How to do a conversion analysis of your website

The work takes an afternoon with numbers you already have. You need a web analytics tool with conversion tracking in place, and ideally a heatmap or session replay tool for the diagnosis step near the end. Nothing else.

One distinction before starting. The metrics and reports this review reads are conversion analytics. The analysis is what you do with them: a pass through the data with a decision at the end.

A conversion is the action your site exists to produce. For a store it is a purchase. For a SaaS site it is a signup or a booked demo. For a service business it is a submitted form or a phone call. Pick one primary conversion, and treat the rest, like newsletter signups, as secondary.

Then confirm each one is actually recorded. Fire a test conversion yourself and watch it arrive. How to track website conversions covers the setup. Every number in the rest of the analysis divides by this definition, so ten minutes of checking here protects the whole afternoon.

The baseline is your overall conversion rate: conversions divided by visits, over a period long enough to smooth out weekday and weekend swings. A full month is the practical minimum. Write down the rate and the absolute count, because a rate can hold steady while the count falls with traffic.

One consistency rule for the whole review: pick a counting basis and keep it. Conversions divided by visits and conversions divided by unique visitors are both defensible rates, but they differ, and a review that mixes them will manufacture changes that never happened. Note which one the baseline uses.

Your own history is the comparison that matters. This month against last month, and against the same month last year if the site is seasonal. Published industry figures are a loose sanity check at best, and what counts as a good conversion rate explains how to read them without being misled.

The site-wide rate is an average of pages and audiences that behave nothing alike, so the next step is splitting it three ways.

By landing page. Rank landing pages by visits and look at the conversion rate of each. The pattern to find is a page with real traffic converting far below the site average. In Google Analytics, the landing page report carries a key events column that does this directly. Most analytics tools have an equivalent view. Bigdelta, for example, shows which channels, UTM campaigns and landing pages drive conversions, and ranks pages so the weak ones surface.

By traffic source. A source can send a third of your visits and a tenth of your conversions. That gap is a finding: the traffic is mistargeted, or the landing page does not match what the ad or post promised. Traffic sources, explained covers how the buckets work.

By device. Sites commonly convert worse on phones than on desktop, but the size of your own gap is the number that matters. A small gap is normal. A mobile rate at half the desktop rate points at a mobile-specific problem, usually the form or the checkout.

Write down the two or three worst combinations you find. "Mobile visitors from paid social on the pricing page" is a fixable brief. "Conversion is low" is not.

For the core journey, build a funnel: the ordered steps from landing to conversion, and the share of people lost between each pair. This locates the weak page more precisely than any page-level rate, because it shows where people who had shown intent gave up. The method, open versus closed funnels, windows and segment cuts, lives in funnel analysis. For this review, one funnel on the primary conversion is enough.

The numbers so far name pages and audiences without explaining them. A pricing page that loses mobile visitors looks the same in a report whether the cause is a broken toggle, an unreadable table or a price that scared people off.

So spend half an hour on the two or three worst pages from steps 3 and 4. Open the heatmap for the page on the device that underperforms, then watch a handful of session replays of visitors who left without converting. You are looking for the moment the visit went wrong. How to see where users get stuck is the routine, and for form pages form abandonment shows how to find the field that loses people.

By now there is a list of problems. Rank them by the conversions a fix would plausibly add, which is traffic multiplied by the rate lift you could reasonably expect. The arithmetic is rough and illustrative, and that is fine, because it only has to order the list.

Say the pricing page gets 2,000 visits a month and converts at 2 percent, and the heatmap showed a plan toggle that does not respond on phones. Fixing it might lift the page one point, which is 20 extra conversions a month. A badly converting blog post with 200 visits offers 2 conversions even if you double its rate. The pricing page wins, and the blog post waits.

This ordering is the whole point of the analysis. Without it, teams fix the page someone complained about most recently. With it, the first fix is the one that pays most.

Make the top fix, then measure the same way you set the baseline: same rate, same period length, the same breakdown that flagged the problem. One change per page per period, or you will not know which change moved the number.

Two cautions when reading the result. Check the traffic mix did not shift in the same period, because a new ad campaign can drop the site-wide rate while every page converts exactly as before. And judge the fix by the conversion count and rate together, on the segment you targeted. A mobile fix shows up in the mobile rate first.

A full pass is a quarterly habit, and worth repeating after anything that changes the site or the traffic: a redesign, a pricing change, a new campaign. Between passes, a weekly glance at the baseline rate and count catches breakage early. Keep the baselines in a simple sheet, because next quarter's analysis starts by comparing against them.

Analysing without a defined conversion. If step 1 is skipped, every rate in the review measures something vague. It happens more often than it should, usually on sites where the "conversion" is a visit to a contact page rather than a sent message.

Comparing rates across different traffic mixes. A month with a paid burst in it converts differently from an organic month, whatever the site does. Compare like periods, or read sources separately.

Fixing low-traffic pages first. A dreadful rate on a page nobody visits is a small number wearing a dramatic costume. Rank by expected conversions gained, not by how bad the percentage looks.

Treating the analysis as a report. The deliverable is the ranked fix list and the retest date. If the afternoon ends with a document and no change shipped, the conversion rate stays where it was.