Conversion work
Conversion work, without the magic buttons.
Conversion rate optimization has a reputation problem, mostly earned by a decade of button colour case studies. The real practice is duller and considerably more useful.
What is conversion rate optimization and how does it actually work?
Conversion rate optimization is the practice of increasing the share of visitors who complete the action a page exists for, by locating where people drop out and removing the reason, rather than by guessing at layouts.
In practice it is a loop: measure what is happening now, form one specific explanation of why, change one thing, then check whether the change held up across enough traffic to mean anything.
On small business sites most of the available gain comes from unglamorous work: a faster page, a shorter form, a clearer next step, a tappable phone number and tracking honest enough to tell you which of those mattered.
What the work actually consists of
Four repeating steps: find out where people leave, work out why, change one thing, and verify. Everything else sold under this name is a variation on those four.
The finding step is measurement. Which pages receive the traffic, which of those produce enquiries, where in a form people stop, and how the answers differ between phone and desktop. Without that, every later step is decoration.
The working out step is a hypothesis, and it should be specific enough to be wrong. Something like people are abandoning the booking form at the insurance question because they do not have the details to hand is testable. Something like the page needs to be more engaging is not.
The changing step should be one change at a time wherever possible, because a release that alters the headline, the form and the layout at once produces a result nobody can attribute.
The verifying step is where most of this goes wrong, and it is the part worth being strict about. A change that coincided with a good month is not a proven improvement.
Find where people actually leave
Almost every site has one dominant leak, and it is rarely where the team assumes it is.
Walk the path in order and put a number against each step. How many arrived, how many reached a page that can convert, how many started an action, how many finished it, and how many of those became customers. The step with the biggest fall is where the work goes.
Doing this once usually settles a long running internal argument. Teams that were about to rebuild a home page often discover that the home page was fine and the booking form was losing two thirds of the people who started it.
Where the drop-offs usually are
Ordered by how often they turn out to be the real problem on a small business site, rather than by how interesting they are.
- Load time on the pages that receive search traffic, which are usually heavier than the home page because nobody inspects them.
- No obvious next step within the first screen on a phone, so the visitor has to scroll and decide at the same time.
- A form asking for more than the first conversation needs, including fields that exist because a system requires them rather than because a person can answer them.
- A phone number that is not a tappable link, on a site where most enquiries are calls.
- Unanswered objections: price, availability, parking, whether you handle their specific situation. If the page does not address it, the visitor resolves it by going elsewhere.
See the full checklist: Where the drop-offs usually areHide the full checklist: Where the drop-offs usually are
- Proof that proves nothing. Stock photography, unattributed praise and vague claims read as filler and cost you the credibility the page needed.
- Form errors handled badly, with messages that do not say which field is wrong or that clear what was already typed.
- Mismatch between the promise that earned the click and what the page actually opens with.
- Inaccessible controls, which silently exclude a share of visitors and usually correlate with confusing design for everyone else.
The most reliable lever is the least exciting one
Google commissioned Deloitte and 55 to measure the commercial effect of mobile load time across 37 brand sites and more than 30 million sessions. It was published in 2020 and it is still the largest public study of its kind.
SourceGoogle, Deloitte and 55, Milliseconds Make Millions, 2020 (37 sites, 30m+ sessions)
Those are correlations from 2020, measured on other people's sites, and we are not going to restate them as a forecast for yours. What they justify is a priority order: before testing a headline, check whether the page is fast, because speed affects every visitor on every page rather than one element on one of them.
Conversion work against a redesign
These get confused constantly, and choosing the wrong one wastes either a lot of money or a lot of time.
| What you are weighing | Conversion work | A full redesign |
|---|---|---|
| What changes | One element, one page, one form at a time | Everything, at once |
| What you learn | Which specific change moved the number | Whether the new site is better overall, roughly |
| Risk profile | Small and reversible | Rankings, tracking and habits all move together |
| What it needs from you | Clean measurement and patience | Content, decisions and a migration plan |
| When it is right | The site broadly works but leaks at known points | The structure, the platform or the positioning is genuinely wrong |
| Typical failure | Chasing noise on too little traffic | Losing traffic at launch because redirects were an afterthought |
If you cannot name which step of the path is failing, you are not ready to rebuild. Diagnose first, because a redesign that fixes an unidentified problem also destroys the evidence of what it was.
