Swim England
A usability evaluation of the national governing body's website. Six swimmers and parents sat down to do seven ordinary things, and revealed that the site's real problem was never a broken feature. It was that nobody could tell who the website was talking to.
I ran the study end to end, and wrote the report Swim England received.
Recruiting and screening participants, writing the task scenarios, moderating every session, coding six transcripts by hand, and turning roughly sixty logged issues into four prioritised recommendations a stakeholder could act on without reading the appendices.
One website serving two completely different audiences, and admitting it to neither.
Swim England is the national governing body: it certifies teachers, supports clubs and facility operators, and also happens to be where a parent goes to find their child a swimming lesson. Both audiences share one site, one menu and one homepage. The evaluation set out to measure visual appeal, findability and usability, and kept returning to the same underlying question the site never answers: is this page for me?
Mixed methods, because a score alone never tells you why.
An explanatory sequential design: quantitative SUPR-Q scores to locate the weak dimensions, qualitative think-aloud and interviews to explain them. Six participants (parents looking for lessons, casual swimmers, competitive swimmers) recruited to cover the range of people who arrive at this site with different intentions.
Finding out what they already expected
Before touching the site, each participant talked through their history with sport-related websites: what they normally come for, what usually goes wrong, how satisfied they end up. That gave me their mental model in their own words, so that later in the session I could tell a genuine usability failure from a mismatch with an expectation they arrived holding.
Seven tasks, thought out loud
First impressions, then visual elements, then four findability tasks: volunteer for an event, become a swimming teacher, find a nearby pool, explore lesson progression. Each had a pass condition written before the session, so success was recorded rather than judged. The think-aloud protocol caught the hesitations that a completion rate hides completely.
Two realistic scenarios, split evenly
Half the participants had to find a pool with disabled access within ten miles for a wheelchair user; the other half, a pool near EC1V 0HB running activities for children aged three to five. Counterbalancing meant the filter and map findings came from two independent groups rather than one repeated path, and both halves hit the same filter logic.
Coding by hand, after deleting the AI's codes
Transcripts came out of Dovetail with automated codes attached, and they were unusable: the model labelled words without understanding what the participant was trying to do. So the first step of coding was deleting all of them, then highlighting and grouping by hand into themes like unclear labels and navigation breakdowns. Slower, and it surfaced issues the automated pass had flattened.
Every issue then went into a rainbow spreadsheet with an identifier, a location, a category, the impact observed, which participants hit it, and a candidate solution. Severity followed Dumas and Redish: Level 1 blocks task completion, Level 4 is cosmetic. Affinity mapping ran alongside to cluster the highlights, which is where the surface-level complaints started resolving into structural causes.
Trusted, and still not coming back.
The SUPR-Q results split cleanly. Every participant rated the information quality well: the content read as credible and correct, which for a governing body is the hardest thing to earn and the easiest to lose. Usability, appearance and loyalty all scored low.
Loyalty was the most damaging result, and the follow-up interviews explained it: people believed the answers were on this site, and still did not want to return to look for them. Credibility was carrying a journey that navigation kept breaking.
The one exception ran the other way: all six named Poolfinder the most useful thing on the site, and said it alone would bring them back.
Plotting category against severity for each area showed where the damage was concentrated. Information architecture dominated everywhere, which is why it became recommendation one rather than a line item.
Three areas, three different failures.
A news column where the value proposition should be.
The first thing in view was the news feed, so participants concluded Swim England was an encyclopaedia or a swimming journal. One read the boxy, densely packed layout as an advertising site. Two decided the audience was professional swimmers, instructors or facility operators, and then questioned whether they should still be there. Nothing on the page was broken; it simply never said what the organisation does for a member of the public.
Not all of it failed. Four of six praised the box gallery components for their clean framing, visible titles and clear calls to action, and the Useful Links menu was liked specifically for staying in view while scrolling. The problem with that menu was vocabulary: labels like Heart of Aquatics and Don't put a Cap on Swimming meant nothing to anyone who did not already work in the sport.
