Conversion rate optimization: the formula, six steps and a worked example

Points and relative lift

A change in conversion rate has two readings, and reports on CRO mix them up often enough to cause bad decisions.

• Change in points = CR new − CR old.
• Relative lift (%) = (CR new − CR old) ÷ CR old × 100.

Report both every time. “Conversion is up 1%” can mean +1 point or +1% relative, which are different results: on the 2% base in our example, +1 point is +50% relative.

Revenue per visitor

Revenue per visitor = conversion rate × average order value. It catches lifts that cost money. Assume a control page converts 2% at a $100 average order, which is $2.00 per visitor (assumption). Assume a variant with a discount banner converts 2.4% at an $80 average order: $1.92 per visitor (assumption, illustration only). In our example, the rate rose 20% relative. Revenue per visitor fell.

Customers and CAC at fixed spend

Hold spend and traffic constant and the chain is short:

• customers = visitors × conversion rate;
• CAC = spend ÷ customers;
• so CAC new = CAC old × CR old ÷ CR new.

In our example, $2,400 a month buys 4,000 visitors (assumption). In our example, at 2% you get 80 customers and a CAC of $30. In our example, at 2.5% you get 100 customers and a CAC of $24.

The same chain runs backwards from a goal. Assume you need 50 customers a month and convert 2%: you need 50 ÷ 0.02 = 2,500 visitors (assumption). In our example, at 2.5% 2,000 visitors are enough. A rate gain lowers the traffic you have to buy.

Step 1. Name the one conversion that pays

The macro conversion is the action that brings money or a firm commitment. Micro conversions are the steps before it. You track the micro steps to find where people leave, and you judge the result on the macro.

Pick the action closest to revenue that happens often enough to count. If paid orders are rare, use the nearest step that predicts them and say so in every report. Secondary metrics still earn a place on the dashboard. Optimizely, drawing on lessons from 127,000 experiments, notes that under-prioritized metrics such as search rate can increase conversions⁶.

Output: one conversion, defined in writing, with its counting rule.

Step 2. Measure the baseline and find the leak

Record the current rate with your chosen denominator over whole weeks, so both weekday and weekend traffic are in it. Then open a funnel report: the share of visitors who pass from each step to the next. The step with the steepest drop is where you look.

Only then open recordings and heatmaps, filtered to that step. Move from a pattern in the heatmap to the sessions behind it, and back. Watching random sessions across the whole site costs hours and produces anecdotes.

Output: the baseline rate and the leaking step.

Step 3. Collect what visitors and customers already say

Start with data you already hold: chat logs, support email, sales call notes, refund and cancellation reasons. Then read public reviews of your business and your competitors, and count complaints by theme: price, quality, reliability, speed. A theme raised by most reviewers outranks a theme raised once. The method is in how to find customer pains in reviews.

Ask people who converted two questions: what persuaded them, and what almost stopped them. The second answer often becomes your next hypothesis. Wording that gets honest answers is in customer interview questions.

Read competitors’ offer and quote pages as evidence: what they show before asking for money, which fees they list, which guarantee sits next to the button. See how to find competitors and read their sites.

Output: a ranked list of complaint themes and objections, each with a count and a source.

Step 4. Write the hypothesis

Build it in three stages.

1. Problem statement: what happens and where, backed by evidence. “Visitors who open the quote form leave at the phone number field; recordings show them pausing there.”
2. Refined statement: narrow it to one cause you can change. “Visitors don’t want a sales call before they see a price.”
3. If–then–because: one change, one expected movement in the macro conversion, one reason.

If we change [X] on [page or step], then [macro conversion] will rise, because [evidence: a complaint theme with its count, a funnel drop, a recording pattern].

Order the list by three things: the strength of the evidence, the share of visits that reach the page or step, and the cost to ship. A change on a step every visitor sees, backed by a counted complaint, and shippable in a day goes first. If the hypothesis concerns what you promise rather than how the page shows it, rework the promise itself with how to write an offer.

Output: a ranked list of hypotheses in this format.

Step 5. Test one change

• Change one thing per test, so you know what caused the result.
• Plan the sample size and the end date before you start, from the baseline rate and the smallest lift worth shipping.
• Run whole weekly cycles, because traffic and intent change by weekday.
• Read the result on the planned date against the planned threshold. The convention is a 95% confidence level⁵.
• Ignore the dashboard’s early “winner” readings. Early results swing, and stopping on a peak books a lift that later disappears.

