Question
Why does qualified traffic arrive and leave?
Not "why is traffic low" — traffic was fine, and the visitors were the right ones. The question that mattered was what happened between arriving and abandoning, which is a question analytics can answer and opinion cannot.
Two stints: June 2020 — Dec 2021, then Aug 2022 — Aug 2024. UI/UX Developer, both times.
Constraint
Remote, a domain I did not know, and a live product with no staging cohort.
Automotive sales was not a field I arrived understanding, and every experiment ran against real customers rather than a test group — which sets a hard ceiling on how adventurous any single change can be, and puts the burden on measurement instead.
The first stint had established the groundwork: a full web application built and maintained, data-oriented UX research applied to content relevance, and website analytics managed through Google Marketing Platform and Hotjar. Coming back in 2022 meant the instrumentation already existed. That is the only reason a four-stage research process was affordable at all.
Decision
Four stages, ordered so that the qualitative work had something specific to ask about.
Decision 01
Instrument first, and comprehensively
Analytics tracking through Google Marketing Platform and Hotjar covering user behaviour, traffic patterns and conversion metrics — before any hypothesis, so the data was not shaped by what we expected to find.
Decision 02
Locate the drop-offs before explaining them
User journey data, heatmaps and session recordings, used to find the specific points where people left or struggled. Position before cause. A heatmap will not tell you why someone hesitated, but it will tell you precisely where, and that turns the next stage from a general enquiry into a specific one.
Decision 03
Then ask why, with usability tests and interviews
Usability tests and user interviews to get behind the numbers — motivations and pain points, the "why" that no funnel reports. Running this third rather than first is the decision: participants were asked about a moment we could already point to.
Alongside it I used affinity mapping to build genuine understanding of the automotive sales industry and its user behaviour patterns — the domain knowledge that made the interview answers legible.
Traded away: early qualitative insight. For the first stretch we had numbers with no explanation, which is uncomfortable to report on.
Decision 04
Validate by A/B test, not by argument
Systematic A/B testing on the critical user journeys to confirm or kill each hypothesis. On a live product with real customers this is the only honest way to change anything, and it settles design debates without anyone needing to win them.
Three findings, three fixes
| Finding | Evidence | Response |
|---|---|---|
| Content relevance | Users bounced from pages whose content did not match their search intent | Content optimisation and improved information architecture |
| Navigation friction | User flow analysis showed heavy drop-off at specific navigation points | Redesigned navigation structure, clearer hierarchy, better guidance |
| Form completion | Funnel analysis found complex forms and unclear requirements blocking key actions | Simplified forms with progressive disclosure and contextual validation |
All three are content and flow problems. None of them is a visual design problem, which is what I would have guessed on arrival and what the instrumentation ruled out.
Working across the handoff
Two things sat outside the research loop and mattered as much. I partnered with backend engineering on API integration and technical scoping for dashboard features, translating business requirements into prototypes — which cut design-to-development handoff time by 20%. And I established and maintained a cohesive design system to hold the experience consistent across the platform.
The API scoping work is the first time I owned a technical conversation instead of handing a file across one. Two jobs later that became the whole role.
Outcome
- 40%Qualified traffic increase
- 30%Engagement improvement
- 18%Signup abandonment reduced
- 15%UX satisfaction scores
- 20%Handoff time reduced
The 40% came from content optimisation and improved search visibility — worth stating plainly, because the content work changed what search engines surfaced as well as what visitors read, and only part of that is design. The 30% engagement figure covers time on site, pages per session and return visitor rates rather than a single metric chosen for looking good.
The 18% is the one I care about. Reducing signup abandonment is the direct answer to the question the project opened with, and it was achieved by redesigning landing pages and the onboarding flow with product and marketing — not by touching the product the users had come for.
Everyone assumed the dashboard was the problem. The instrumentation said the landing pages were, and the landing pages were not owned by anyone.
google marketing platform · google analytics · hotjar · heatmaps · session recordings · funnel analysis · a/b testing · affinity mapping · design systems · api scoping