Stress peaks the moment students commit to a room they haven't seen
Average self-rated stress when signing for a room, the highest of the five stages we mapped, up from 3.1 when applying. Moving in brings no relief for half the group: three of eight rated it 9 out of 10.
How might we use user research to reduce the barriers international postgraduates face in finding suitable accommodation before they start?
Problem
Short supply, high demand and rising costs. One-year stays, no UK guarantor, and a room secured remotely.
Opportunity
Not more housing: students who go into the market better equipped, so the room they find is the room they expected.
Scope
From applying to the course to the first weeks in the room. New building was out of scope.
Scaling the "how might we" question into a research plan
Before choosing methods, I broke the brief's question into parts and branched each one, so every method had a clear job and every finding traced back to the question. That tree wasn't fixed: it was built before fieldwork, grew as each method took us deeper into the cascading problems behind unsuitable accommodation, and, with the desk research, gave us the lens we coded our findings with.
How might we use [method] to [goal] for [who] in [where]?
| Part | How I branched it |
|---|---|
| Method | User research → behavioural or attitudinal → qualitative or quantitative → a named method (autoethnography, focus group, interviews) |
| Goal | Reduce barriers to suitable accommodation → facilities, location, price |
| Issue | Barrier types → PESTLE for the wider environment (desk research) → AEIOU for what we observed (fieldwork) |
| Who | International postgraduates → by nationality and language |
| Where | Bristol and the city → previous universities, and competing universities |
We ran six autoethnographies, interviews (including a Students' Union officer) and a focus group, then mapped the journey stage by stage. I wrote the focus-group question guide and facilitator guide, and built the AEIOU framework for the journey map.
| Stage | Methods |
|---|---|
| Planning | How-might-we scaling; research matrix (the 10 questions); opportunity solution tree; brainstorming with secondary research, to deepen the tree; PESTLE; AEIOU |
| Desk research | Secondary research (sector reports, literature); HESA data; competitor analysis (2026 follow-up) |
| Fieldwork | Autoethnography; semi-structured interviews (students and experts); focus group with a co-designed journey map |
| Combined by me | A journey-map focus group with PESTLE and AEIOU prompts placed on each stage, alongside stress ratings; an AEIOU analysis grid by journey stage and participant |
| Analysis | Thematic coding; affinity mapping; journey mapping; persona; prioritisation |
Six insights behind the peak
Committing sight unseen
"I didn't actually see the room and had to pay in full; everything is unknown."
Information at the wrong time
Students had websites, but not timely, trustworthy guidance before deciding.
Trust comes from peers
"There aren't any reviews based on real experience."
Financial rules exclude by default
No UK guarantor or credit history, and limits on paying rent upfront.
Time is a hidden barrier
"Uni didn't reply to my application, so I started finding a private flat."
Access needs can't be searched for
No listing showed desk size, room layout or window position, so students who need them can't check before signing.
Wei, 24, signing from abroad
Coming from China for a one-year MSc. She applies for university housing as soon as her offer is confirmed, then hears nothing. She joins WeChat groups and asks friends, then signs for a studio seen only in photos and pays the year upfront, with no UK guarantor. The room is fine, but further from campus than she thought.
She needs
- To know what to do, and when
- A way to check a room is real
- A route through guarantor rules
- Someone she trusts to ask
Which problem is biggest?
I re-analysed the team's data to rank problems by reach, stress and consequence, and added a confidence rating: how much evidence sits behind each row, not how serious it is.
| Problem | Reach | Avg stress | Consequence | Confidence |
|---|---|---|---|---|
| Time: late replies, deadlines, rushed choices | 5 of 8 | 7.4 | Refused university housing; signed in a hurry | Medium |
| Room or distance not as expected | 5 of 8 | 6.8 | Three rated moving in 9 out of 10 | Medium |
| Money: price, upfront payment, cost to family | 5 of 8 | 5.4 | Working full time to afford rent | Medium |
| Signing for a room unseen | 4 of 8 | 5.3 | Nervous commitment | Medium |
| Information hard to find or judge | 2 of 8 | 7.0 | Unsure where to live | Low |
| No trusted reviews | 2 of 8 | 2.5 | Relied on friends | Low |
| Access needs | 1 of 6 autoethnographies | n/a | Left the course | Very low |
Time leads on this data. Access has the worst consequence, but the focus group never asked about it, which is the story of the next section. Indicative: small samples.
How I told the client, and what I changed
While reporting the data, I noticed a finding missing from the discussion, one the focus group never covered. The access barrier came from my own autoethnography: I needed a room with a window and a suitable desk, couldn't find or check one easily, commuted instead, and it affected my health enough that I left the course. Two sources pointed at it, and our coding framework listed disability as a barrier, yet it never reached the final insights. When I noticed access was missing from the team's analysis, I used Claude to trace back through the sources and find where it dropped out.
What happened
- We built the coding framework by breaking the "how might we" statement down into each of its elements, before fieldwork. But we didn't carry it through consistently into the interviews, diaries and focus group, because no guide checked that every part of it was covered.
- So the focus group ran before the access finding was coded. The gap only showed later, when I noticed my own autoethnographic experience was missing from the rest of the team's understanding.
- Then, in synthesis after fieldwork, my case became a "hybrid commuting student" and the reason was dropped. The likely cause is a common AI-assisted coding error: the model obscures health details as private.
Why it matters
- It was the study's only documented case of an access barrier, and it had the worst consequence in the study: leaving the course due to a health condition worsened by unsuitable accommodation.
