◀ Course contents Part 2 · Module 2-08

The 5-Step User Research Process

A repeatable recipe for learning from users, without a research degree

Good research isn't about fancy labs; it's about asking a sharp question, putting it in front of the right people the right way, and turning what you hear into decisions. Five steps, every time.

Ready?

1

Define the Objective

Before choosing a method, before recruiting anyone, there is one question that decides whether a study is worth running: what will we do differently depending on the answer?

This sounds obvious and is routinely skipped, because research is culturally virtuous — proposing it looks diligent and objecting to it looks incurious. So studies get run to confirm decisions already made, and the findings arrive to polite nods and no consequences.

A well-formed objective has three parts: the decision it will inform, the question that would move that decision, and the threshold at which you would act differently. “We are deciding whether to build the import wizard. We need to know how people currently get data in and how painful it is. If most already have a workable path, we deprioritise it.”

That third clause is the one people leave off, and it does the same work as a threshold in a hypothesis. Writing down in advance what would change your mind is what stops research becoming an exercise in finding support for what you already wanted.

The shape of a research study Five steps left to right: define the objective, choose a method, recruit participants, conduct the sessions without contaminating them, then synthesise findings. 1 Define the decision it serves 2 Choose follows the question 3 Recruit the right people 4 Conduct without leading 5 Synthesise patterns, not quotes
Five steps, and the ordering is the entire content of the drawing. The method is panel two, and it is downstream of panel one — a team that opens by booking interviews has settled the shape of its answer before writing down the question, and will finish the study holding a folder of transcripts and no decision.

Everyday example, texting a friend for advice

When a friend texts "help me decide," the useful first question is: "what will you actually do differently depending on my answer?" If they've already booked the flights, your opinion on the destination is just chit-chat. Research is the same: before you spend anyone's time, know the real decision waiting on the other end. "Understand our users" leads to a nice chat and no change. "Find out why trials die in week one, so we know whether to fix onboarding or the price" has a decision attached, so every finding will actually land somewhere.

The litmus test

Ask: "What will we decide differently depending on the answer?" If nothing changes either way, the research is entertainment.

Weak: "Understand our dashboard users." Strong: "Find out why weekly dashboard usage is flat, to decide between a redesign and better onboarding."

The study that would have changed nothing

A team proposed eight interviews on a feature already committed, designed and scheduled. Asked what they would decide differently depending on the answer, nobody could say. The study was cancelled and the same effort went to a question that was genuinely open.

Quick check

Which of these is a strong research objective?

2

Choose the Method

Methods are not better or worse. They answer different questions, and the most common mistake is choosing one before knowing which question you have.

The first split is what people say versus what people do. Interviews and surveys give you the first. Observation, usability tests and analytics give you the second. When they disagree — and they frequently do — behaviour is the more reliable witness, because people explain their own actions with reasons rather than causes.

The second split is exploratory versus evaluative. Exploratory research asks what problems exist at all; you use it when the question is open. Evaluative research asks whether a specific thing works; you use it when you have something to put in front of someone.

A rough guide: to learn why, talk to people; to learn how many, count them; to learn whether this design works, watch someone try it. Teams that start by picking a method are choosing an answer shape before they have a question, which is how you end up with sixteen interviews nobody can act on.

Choosing the method from the question A chooser matching questions to methods: why needs conversation, how many needs counting, and whether a design works needs watching someone attempt a task. What do you need to know? why do they do this? Talk to people — interviews, contextual enquiry how common is it? Count them — analytics, survey does this design work? Watch someone try it — usability session
One team asked why people abandoned a flow and ran a survey. The answers came back articulate, plausible and reconstructed after the fact; five recordings later showed a validation error nobody had mentioned. They took the middle branch to answer a question on the left — the survey counted something, accurately, and it was not the thing they needed.

Everyday example, the right tool for the job

You wouldn't use a hammer to tighten a screw. Research methods are the same, each fits a different kind of question. Want to know how many people prefer tea over coffee? Poll a big crowd (a survey). Want to know why your friend quit coffee? Sit down and talk (an interview). Want to know if someone can actually assemble your flat-pack chair? Hand them the box and watch (a usability test). Want to know which of two menus sells more? Try both and compare sales (an A/B test). Pick the tool that matches the question, not the one you're most comfortable holding.

Why? How?

Interviews & field studies

Understand motivations, workflows, and context. Small numbers, deep answers.

Can they use it?

Usability tests

Watch real people attempt real tasks. About five users finds most issues.

How many?

