Improve conversion rate
A conversion rate cannot be judged without knowing what the traffic was
The same page can convert at one per cent and at eight, depending entirely on who arrived and what they were promised on the way in. Before changing anything on the page, work out whether you have a traffic problem, a message-match problem, a friction problem, a trust problem or a path that is simply broken.
Before you change anything
The number on your dashboard is probably four numbers averaged
A conversion rate is a fraction. The top is whatever you decided counts as a conversion; the bottom is whatever sessions the tool happened to include. Both halves are choices, and neither is usually made deliberately, which is how a site-wide percentage ends up meaning almost nothing.
Consider what a single figure blends together. Brand searches from people who already intended to buy sit in the same average as a broad display campaign. Blog readers researching a term sit alongside people on a pricing page. Mobile sessions on a poor connection sit alongside desktop sessions in an office. The composite moves whenever the mix moves, and the page can be entirely innocent.
So the first job is not to improve the rate. It is to split it: by traffic source, by device, by page template, and by whether the conversion counted is an enquiry or something softer like a newsletter signup. Very often the split is the whole diagnosis, because one segment turns out to be dragging everything else down and nobody had seen it separately.
Only then is the question worth asking. Not "is one per cent bad", which has no answer, but "why does this segment convert at a third of the rate of that one, when the page is the same".
Differential diagnosis
Five causes, and the evidence each one leaves behind
Once the rate is segmented, the cause is usually one of these five. They are listed roughly in the order they are worth checking, which is not the order in which they are usually proposed.
- The traffic was never going to convert.
- The page is fine and the audience is wrong. The tell is a low rate on one source and a normal rate on the others, or a lot of sessions with almost no scroll depth and immediate exits. Broad match, display placements, an article ranking for an informational term, or a social post that went further than intended will all deliver people who had no purchase in mind. Changing the page will not repair a mismatch that happened before anyone arrived.
- The page does not answer the promise that brought people to it.
- Message match. The ad or the search result offered something specific and the page opens with a general statement about the company. The tell is a high exit rate within a few seconds on a page that performs normally for other sources. This is the cheapest thing on this list to fix and the most frequently broken, because the person who wrote the ad and the person who owns the page are often not in the same conversation.
- Something in the way is making it hard work.
- Friction: a form asking for eleven fields when three would do, a required phone number people will not give, an account demanded before checkout, a page that takes six seconds to become usable on a mid-range phone. The tell is people reaching the point of action and stopping there — form starts without submissions, carts without checkouts. This is measurable rather than speculative, which makes it the most satisfying cause to have.
- They believe you can do it and not that they should trust you with it.
- A trust gap. The tell is a long, engaged visit that ends without action, often with a detour to your about page or a search for your name elsewhere. The missing pieces are usually specific: no price guidance at all, no named humans, no evidence of comparable work, no answer to the obvious objection, no visible terms. Design polish does not substitute for any of these, and adding more persuasion tends to make the gap wider.
- The path is genuinely broken.
- The form errors silently on a particular browser. The confirmation never sends. The phone number is not a link on mobile. The payment step fails for one card type. The consent banner blocks the button underneath it. This is rarer than the other four and worth ruling out first anyway, because it is the only one where the fix is free and the loss is total for everyone affected.
Matching the fix to the cause
What each cause responds to, and what it ignores
The most expensive mistake in conversion work is applying the right fix to the wrong cause, which produces a large invoice and a flat line.
| Dimension | Responds well to | Wasted on it | Where to look for evidence |
|---|---|---|---|
| Wrong traffic | Targeting, keyword and placement changes upstream of the page | Any change to the page itself, including a full redesign | Rate by source and by campaign, exit speed, scroll depth |
| Message mismatch | Rewriting the opening of the page to repeat the promise made | New photography, animation, layout experiments | Rate by landing page and by ad group, exits in the first seconds |
| Friction | Fewer fields, fewer steps, faster rendering, clearer errors | More persuasive copy above a form nobody can complete | Form starts against submissions, field-level abandonment, device split |
| Missing trust | Price guidance, named people, comparable work, plain terms | Louder calls to action and more urgency devices | Long engaged sessions ending in nothing, detours to about pages |
| Broken path | Testing the actual journey on real devices and fixing what fails | Anything at all until the break is found | Error logs, cross-browser testing, submitting the form yourself |
The order to work in
Cheap and certain first, expensive and speculative last
Complete the journey yourself, on a phone
Real device, mobile data, a browser you do not normally use, an address nobody recognises. Go all the way to the confirmation. This takes fifteen minutes and finds the broken-path cause outright often enough to justify doing it before anything else.