If traffic is too low to reach the sample, ship the change. Then compare it with the baseline over the same number of whole weeks. Label that a before/after reading. It is weaker evidence than a split test, and the report should say so.

Output: a test plan written before launch. After the test, a result stated in points and relative lift, with its confidence.

Step 6. Ship, then check the money metrics

Before you call a change a win, check four numbers against the baseline:

• revenue per visitor;
• average order value;
• CAC at the current spend;
• refund or cancellation rate over the following weeks.

A higher rate with smaller orders or more cancellations can lose money. Log every result, failed tests included, so the next hypothesis starts from what you learned.

Landing pages

One page, one offer, one action. The headline answers the query or ad that brought the visitor: someone who searched for a moving quote should see the quote form and a price range on the first screen. In HubSpot’s comparison of signup forms, landing pages have the highest average conversion rate, 23%, because each is built for a single action¹. This section covers improving a page you already run; to build one from zero, see a landing page to test an idea.

Calls to action

Put one primary action on each screen. The label names the outcome: “Get my quote” tells the visitor what happens, “Submit” does not. Placement can outweigh design. In a HubSpot analysis by Pamela Vaughan, reported by Moz in 2016, 47–93% of a blog post’s leads came from anchor-text CTAs inside the text, and about 6% from the end-of-post banner⁷. That data comes from blog posts, so on a sales page treat placement as a hypothesis to test.

Forms

Every field has to earn its place. Ask for a phone number only if you will call, and explain why next to the field: “so the crew can reach you on moving day.” You can split a long form into steps, easiest questions first. Test a shorter version against the current one and watch which field people abandon.

Checkout and deposits

Checkout is where many buyers leave: Baymard Institute puts the average documented cart abandonment rate at 70.22%, an average of 50 studies². For a service, the deposit step is the checkout. Before the payment field, show the full total and what the deposit covers. Also state the refund terms and the next step. In the worked example below, unexpected extra fees appear as a reported trust risk.

Mobile

Run the whole flow on a phone yourself: tap targets, the keyboard type on each field, pop-ups that cover the button, scripts that fail. Speed is part of the flow. Google’s research shows that 53% of visits are likely to be abandoned if pages take longer than 3 seconds to load⁸. Visitors seldom report broken scripts in surveys; you find them in the funnel as a drop that appears on mobile only.

The competitors a renter compares

The check found 9 direct competitors, example by our method^ex. Five of them carry a measured visibility share:

The top three sit within two points of each other. The check lists a risk that TWO MEN AND A TRUCK already covers about 45% of the offer, so renters may pick the established brand (example by our method)^ex. A new mover’s quote page has to show something the known brand’s page leaves out. Read each rival’s quote flow with competitor analysis before you write the hypothesis.

Complaints and the hypothesis

Complaint themes in the check, example by our method: Price 4^ex, Quality 1^ex, Reliability and trust 1^ex, Speed 1^ex. That is 4 + 1 + 1 + 1 = 7 complaints, and price accounts for 4 of the 7 (example by our method).

Two of the listed risks point at the same step, example by our method. Price complaints are the most frequent, so the hourly rate may lose price-sensitive renters^ex. Unexpected extra fees are a reported trust issue that could hurt bookings if hourly totals grow beyond the quote^ex.

If the instant quote shows the full estimate, with every fee listed, before the deposit step, then more quote requests will turn into paid deposits, because price is the most frequent complaint theme in the check and unexpected fees are a listed trust risk.

The primary metric is the share of quote viewers who pay a deposit. The companion metric is the cancellation rate after the deposit, which shows whether a clearer estimate brings firmer bookings. If renters reject the rate itself once they see the full total, the page cannot fix that; work on the price with how to set a price.

The limit: no demand data

The check found no demand source, example by our method^ex. Its first risk reads: local Denver demand is unmeasured, and the number of renters looking for two-mover crews is unknown (example by our method)^ex. Without traffic figures there is no sample size and no test length to promise.

So the first month is a baseline. Count each step from quote page visit to paid deposit, including instant quotes shown, against one denominator. Then choose between an A/B test and a before/after reading. The ways to measure interest before spending on the page are in check demand before you launch.

The money

If the full-estimate change lifts deposits by an assumed 20% (assumption), CAC moves from $30 to $25 and payback from 0.5 to 0.45 months, as the table in the previous section shows (example by our method)^exex.

Run the same check on your offer: Check your idea for free.