- It wasn't a lone signal. The autoethnography and the Students' Union officer interview both pointed at it, but it never made it into the guided discussions, so nobody else was asked.
- A finding that drops out quietly can't be weighed by the client. They'd act on an incomplete picture without knowing it.
The diagnosis: where the signal went, stage by stage
I traced the access finding through the study like a system, stage by stage. It entered the data through my autoethnography and the Students' Union interview, but no stage before or after carried it all the way through.
What I told the client
A client paying for research is paying to act on it, so I raised the gap early, rather than leaving them to find it later.
- Before they acted on the findings
As soon as I confirmed the gap, and before any sign-off or redesign started, I told them what was missed, how it was missed, and what it might change.
- I added it back, clearly marked as low evidence
I restored it to the insights with its sources (one autoethnography and one stakeholder interview) and a low confidence rating, and said what it does and doesn't prove: the consequence is serious, but we don't yet know how many students it affects. None of the other findings change.
- I offered a targeted follow-up at no extra cost
The miss was in our process, so the client shouldn't pay to fix it. The follow-up measures only this question: how many students need room details or adaptations before signing, how many get them, and what happens when they don't.
What I changed: how I run research now
Not a one-off fix: a standing rule for every study, so a finding can't fall between methods again.
- Synthesis between methods, before the next guide is fixed
Each method's data is coded before the next method's guide is finalised. No more running every method in the same week and analysing afterwards.
- Every guide is checked against the framework
Before fieldwork, each part of the "how might we" breakdown and coding framework is mapped to at least one question in every guide, and any gap is closed or written down as a deliberate choice.
- A manual search for sensitive themes across all data
Before synthesis, I search all raw data by hand for health, disability, money and safety, the themes most likely to be hidden by participants or dropped by tools.
- A human check of every AI-coded source against the original
AI speeds up coding, but each coded source is compared with its original before it counts. Anything the model left out is restored.
- Consented, anonymised codes instead of deletion
Sensitive details become anonymised codes, with the participant's consent, rather than being deleted as private. The fact that a barrier exists stays in the analysis, even when the personal detail doesn't.
The finding wasn't missing from the data. It was lost between methods. None of the other findings change.
Design recommendations
| Recommendation | Answers | What good looks like |
|---|---|---|
| A pre-arrival housing timeline | Time | What to do each month, and a fallback if there's no university room by a set date |
| Verified remote viewing | Sight unseen | Video or 360 tours and recent student reviews for listed properties |
| One clear guarantor page | Money | Guarantor services, deposits and rent rules, in plain English and the main languages |
| A peer channel by building | Trust | Incoming students can message current residents |
| Honest location details | Moving in | Walking and bus times to campus, nearest supermarket and hills on every listing |
| Searchable room details | Access | Desk size, room layout and window position on every listing, and a way to ask for adaptations before signing |
My recommendation on access: measure it before it causes damage nobody sees coming. Count how many students need room details or adaptations before signing, how many of those requests are met before arrival, and how many room moves, early tenancy ends and withdrawals follow an unsuitable room. Track two causes separately: a pre-existing health condition the room didn't suit, and health problems caused or made worse by poor living conditions, such as damp, mould or no daylight. Study now means hours on laptops and phones, so room details matter: a desk deep enough to keep screens at a comfortable distance, and a window, because the 20-20-20 rule for screen use means looking at something about 6 metres (20 feet) away for 20 seconds every 20 minutes. Universities lose fees, providers lose tenants and the money spent refitting rooms whose furniture doesn't suit, parents lose investments, and students lose the most.
I checked what the booking websites show
In November 2026 I compared five providers' booking pages against the six things students told us they needed. That follow-up is now its own case: 05b · What the booking websites show →
What I'd do with more time
0 · Ideate early
- Brainwriting on each "how might we"
- Grow the opportunity tree into solutions and tests
- SCAMPER on the methods themselves, to uncover new combinations before they go into the guides
- Prototype discovery inside the sessions: test rough concepts with participants, as in Fruityard's discovery service
- Concept cards as probes for the next phases
1 · Measure access
- Freedom of Information requests on adjustments, room moves and withdrawals
- Survey via the Students' Union
- Interviews with disability services and providers
2 · The journey over time
- Diary study from offer to arrival
- Photo study: the listing photo next to the real room
- Mystery shopping providers with access questions
3 · Design and test
- Card sort the room details students need
- Co-create one listing with students and providers
- Usability test, then A/B test it
How I use AI in research: it speeds up capture, coding and first-pass synthesis, and I check every AI-coded source against the original. This project is why: an AI step can quietly drop the finding that matters most.
A year on: reviewing it as a fresh case study
I'm only finishing this study now, so I reviewed it the way I'd review a client's: as a system, step by step, to find where the access finding fell through. The question tree, the guides and the coding frame all had a place for it. It was lost in the handovers between them: fieldwork and coding ran in the same week, my access prompts reached the focus-group map at only one stage, and the ethics plan steered student research away from health.
So I rebuilt the work as one opportunity tree, which joins the team's research matrix, my 2025 scaling of the question, and what I'd add now, with every item marked by where it came from. Two things stand out. The team's questions and mine covered mostly the same ground with different blind spots. And neither gave health and access a branch of its own.
For the University, the brief's client
- Restore the access finding
- Measure it with a free targeted follow-up
- Then build the pre-arrival timeline
For housing providers
- Co-design one listing with searchable room details
- Test it with students
For any team using this method
- Write the autoethnographies first and use them to test the guides
- Derive the coding from the tree
- Code between methods
- Add a senior check on top of peer review