A promotional banner recruiting volunteer working group members sat on the homepage until 10 March. Five participants tested while it was live and found volunteer opportunities smoothly. It was removed before the sixth session, and that participant struggled visibly before eventually succeeding. Same site, same task, one component's worth of difference. The task everyone failed, meanwhile, was the one with no banner at all: becoming a swimming teacher. Participants clicked Teacher, drowned in undifferentiated information, found no job-application CTA anywhere, and concluded they must be on the wrong page.
Four participants also flagged the grey navigation bar at the top of every page: they read it as a cookie notice or something belonging to another site, and skipped past a whole tier of the site's key functions without registering it existed.
The one thing they loved, held back by a scroll bug.
Poolfinder matched people's mental models almost exactly: simple layout, obvious controls, results where you expect them. All six named it the site's most useful feature and the reason they would come back. Which made its unresolved issues the most expensive ones in the study: this is the feature with something to lose.
One participant hit a bug that froze page scrolling entirely, cutting off map navigation mid-task. Three could not work out the zoom, so their searches returned less than the area actually held. The accessibility filter options were hard to interpret, and the filter logic was strict enough that reasonable searches came back with very few results, or none: the answer existed, the query just could not express it.
What they asked for was consistently practical: opening hours, photos and water temperature per venue, a bigger map panel, more icons on the venue tags they already found easy to scan, and a real Book a Pool route. The existing Venue info link disappointed everyone who tried it: it never led to the facility detail they had gone looking for. The banner ad above the map was noted, unprompted, as a distraction.
Six out of six arrived and did not believe they had arrived.
Every participant found the page hard to reach: the link sat in the low-visibility grey bar they had already learned to ignore. Then, having reached it, none of them was confident it was the right page. There was no clear heading to confirm it, the labels pointed elsewhere, and the layout gave no structure to hold on to.
The video thumbnail in the most prominent position showed children, and one adult casual swimmer concluded from it that Swim England only teaches kids, a whole audience written out by one image choice. Three did not connect the label Learners to themselves, and the side menu gave them no help. People expected lessons laid out by skill level; instead they got dense prose to wade through, and reported feeling fatigued rather than informed.
Several genuinely wanted to sign up, for themselves or their child, and could not find a button or link that would let them. Parents and ordinary swimmers left the page believing it was not built for them.
Four moves, ordered by severity rather than by effort.
Roughly sixty issues do not make a roadmap. Grouping them by cause left four, and the order matters: the first two remove whole classes of the others, so shipping them in reverse would mean redesigning surfaces that are about to move.
Reorganise the information architecture
Unclear labels and structural inconsistency produced the majority of issues, in every area tested. Rebuild the site map, then validate it with card sorting and tree testing on general consumers rather than staff: the people who wrote the labels are the only ones the labels currently work for.
Split the user flow in two
One flow for professionals (instructors, clubs, facility operators) and one for the public, including parents. Each group then sees only what is relevant to it, which fixes the audience confusion and the content overload in a single move, because they were always the same problem.
Modernise the UI
Cleaner layout, decisive colour and weight to mark what matters, and the two stacked navigation bars merged into one: the grey upper bar is currently mistaken for a cookie notice. Leaning into the blue of the association's own identity also does brand work the current palette does not.
Give articles a format
Large headings, honest thumbnails, icons, and font sizes and weights that hold up for every reader. Scannability was the difference between participants absorbing a page and abandoning it, and it is the cheapest of the four to ship.
To check the four moves survived contact with a real screen, I redrew the four pages the study broke on. Every callout is traceable to a numbered finding, so each design decision can be argued from evidence rather than taste.
By the sixth session the issues were repeating, which is the signal to stop, but saturation on six participants tells you what is broken, not how often or for whom. The severity ratings are my judgement calls applied consistently, not measured frequencies. Anything downstream of them inherits that.
The banner that vanished mid-study taught me the most. Five people sailed through a task and one struggled, for no reason inside their own behaviour: the site had changed underneath the study. It made the difference between a working component and a working structure impossible to miss: findability that depends on a promotion is borrowed, not built.
I would also fight harder for the unglamorous recommendation. Modernising the visual design is the one stakeholders reach for first, and on this site it would have repainted an architecture that was the actual cause. Learning to argue for sequencing (this before that, and here is what it costs to invert) mattered more than any single finding in the report.