Surveys & analytics

Measure scale and frequency across the whole audience.

Which is better?

A/B & fake-door tests

Compare options with behavioral evidence, not opinions.

The method that could not answer the question

A team wanted to know why people abandoned a flow and ran a survey. The responses were articulate, plausible, and reconstructed after the fact — people explain their own behaviour with reasons rather than causes. Five session recordings showed a validation error that nobody had mentioned, because nobody had noticed it consciously.

Quick check

Your question: "Why do restaurant owners still manage bookings on paper instead of using apps like ours?" Best method?

3

Recruit the Right People

Who you talk to determines what you can conclude, and recruiting is where more studies are quietly invalidated than at any other step.

So work backwards from the decision. If you are deciding about onboarding, you need people who onboarded recently, including the ones who gave up. If you are deciding about a segment you do not yet serve, your existing customers are the wrong population by definition.

Write screening criteria before you start and hold to them, because the temptation to accept a willing near-match is strongest when recruiting is slow. Five of the right people beat twenty of the wrong ones, and the twenty will feel more convincing, which is the danger.

Two biases are worth naming. Availability: the people who answer are unusually engaged. Incentive: pay too much and you attract professional participants; pay nothing and you attract only the delighted and the furious. The quiet majority in the middle is the hardest group to reach and usually the one the decision is actually about.

Who you talk to decides what you can conclude Two panels. Recruiting whoever is easiest to reach produces a sample defined by the thing being studied; recruiting backwards from the decision reaches the people who can actually answer. Easiest to reach the sample is the bias In-app message → engaged users Studying churn, asking the retained The quiet majority never answers Twenty of the wrong people Backwards from the decision harder, and valid Recruit those the decision affects Include the ones who gave up Screen before you start Five of the right people
One study recruited from an in-app message and heard from satisfied users. The question was about churn — everyone who could have answered had already left and could not see the message. The sample was defined by the very thing being studied.

Everyday example, taste-testing kids' cereal

If you're designing a breakfast cereal for children, you test it on kids, not on their parents, a parent guessing what a 7-year-old likes is worthless data. Same logic everywhere: to learn why people quit your gym, talk to ex-members, not the regulars still showing up happily. The easiest people to grab (your friends, your teammates, your biggest fans) are almost never the right ones, and their politeness and insider knowledge quietly poison the answers. Fish where the fish are: recruit the exact people who did the thing you're studying.

1

Screen by behavior, not just demographics

Studying churn? Recruit people who actually canceled. Studying invoicing? Recruit people who send invoices, not people who say they might.

2

Beware convenience samples

Teammates, friends, and your biggest fans are the easiest to book and the least representative. Their answers are polluted by politeness and expertise.

3

Small incentives work

A modest gift card recruits honest strangers. It costs less than one wrong feature.

Recruiting the people who were easy to reach

A study recruited from the in-app message and heard from engaged, satisfied users. The question was about churn. Everyone who could have answered it had already left and could not see the message — the sample was defined by the thing being studied.

Quick check

You're researching why users cancel subscriptions. Who do you recruit?

4

Conduct Without Contaminating

In a research session, the largest source of error is the person running it. Everything you say shapes what you hear, and most contamination is well-intentioned.

The dominant rule is to ask about the past, not the future. “Would you use this?” produces a polite yes that predicts nothing. “Walk me through the last time you had to do this” produces a story with details in it, and details can be checked.

Then be quiet. Silence is the most productive instrument in the room and the hardest to use, because a pause of four seconds feels to the moderator like a failure and to the participant like an invitation. Most of the useful material arrives after the point where you wanted to fill the gap.

And do not defend the design. When someone struggles, the instinct is to explain how it works — which ends the observation and converts the session into a demo. The struggle is the finding, and explaining it away is how a team leaves a session feeling reassured and knowing less than when they walked in.

Asking without contaminating A checklist of moderating habits: ask about the past rather than the future, avoid leading questions, stay silent, and never explain the design when someone struggles. Ask this Contaminates "Walk me through the last time…" "What were you expecting to happen?" Silence after a question Let the struggle happen "Would you use this?" "Was that confusing?" Filling every pause Explaining how it works
One moderator asked “was that confusing?” after every task and reported eleven confusing steps; a second asked what people expected and found three. Every line in the right-hand column is the moderator being helpful, which is exactly what makes them hard to stop doing — and why a contaminated session comes back agreeing with whoever ran it.