Why most small sites should not run split tests
A split test compares two versions and asks whether the difference is real or chance. Answering that honestly needs enough conversions on both versions that a handful either way cannot swing the result, and most local business sites do not produce that volume within a sensible timeframe.
Run one anyway and you get a result that looks decisive and is not. Teams then ship the losing variant with confidence, and the next test contradicts the last one, which is exactly what you would expect from noise.
Read the full breakdown: Why most small sites should not run split testsHide the full breakdown: Why most small sites should not run split tests
The honest alternative for lower traffic sites is sequential improvement with clear before and after windows, applied to things that are obviously broken rather than marginally different. Fix the four second load time, shorten the nine field form, make the phone number tappable, answer the price question. You are not proving a two percent lift, you are removing a defect.
Where testing does earn its place is on high traffic paid campaigns, where the volume exists and the clicks are being paid for individually. Even then, run one change at a time and decide the measurement window before starting rather than stopping when the numbers look good.
One more discipline worth keeping: write down what you expected before you look. It is the cheapest protection there is against reading a story into a random month.
Some pages convert badly because the audience is wrong
Before optimising a page, check whether it was ever being shown to people ready to act.
A page attracting researchers will convert at a fraction of a page attracting buyers, and no amount of layout work closes that gap. The fix there is the traffic, the keyword targeting or the offer, not the button.
This is also why the search terms report and the channel breakdown belong in conversion work. Sometimes the most effective conversion improvement available is switching off the campaigns that were bringing the wrong people.
Read next
The pieces this work depends on.
What a good conversion rate is
Why the benchmark question has no honest answer, and what to measure instead.
What a good conversion rate isWhat a call to action is for
The single element most often rewritten and least often the actual problem.
What a call to action is forMaking a site load faster
The most reliable conversion lever, in the order that finds the bottleneck.
Making a site load fasterTracking where leads come from
The measurement this whole practice sits on, including tracked calls.
Tracking where leads come fromThe web design guide
How the build itself removes most of these problems before they exist.
The web design guideFind the leak before you rebuild
A free audit walks the path from arrival to enquiry and reports the step losing the most people, with the numbers behind it.
Straight answers.
How long does conversion work take to show results?
Removing an obvious defect, such as a slow page or a broken form on mobile, can show up within weeks. Anything subtler needs enough traffic to separate a real change from a normal fluctuation.
Decide the measurement window before you start, or you will be tempted to stop at whichever point the numbers look best.
Do I need special software for this?
Not to begin with. Analytics that records the actions you care about, call tracking, and a way to see where people stop in a form will cover most of the useful work.
Heatmaps and session recordings can help you form a hypothesis, but they show what a handful of people did rather than what most people do, so treat them as clues rather than evidence.
Is A/B testing the same as conversion rate optimization?
No. Testing is one method for verifying a change. Most of the practice is measurement and diagnosis, and on lower traffic sites the verification step usually has to be a before and after comparison instead.
Presenting a split test as a requirement for conversion work is a good way to spend money on a result that will not be statistically meaningful.
What is the single most common fix?
Shortening a form and making the phone number tappable on mobile. It is unglamorous, it takes an afternoon, and it regularly outperforms anything else on the list for local businesses.
Page speed is a close second, and it is the one that improves every page at once rather than one of them.
Can conversion work hurt my SEO?
It can if changes strip out content, remove internal links or break structured data in the pursuit of a cleaner page.
Keep the substance and change the presentation. A page can be far simpler to use without being thinner.
Should I hire someone or do this myself?
The diagnosis is worth doing yourself first, because it costs nothing but attention and it prevents you buying a solution to a problem you do not have.
Bring in help when the fix needs development work, when tracking is unreliable, or when you want the measurement designed by someone with no stake in the result looking good.
Where this comes from.
Primary documentation and published research behind the guidance on this page.
- Google Ads Help: About conversion measurement (opens in a new tab)
- Google Analytics Help: Engagement rate and bounce rate in GA4 (opens in a new tab)GA4 defines bounce rate as the inverse of engagement rate, not as a single-page visit.
- Google web.dev: Web Vitals (opens in a new tab)The current thresholds for LCP, INP and CLS, and the 75th percentile rule.
- W3C: Web Content Accessibility Guidelines (WCAG) 2.1 (opens in a new tab)
Talk to the team
A short call, a look at how the business currently shows up, and a straight answer on what we would do first.