Website Conversion Rate Optimization: A Practical System

Conversion rate optimisation has a reputation problem. A decade of "27 quick wins" posts turned a serious discipline into a game of button colours. Done properly, CRO is closer to diagnostics: you find where money leaks out of the funnel, you form a hypothesis about why, and you fix it in the order that pays best.

Here is the system we run, in the order we run it.

1. Instrument the funnel before you touch anything

You cannot prioritise what you cannot see. Before any redesign, we make sure every meaningful step fires a clean event with the parameters we will need later:

• Page view with template type and traffic source
• Key content interactions — filter used, size selected, calculator completed, video played to 50%
• Form start, form field errors, form submit
• Add to cart, begin checkout, each checkout step, purchase with value and items
• Micro-conversions: phone tap, email tap, directions tap, PDF download

Two details that save weeks later: fire events server-side where you can, and include a stable session and user identifier so you can join behaviour to revenue in the warehouse rather than trusting a platform report.

2. Diagnose with four lenses

Every conversion problem we have ever fixed sat in one of four buckets. Work through them in order.

Clarity

Can a stranger tell, within five seconds, what you sell, who it is for and what to do next? Run a five-second test with people outside your company. If they cannot answer, no amount of speed work will help.

Speed

Largest Contentful Paint above 2.5 seconds on mobile 4G reliably costs conversions. Interaction to Next Paint above 200ms makes forms feel broken. Fix images, fonts, third-party scripts and layout shift in that order — in our experience third-party tags cause more damage than any framework choice.

Friction

Count the fields, the clicks, the required decisions and the moments where a user has to leave to find information — a size chart, a delivery estimate, a price. Every one is a chance to abandon.

Anxiety

What is the visitor afraid of? Wrong size, hidden fees, slow support, being spammed, choosing wrong. Anxiety is answered with proof placed exactly where the doubt occurs, not with a testimonial slider in the footer.

3. Prioritise with ICE, and skip testing the obvious

Score every idea on Impact, Confidence and Ease from 1 to 5, multiply, and sort. It is crude and it works, mostly because it forces the team to argue about confidence explicitly.

Some items should never be tested — they should just be fixed. A broken mobile form, a 6-second page, a missing delivery estimate, an unreadable contrast ratio. Testing a defect wastes traffic you could spend on a real question.

4. Run tests that can actually conclude

Three rules keep experiments honest:

1. Calculate sample size first. Baseline rate, minimum detectable effect, power 80%, significance 95%. If the calculator says nine weeks and you have four, do not run the test — ship the better-reasoned version instead.
2. Fix the end date before you start. Peeking and stopping at the first green result is the single most common way agencies manufacture wins that evaporate.
3. Run full weekly cycles. Weekend behaviour differs from Tuesday behaviour. Always run in multiples of seven days.

For low-traffic sites — under roughly 1,000 conversions a month — classic A/B testing rarely reaches significance. Use sequential redesign with holdout periods, qualitative research and hard-to-argue-with repairs instead. Pretending you can test with 40 conversions a month is the expensive option.

5. Forms: the highest-yield surface on most sites

Forms are where intent goes to die. The fixes are unglamorous and reliably profitable:

• Ask for the minimum. Every optional field you delete typically lifts completion by a few points.
• Label above the input, never placeholder-only — placeholders vanish and break error recovery.
• Validate on blur, not on submit, and put the error message next to the field.
• Use the correct type and autocomplete attributes so mobile keyboards and autofill behave.
• Never clear the form on error. It is 2026 and this still happens.
• Say what happens next: "We reply within one business day" converts better than "Submit".

6. Checkout and the moment of truth

For e-commerce, checkout deserves its own audit. The changes that moved the needle most across our client base:

• Total cost visible early. Unexpected shipping cost remains the number one abandonment reason, year after year.
• Guest checkout first. Offer account creation after the order is placed, not before.
• Express payment above the fold. Wallet buttons remove a dozen keystrokes on mobile.
• One page, or clear step indicators. Ambiguity about how much is left causes drop-off on its own.
• Inline error recovery on card failures, with the reason stated plainly.

Cutting a four-step checkout to one page raised Northwind Supply's mobile conversion rate by 41% — more than every copy test we ran that year combined.

7. Performance work with a budget

Set a performance budget before design starts: a target LCP, a JavaScript weight ceiling, a maximum number of third-party scripts. Then hold the line. The usual wins, in order of return:

1. Serve correctly sized, modern-format images with explicit width and height.
2. Self-host fonts, preload the one you need, and use font-display: swap.
3. Audit third-party tags quarterly and delete the ones nobody can name an owner for.
4. Defer non-critical JavaScript; hydrate interactive components only where they exist.
5. Cache aggressively at the edge and keep the HTML response under 100ms server time.

8. Measure in revenue per session

Conversion rate alone can mislead. A discount lifts conversion rate and destroys margin. A higher-quality traffic mix can lower conversion rate while increasing revenue. Track:

• Revenue per session, segmented by device and channel
• Average order value and units per order
• Contribution margin per session where you can get cost data
• Step-through rates for each funnel stage, so you know where a change actually landed

9. Turning it into a programme

One-off CRO projects produce one-off results. The version that compounds looks like this: a monthly cycle with a research week, a build week and two measurement weeks; a documented backlog scored with ICE; a decision log recording what you tested, what happened and what you concluded — including the losers, which are usually more informative than the wins.

After a year you will have something more valuable than any individual uplift: an evidence base about your own customers that no competitor can copy from a blog post.