How to do UX research on a budget: five methods, their limits

If you want to know how to do UX research without a lab, start with one question, pick the simplest method that can actually answer it, and write down how that method could be wrong before you trust it. Five methods cover most small-budget work: short interviews, pop-up sessions, surveys, free behaviour analytics and remote card sorts. Each fails in a predictable way, and a one-page report that names the failure is worth more than a long deck that hides it.

If you are coming to UX from visual work, the earlier note, From graphic design to UX: a staged path, covers what carries over. This one is about the research itself.

Say what you want to know, then choose

Nielsen Norman Group sorts research methods along three lines: what people say against what they do, qualitative against quantitative, and how far the participant is using the real product ¹. The split matters because attitudinal methods are "limited by what people are aware of and willing to report" ¹. Qualitative methods tell you why and how to fix; quantitative ones tell you how many and how much ¹.

So write the question first. "Why do people abandon the KYC step?" is a behaviour question. "Which of these two labels do people expect?" is a mental-model question. They need different methods.

How to do UX research with five simple methods

Method	Good for	Where it misleads
Interviews	Needs, pain points, how people think about a task	Reported, not observed, behaviour; memory gaps
Pop-up sessions (about 10 min each)	One or two tasks, hard-to-reach people	Short sessions, random participants
Surveys	Counting how common an attitude is	Self-report; wording steers answers
Behaviour analytics	Where people click, scroll and drop off	Shows what happened, not why
Remote card sort	How people group your content	No context, one level of categories

Interviews

Interviews are for learning who users are, what they struggle with and how they think about a task ². They are not a usability test: participants usually don't touch a design ². NN/g lists how they go wrong: people report behaviour rather than show it, "human memory is flawed", people leave out details they think don't matter, some present themselves in a good light, and leading questions compromise the data ².

When it's wrong: when you use interviews to decide whether a screen works. Ask about the last time they did the task, not what they would do in future.

Pop-up (guerrilla) sessions

The UK Government Digital Service's manual calls this pop-up research. You recruit people at a venue on the day, aim for sessions of around 10 minutes, focus on one or two tasks, and always work in pairs ³. You get informed consent from each person first, and the manual advises against offering money ³. The guide dates from February 2017, but it is still the government's published guidance.

When it's wrong: the manual says short sessions, random participants and recording problems "mean you'll always need to combine it with other research methods" ³. People in a café are not your users unless your users sit in that café.

Surveys

A survey measures attitudes at scale, usually with closed questions ¹. That is its strength and its limit: it counts what people say, not what they do ¹. NN/g's advice is to write neutral questions, avoid loaded words such as "essential" or "useful", define your terms, and pilot the survey with a few people thinking aloud before you send it ⁴. That article is from 2016; the advice has not aged.

When it's wrong: when you treat open answers from whoever replied as the view of all your users. NN/g notes such results are "rarely representative for the whole target audience" ⁴.

Free behaviour analytics

If the product is live, you can watch behaviour without recruiting anyone. Microsoft Clarity offers session recordings, heatmaps and funnels, and Microsoft's documentation calls it "a free service forever" ⁵. The same page says it isn't meant for sites or apps aimed at minors, and points to its masking and consent controls ⁵. Any analytics needs the site instrumented properly in advance ¹.

When it's wrong: analytics shows what happened, not why ¹. Clicks piling up on something that isn't a link tell you there is a problem. An interview or a test tells you what people expected.

Remote card sorting

Card sorting shows how people group your content, which helps you build navigation that matches their mental models ⁶. NN/g's 2024 guidance is at least 15 participants for a qualitative sort and 30 to 50 for a quantitative one, with about 30 to 50 cards to avoid fatigue ⁶. Remote, unmoderated tools run the study without a facilitator ⁶.

When it's wrong: a card sort has no context, shows only one level of categories, and past 50 cards people start dumping items into "Miscellaneous" ⁶. To check a structure you already have, NN/g points you to tree testing instead ⁶.

Read other people's research the same way

Most of what you read about UX research is secondary: someone else's survey, summarised. Read it as a method, not a headline.

Take Maze's Future of User Research Report 2025. It reports that 58% of respondents use AI tools and calls that "a 32% increase" on its 2024 report ⁸. Before you quote it, check four things:

• Who asked? Maze sells a user research platform, so it has an interest in research demand looking strong.
• Who answered? 800 product professionals, 47% researchers and 38% designers, surveyed over December 2024 and January 2025 ⁸.
• Where? 36% North America and 33% Europe ⁸. That says little about a product team in Pune.
• What does the number mean? "A 32% increase" doesn't say whether it is relative or in percentage points ⁸, and the page carries no publication date.

The figure may be fine. Just say what it is: a vendor's survey of its own field, mostly from two regions.

Report what you found on one page

The GOV.UK manual separates observations from findings. Observations are what you saw or heard, written exactly, "not what they think it means"; findings are short statements written once observations are grouped into themes ⁷. It suggests about one hour of analysis for every two hours of research, done while it is fresh, with more than one person to reduce bias ⁷. That guide is from 2016.

A one-page report that holds up:

• The question, in one sentence.
• Method and who: how many people, how you found them, when.
• What we saw: five to eight observations, quotes or behaviour, verbatim.
• What we think it means: three findings, each linked to its observations.
• How this could be wrong: the method's failure mode, from the table above.
• What we'll do next: design changes, and the question for the next round.

Keep this page. It is the first real piece of your portfolio, and it shows a hiring manager how you think, which a polished screen doesn't. The user research module of TTFA's UI/UX Designer program works through these methods, with personas and user flows, on project briefs.

What to do with this

Pick one product you use at work, or a public service site. Write one question. Run five short interviews or one pop-up afternoon, and write the one-page report the same week. Then add the line most people skip: how this method could have misled you.

What we don't know: there is no published, method-stated data on how Indian product teams run research on small budgets. Treat global surveys as context, not as a picture of your market.

The next note in the series is about usability testing with five people: how to run a session, what to write down while it happens, and how to decide what to change.

Frequently asked questions

Can I do UX research without users to recruit?

You can start with analytics on a live product and with published research, but both tell you what happened, not why. Pop-up sessions at a relevant venue are the quickest way to reach real people.

How many people do I need for interviews?

There is no single right number. Stop when new sessions stop telling you new things, and state your number in the report so readers can judge.

Is a survey enough on its own?

Rarely. A survey counts what people say, and answers from whoever replied don't represent all users. Pair it with something behavioural, such as analytics or a test.

Do I need consent to record a session?

Yes. Get informed consent from every participant before you start, as the GOV.UK manual asks for pop-up research. Tell them what you will record, why, and who will see it.