A pattern that started on one page and ran the whole site

CEL Solicitors is a UK consumer law firm on a no-win, no-fee model handling fraud and scam recovery, housing disrepair, and personal injury claims. The people who land on it aren’t shopping around. They’ve usually just lost money, or they’re living in a house their landlord won’t fix, and they’re scared they’ve left it too late. That’s the frame of mind the site had to meet, and it wasn’t meeting it.

The brief I was handed was straightforward: fix conversion on the service pages. The trap in that brief is that it sounds like a page problem. Find the worst page, redesign it, move on. So before I designed anything, I audited every core service page against the same rubric to see what I was actually dealing with.

What I found changed the shape of the project. It wasn’t one bad page. It was the same handful of failures, repeating on every page. A page-by-page fix would have patched the symptoms four times over and missed the cause. So I stopped thinking about redesigning pages, and started thinking about designing one pattern: proving it once, then rolling it out everywhere.

I ran every page through the same lens: usability heuristics, accessibility and content clarity, scored keep / review / needs improvement. Using one rubric across all of them is what turned four separate critiques into a single diagnosis. Four failures, and they showed up everywhere:

The brief was to fix conversion on the service pages, and the head of marketing wanted the fraud page turned around first. Proposing a site-wide pattern instead meant asking for more time before anything visible changed, which was a hard sell with paid traffic running.

So I didn’t argue for it in the abstract. I put the audit scorecards for every page side by side in a 25-minute review and let the repetition make the case: if the same failures sit on every page, fixing one page leaves the cause everywhere else.

We agreed to pilot on fraud with the explicit goal of proving a pattern. The business kept its urgency, and I got the scope.

You can’t design a pattern for a journey you haven’t seen whole. So before I touched a single page, I mapped the entire fraud claim end to end, from the moment someone finds CEL, usually mid-crisis and straight off a search or a news mention, through eligibility, triage, and the long stretch of case management, to the outcome and what happens after it.

Mapping it whole showed me two things a page-by-page view never could. First, the emotional shape of it: where trust has to be won (early, before anyone fills in an eligibility form) versus where it has to be held (through the quiet, anxious weeks of case management, when nothing visible is happening).

Second, it forced me to design the paths most sites quietly ignore: the unsuccessful outcome, where someone we can’t help is still signposted to Action Fraud and Citizens Advice rather than left at a dead end, and the advocacy loop, where a happy client’s review becomes the first trust signal the next frightened person reads. That loop is the whole reason trust signals mattered so much. The output of one journey is the input to the next.

From there the pattern nearly wrote itself, because it was just the four audit failures turned the right way round: lead with the answer to “can you help me,” put the trust signals where the decision actually happens, write for a stressed person scanning on a phone, and make the numbers consistent enough to carry real weight.

I placed each fix using analytics alongside the interviews, surveys, session recordings, and reviews where people were really falling away, not where I assumed they were. And I also treated WCAG 2.1 as part of the build, not a clean-up pass at the end.

And I built it all on the fraud and scams service page first. That was where the business wanted to start, and I agreed for a reason of my own: it was the hardest place to be wrong. Fraud is CEL’s flagship vertical and carries the most paid traffic, so if the pattern held there, under real pressure and real spend, it would hold anywhere. Prove it where it counts, then scale.

The old one was a dark, dense wall on a single scroll. It asked for things like “an exact amount lost” from someone who has just lost money and may not have a clean figure to give. That’s a lot to ask at the worst possible moment, and it’s exactly where anxious people quietly give up..

The redesign didn’t make it prettier so much as make it kinder. I broke it into steps with a trust rail that stays visible, and changed two small things that matter more than they look:

Neither is a visual change. Both remove a moment where someone stalls. That’s the pattern in miniature: meet the person where they are, and take the friction out of the exact spot where fear turns into a closed tab.

The fraud and scams service page was the pilot. Once it held up, I extended the same pattern (layout, trust-signal placement and content approach) across the rest of the site: crypto, housing, disputes and personal injury. At that point the fraud redesign stopped being a redesign and became the template the whole site runs on.

That’s the part I’m proudest of, and it’s the part that’s easiest to verify. You can open any of those pages and see the same thinking at work.

A rebrand landed around the same time as this work, and the decision to ship it with the UX changes rather than staggering them was a shared one made with the business, not by me alone. What I owned was the design side of it: I led the UX and the website redesign through that combined release, which meant making the harder path work rather than the safer one.

Shipping everything at once as a single, bigger disruption instead of two smaller ones spared users and the team from two separate rounds of change, and meant we never put a new look on top of an unfixed journey. But it carried a real cost, and it landed on me to manage.

When the look, the words, and the flow all move together, it’s much harder to say which one moved a given number. That’s exactly why I ran the changes that mattered as controlled A/B tests. Those results stand on their own, whatever the rebrand was doing alongside them. The new identity was the context I designed inside, not the thing I was measuring.

I had good research to lean on: reviews, session recordings, interviews and surveys. But the site needed fixing fast, so I let the analytics point me at where people were dropping off and moved from there. If I did it again, I’d hold the qualitative and quantitative against each other properly before deciding where the fixes went. Analytics tells you where people leave.

Interviews often tell you the thing that made them leave happened a step earlier, somewhere the numbers never flagged. Under time pressure I let the drop-off data lead, and I think proper triangulation would have sent some of the effort somewhere slightly different.

Each change traced back to a specific drop-off point in the analytics and the research, and the key ones were validated with controlled A/B testing before rollout. Behaviour, not opinion, deciding what shipped. I still work at CEL, so I can’t publish the performance figures. What follows is how each change was tested, which I’m happy to walk through in detail.