in the experimentation section shows how to work it out from the baseline rate and the smallest effect worth detecting. Below that level the programme is still worth running: research, usability testing and fixes shipped without a test produce most of the early gains on low-traffic sites.

What is a good conversion rate?

There is no useful universal figure. Rates differ by industry, by traffic source, by device, by price point and by what counts as a conversion, so a published average says little about any particular site. The comparison that matters is against the site's own baseline, segmented the same way every time, and the direction of travel after each change. The conversion rate calculator turns raw counts into rates so that those comparisons are consistent.

Terms used on this page

Tools to try

Frameworks

See it applied

Sources

  1. How Not To Run an A/B Test — Evan Miller

Related