Everyday example, good reporter vs. bad reporter

A bad reporter asks, "You must have been furious about that, right?", and gets back whatever they suggested. A good reporter asks, "What happened next?" and then goes quiet, letting the person fill the silence with the real story. In research you want to be the good reporter: ask about what actually happened ("tell me about the last time you…"), keep questions open, and resist the urge to jump in. The moment you hint at the answer you're hoping for, polite people hand it right back to you, and your data is worthless.

1

Past behavior over future intent

"Walk me through the last time you…" beats "would you ever…?" People are historians of their own lives and terrible fortune-tellers.

2

Open questions, then silence

"What happened next?", then wait. Silence is uncomfortable; users fill it with the good stuff.

3

Never pitch, never lead

"Don't you think this is easier?" teaches participants the answer you want. In usability tests: give the task, then stay quiet while they struggle.

Leading without meaning to

A moderator asked "was that confusing?" after every task. Participants agreed each time, and the report listed eleven confusing steps. A second study asked "what were you expecting to happen?" and found three. The first study had measured the question, not the product.

Quick check

Which interview question will produce the most trustworthy answer?

5

Synthesize and Share

Raw notes are not findings, and findings are not insight. Synthesis is the work of climbing that ladder, and it is where most studies stop one rung too early.

The ladder looks like this. Data: “40% of users abandon checkout at the payment step.” Finding: “most of those abandonments follow a validation error on the card field.” Insight: “people do not realise the card was rejected for a formatting reason, so they assume the payment failed and stop trying.” Only the third sentence tells anyone what to build.

Getting there is mechanical more than inspired. Put every observation on its own note, cluster them by similarity without naming the clusters, then name each cluster once it has formed. Then look for the tension inside each — the place where what people want and what they do disagree. That tension is where the insight is.

Twelve interviews once produced ninety quotes and a summary nobody acted on. Clustered, the same material showed one tension repeated in eleven of the twelve: people wanted the tool to make a decision for them and did not trust it to. That sentence changed the roadmap; the ninety quotes had not.

From notes to something that survives Four synthesis steps: one observation per note, cluster before naming, name the tension inside each cluster, then write findings a roadmap could contradict. 1 One note each observations, not opinions 2 Cluster first name them after they form 3 Find the tension where want and do diverge 4 Write to be contradicted three sentences, not thirty
Steps one and two are clerical, and step three is the only place judgement enters. Step four is a constraint, not a note about formatting — a finding written so that nothing could contradict it cannot be confirmed either, and it will be nodded at once and never cited again.

Everyday example, the chef, not the grocery bag

A chef doesn't dump a bag of raw ingredients on your table, they cook and plate a dish you can actually enjoy. Handing stakeholders raw transcripts is dumping the grocery bag: you're making them do the cooking (and they won't). Your job is to synthesize the messy sessions into a few complete, well-plated insights, and to show the evidence, a 30-second clip of a real user getting stuck beats a 30-page report every time. Bring the dish, cooked and served, into the room where the decision gets made.

1

Cluster into themes

Affinity-map the observations and let patterns emerge (the full technique lives in Driving User Insights).

2

Write insight statements

[Observation], because [reason], which means [implication], each tied back to the original research objective.

3

Share with evidence attached

A 30-second clip of a real user failing beats a 30-page report. Bring quotes and clips into the room where the decision happens.

Findings that outlived the deck

A study was presented once, applauded, and never referenced again. The next study ended differently: three findings, each written as a sentence a roadmap could contradict, pinned in the team channel. Two were still being argued with a year later, which is the only evidence that research was used.

Quick check

You finished eight interviews and the findings challenge the roadmap. What's the most effective way to share them?

Drill what you learned
Scenario 1 easy

Your manager says: 'Go do some user research on our new dashboard.' Nothing more specific.

What's your first move?

Scenario 2 easy

You have one week and almost no budget to find out whether people can complete checkout in your new app.

What do you do?

Scenario 3 easy

A teammate proposes validating a new feature idea by surveying your most engaged power users: 'They know the product best.'

What's the problem?

Scenario 4 easy

Your question is "How many of our users would use a dark mode, and how strongly do they want it?" A researcher plans five in-depth interviews.

Is the method a good fit?

Scenario 5 easy

A usability test is scheduled with 50 participants to "be statistically valid," which will take a month to run and analyze.

What's the more efficient approach?

Scenario 6 medium

To study why users churn, a PM recruits from the list of users who are still actively subscribed, because "they're the easiest to contact through the app."

What's the recruiting error?

Scenario 7 medium

In a usability test, the moderator says before each task: "This next part is really easy and intuitive, you'll love it."

Why does this ruin the test?

