What 36 users knew that the team didn't

A new holiday booking platform about to launch — and nobody had watched a real user try to use it

Holidays by flydubai was preparing to launch a full holiday booking product — not just flights, but packages combining flights, hotels, room selection, excursions, and in-flight add-ons into a single end-to-end booking flow. Complex by nature. High-stakes for users making real financial decisions.

The design team had built something visually polished. But polished surfaces can hide deep usability problems. My job was to find them before real users did — and turn what I found into clear, prioritised direction for the product team.

I was accountable for this study end-to-end: study design, participant recruitment, session facilitation, data analysis, and the final report with recommendations. I co-led Team 1 in parallel with Team 2, running sessions simultaneously across four days to test at scale within a tight pre-launch window.

Setting up a test that would actually tell us something useful

I chose a moderated think-aloud protocol — not unmoderated remote testing, not a survey. A holiday booking flow is complex and multi-step. We needed to understand not just where users failed, but what they were thinking and expecting in the moments before they failed.

I recruited across nationalities, professions, and travel experience levels — South African, Scottish, Kenyan, Egyptian, Jordanian, Portuguese, British, Indian, Filipino, Russian, Hungarian, and more. flydubai serves a diverse international customer base. A test run only with Indian participants in Dubai would have been locally convenient but analytically misleading.

Who we tested with — and why it mattered

The three segments reflected fundamentally different mental models coming into the booking flow — not just demographic differences. A single-segment test would have hidden the fact that some features worked well for one persona and actively failed another. The divergent results were the finding.

The data told a clear story — once we knew where to look

Task completion rates by persona · UT Report p.7

Error rates per task · UT Report p.8

The most useful data is often what people say out loud

Think-aloud sessions generate something no analytics dashboard ever will: the unfiltered moment of confusion, the exact point where a user decides to push through or give up. Each quote maps directly to a structural failure in the flow.

Three critical failures, three structural interventions

The three most severe findings each required not a UI patch but a structural rethink. Here is the direct evidence-to-recommendation chain.

Participants view on booking a holiday

Participant-validated booking flow · UT Report p.16

Eight high-severity recommendations, each tied directly to evidence

Every recommendation was tied directly to an observed failure — not a heuristic judgment. Delivered with clear justifications to the IA and development teams, rated by impact: HIGH = blocking user success at critical funnel points. MEDIUM = significant friction but recoverable. LOW = quality-of-life improvements that matter post-launch.

From test findings to product decisions

What this study taught me about research at scale

Running 36 moderated sessions over four days with a diverse, international participant pool is a logistical challenge as much as a research one. The coordination overhead — parallel teams, scheduling, no-show management, consistent protocol across facilitators — is real and often underestimated.

What I'd do differently: build a more explicit affinity mapping session between days 2 and 3, when patterns were starting to emerge but we were still mid-collection. Early patterns were already pointing clearly at the room modification problem. Catching that in real-time and probing deeper in subsequent sessions would have produced richer insight on the most critical failure point.