You get: A completed journey, or a named break in it
Split the rate until it stops being an average
By source, by device, by template, by conversion type. Stop when each segment is describing one kind of person doing one kind of thing. If one segment is dragging the composite down, you may already have your answer without touching the page.
You get: A segmented view rather than a single figure
Read the promise that brought people in
Put the ad, the search result title and the first screen of the landing page side by side. Read them in that order, as a stranger would. Where the third one does not obviously continue the first two, you have found a message-match problem worth fixing this week.
You get: A promise-to-page audit for your main entry points
Measure the point where people stop
Instrument form starts separately from submissions, and steps separately from completions. Knowing that forty per cent of people begin the form and eight per cent finish it is a different problem from nobody starting it, and the two lead to opposite fixes.
You get: Step-level instrumentation on the conversion path
Decide what you can honestly test
Work out how many conversions per variant you can realistically gather in a sensible window. If that number is small, do not run tests — make reasoned changes one at a time and read them over quarters. Say which approach you are taking before you start, not after you see the result.
You get: A stated testing or change approach with its limits written down
Accountability
How we read a conversion result, and what we refuse to call a win
Conversion work is unusually easy to fake, because the number moves on its own and there is always a change to point at.
What we hold to
- Deciding the stopping point and the success measure before a change goes live
- Reporting results by segment, including the segments where the change made things worse
- Saying plainly when your volume does not support testing, and proposing something else
- Measuring to the outcome that pays, not to the click that precedes it
- Keeping a written record of every change and its date, so the trend line can be read later
What we will not do
- Quote an industry benchmark as though it were a target for your site
- Declare a winner from a few days of data because the line looked good
- Recommend a redesign as the first response to a low conversion rate
- Promise a percentage uplift before we have seen the traffic behind the rate
- Report a rate improvement that came from removing traffic without saying so
Questions
Conversion questions with uncomfortable answers
What is a good conversion rate?
There is no usable answer to that question, and any figure quoted to you should be treated as marketing rather than evidence. A rate depends on what counts as a conversion, which sessions are in the denominator, what the traffic was buying, how considered the purchase is, and how the tracking is configured.
The comparison that is genuinely useful is against yourself: the same page, the same traffic source, the same conversion definition, this month against last. That comparison can be wrong for boring reasons, but at least the reasons are knowable.
Our rate dropped and nothing on the site changed. What happened?
Look at the traffic mix before you look at the page. A new campaign, a broad-match expansion, a piece of content that started attracting a different audience, or a shift toward mobile will all move the rate without anything on the page changing at all.
Then check the tracking. A tag that fires twice, a consent banner change, or a redirect that drops the conversion event will produce a clean-looking drop that has nothing to do with behaviour.
We do not have much traffic. Can we still test?
Not in the way testing is usually described. Detecting a small improvement reliably needs a number of conversions per variant that most sites simply do not produce in a reasonable timeframe, and running a test anyway produces a confident-looking result that is noise.
What works instead is fewer, larger, reasoned changes, evaluated over a longer window, with qualitative evidence — session recordings, support enquiries, sales calls — carrying more of the weight. It is less satisfying and considerably more honest.
Do we need a redesign to improve conversion?
Almost never as a first step, and a redesign is the single most common way a working conversion rate gets destroyed. A full rebuild changes hundreds of variables simultaneously, so if the rate moves afterwards nobody can say which change did it.
Redesign when the current site genuinely cannot express what you need to say or cannot be edited without a developer. Otherwise, fix the specific cause you have diagnosed and keep the rest still.
How long should we leave a test running?
Long enough to cover whole weeks, because behaviour on a Tuesday and a Saturday differ, and long enough to reach the number of conversions you decided on before you started. Deciding the stopping point in advance is the part that matters.
Stopping when the result looks good is how most in-house testing programmes accumulate a folder of winners that never showed up in revenue.
Send us the rate and the traffic behind it
Give us analytics access and the three pages that matter most commercially. We will segment the rate, complete the journey ourselves on a real device, and tell you which of the five causes you are looking at before anyone proposes a change.
If we don't deliver the work we agreed to deliver for reasons within our control, you don't pay for the undelivered work. Read our guarantee
Related
Where to go next
- structured conversion workWhere the diagnosis on this page turns into a testing or change programme.
- the interface and flow itselfFor the friction and broken-path causes, which are design problems.
- page speed measured on real devicesA page nobody waits for has no conversion rate worth discussing.
- the campaigns that decide who arrives
- a site producing no enquiries at all
- the wider question of where customers are lost
Last updated · Reviewed by Zubair Afzal