Scenario 8 medium

A stakeholder wants to run research to "prove that users want the feature we've already decided to build."

What's wrong with this framing?

Scenario 9 medium

During an interview, a user goes quiet for eight seconds after you ask about their last frustrating experience. The interviewer feels awkward and jumps in with "was it the loading speed?"

What did the interviewer do wrong?

Scenario 10 medium

You need to decide between two onboarding designs. You have live traffic and can measure which one leads to more completed setups.

Which method best answers 'which is better?'

Scenario 11 hard

Leadership wants "big, statistically-significant research" before any decision, but you have a small user base and a one-week deadline for a design choice.

What's the pragmatic, honest approach?

Scenario 12 hard

A researcher recruits participants for a study on a professional accounting tool by posting in a general "make money online" forum, because it was quick and free.

What's the risk?

Scenario 13 hard

After great interviews, a PM emails stakeholders a 40-page report with every transcript. Two weeks later, the roadmap decision ignored all of it.

What's the synthesis-and-share failure?

Scenario 14 hard

You want to understand the real workflow of warehouse pickers, but you plan to interview them over video calls during their lunch break.

What might a better method capture that interviews miss?

Scenario 15 hard

A PM insists that talking to just five users can never be trusted and that no research is worth doing unless hundreds participate.

What's the balanced, correct view?

Drilled it. Now apply it to a real situation.

Put it to work
From lesson 1

Intuit went home with people

Intuit

Intuit's "Follow Me Home" programme, run from the earliest days of Quicken, sent employees to watch customers use the product where they actually used it — at a kitchen table, with their real bank statements, in their own mess.

Scott Cook has described the practice as foundational. The objective was never "get feedback"; it was to see what happens in the real setting, which no lab and no survey reproduces.

What did the objective being that specific make possible?

From lesson 2

Four questions, four methods

A parcel-locker network

Four questions on the research plan. Using one method for all four is the most common way a research budget gets wasted.

Match each question to its method, in the order the questions are listed.
1. Why do people abandon at the locker rather than completing collection? 2. Which of two door designs is faster to open? 3. How many customers own a car? 4. Is there an unmet need around returns we have not seen?

Drag the rows, or use the arrows, then check.

  1. Contextual observation at lockers Question 1. Abandonment happens in a place, in weather, in a queue, often holding something. Only being there reveals why, and people cannot reliably report it afterwards.
  2. Usability test with a prototype and a stopwatch Question 2. A specific, comparative, measurable question about an interaction. Small n, controlled task, complete metric.
  3. Survey to a representative sample Question 3. A factual question about incidence across a population. This is the one thing surveys are genuinely good at.
  4. Open-ended discovery interviews Question 4. You are looking for something you cannot yet name, so the method has to allow the participant to take you somewhere you did not plan to go.
From lesson 3

The recruit that decided the answer

A hospital discharge app

Research question: why do patients miss follow-up appointments? Two recruitment plans are proposed.

The two plans
Plan APlan B
SourceApp users who opted into researchAppointment records, all patients
ScreeningWilling and available for a 45-min callMissed at least one follow-up in 6 months
Sample12 people12 people
Incentive£30 voucher£30 voucher, plus offer of a home visit

What is wrong with Plan A for this question?

Select all that apply — there are 3 to find.

From lesson 4

Four moments in one session

A children's reading app

Five parents are testing a new reading-progress screen. Four moments from the transcripts. Two of them are the researcher damaging their own data.

From the session transcripts
#Moment
1"So this shows your child's progress — pretty clear, right?"
2Participant pauses for 20 seconds; researcher says nothing
3"What were you expecting to happen when you tapped that?"
4"That's a bug, ignore it — imagine it worked and carry on"

Which moments are contaminating the session?

Select all that apply — there are 2 to find.

From lesson 5

A readout nobody acted on

An insurance claims portal

Six weeks of research, eighteen participants, a 60-slide deck delivered to a room of fourteen people. Three months later, nothing in the product has changed and two of the findings have been independently rediscovered by another team.

Work out why good research produced no change.

  1. Step 1 of 3

    The deck was organised by participant — one section per person. What does that cost?

    Restructured by finding, the same material was eleven slides with the participant quotes as supporting evidence.

  2. Step 2 of 3

    No finding named an owner or a decision. What does that produce?

    Two findings got owners and dates. Both shipped within the quarter.

  3. Step 3 of 3

    Two findings were rediscovered by another team. What does that indicate?

    Findings moved to a searchable index, one page each, tagged by product area.

Notification