User Research Methods: Pick the Right One for Your Question

Most people asking about user research start from the wrong question. The useful question is not "what are the research methods?" but "what don't I know right now that will cost me money if I build without it?" The method follows from that, and nothing else decides it.

If your question is "why do customers stop at the payment step?", you need an interview or a test session with five people, and you will have an answer within a week. If your question is "what percentage of them stop?", you need measurement data or a survey with dozens of responses, and neither will tell you the reason. Confusing the two wastes more research budget in small companies than anything else: five interviews get presented as a statistic, or a survey is asked to explain behaviour nobody watched.

This guide gives you a table that connects your question to a method, a sample size and the things that drive its cost; a sourced correction to the most widely repeated number in the field; and a way to turn what you hear in a session into an actual change in the design.

Start From the Question, Not the Method Name

The table below is ordered by the question you are trying to answer. Participant counts come from Nielsen Norman Group's published recommendations wherever one exists, and you will find the sources in the sample-size table further down. The cost column describes what drives the bill for a small team doing the work itself; it is not a published price.

Two notes before you choose. First, analytics do not work retroactively: reading behaviour requires instrumenting the site or app in advance, so if you never configured the events, last month's data does not exist no matter how hard you look. Read what to actually measure in Google Analytics 4 before you rely on that row. The same applies to A/B testing on a live store, covered in detail in ecommerce conversion rate optimisation.

Second, there is a gap between what people say and what they do, and the two often differ. Surveys and focus groups measure what people say; tests and analytics see what they do. That is exactly why Nielsen Norman Group describes focus groups as the least useful option for usability purposes: you hear a polished group opinion, not behaviour.

The "85% of Problems With Five Users" Number: Where It Came From and What It Requires

This number appears in almost every article on user research, in Arabic and English, and it was in the previous version of this page. It gets repeated in a form that makes it look like a general law. It is not one.

What the Original Paper Actually Said

The reference is Jakob Nielsen and Thomas Landauer's paper, "A Mathematical Model of the Finding of Usability Problems", published at INTERCHI'93. The paper did not say that five users find 85%. What it said is that problem discovery across 11 studies can be modelled as a Poisson process, and that the model is useful for estimating how many evaluations are worth their cost. In its medium example, it concluded that 16 evaluations would pay for themselves, with the highest benefit-to-cost ratio at four.

The 85% figure appeared seven years later, in Nielsen's 2000 article on the institute's site. There he plotted the curve of the formula N(1−(1−L)ⁿ), in which N stands for every problem the design contains and L for the share that a single user uncovers. He set L at 31%, which he describes as an average across a large number of projects, and at five users the curve reads about 85%. The number is a model output at an average value, not a field measurement.

What Happens at Either End of the Range

That is where the trouble is. In the CHI 2001 panel paper that Nielsen co-authored with Carl Turner and James Lewis, the average probability of discovering a given problem across those 11 studies is reported as 0.33, but ranging from 0.16 to 0.60. Apply the same formula at both ends and five users give you 58% in the worst case and 99% in the best. That is the distance between a team that was right to feel reassured and a team that left half its defects hidden.

Independent empirical work confirms it. In Laura Faulkner's 2003 study in Behavior Research Methods, 60 users were tested and random sets of five were drawn from them: some sets found 99% of the problems and others only 55%. At ten users the lowest result rose to 80%, and at twenty to 95%.

So the correct statement is this: in a qualitative usability test, on one interface with one comparable group of users, five participants uncover around 85% of the problems on average, and the actual result can fall to half that. Nielsen himself set the condition in the same article: the model applies only when your participants resemble one another and approach the product in much the same way. The number does not apply to interviews, to surveys, or to any numeric measurement.

There is a third warning worth noting: estimating that probability from a small sample tends toward optimism. Put differently, when you test with five and feel you covered everything, that feeling is itself a symptom of the small sample.

How Many Participants per Study Type

The practical rule that comes out of all this: three rounds of five beat one round of fifteen, because the goal is to improve the design, not to document its faults.

Zero-Budget Research: Start From the Data You Already Own

Most writing on user research assumes a budget and a team. But the owner of a store or a clinic in Cairo or Jeddah is sitting on a pile of data nobody has read, and it costs nothing:

• Customer service conversations on WhatsApp. Read the last 50 and pull out the recurring question; every question that repeats is information that should have been in the interface.
• What visitors type into your site's search box. The words they search for and do not find are a ready-made list of what is missing.
• Return reasons and order cancellation reasons. If "the size was wrong" tops the list, the problem is on the product page, not in shipping.
• The objections your sales rep hears every day, which are the closest thing you have to free, continuous field research.
• The clinic's booking call log. How many times did a patient call to ask about something that is on the website but they could not find?

Free Tools That Are Enough to Start

After your own data come the tools. Microsoft Clarity gives you heatmaps and session recordings, and its official page states that it is free with no traffic limits. Google says it offers Google Analytics tools free of charge, which is enough to measure the events you care about. Figma's Starter plan is free and covers building and testing an interactive prototype. The interview itself needs nothing more than a video call, screen sharing, and recorded consent from the participant.

On the other hand, do not assume every research tool has a free tier. Specialist card-sorting tools such as Optimal Workshop start at $199 per month billed annually with no free plan, as of September 2026. For a zero-budget card sort, use sticky notes in a room, or a shared FigJam file.

What a Zero Budget Does Not Buy

Honesty helps more than enthusiasm here. A zero budget gives you observations; it does not give you a representative sample. Everyone you reach for free is an existing customer satisfied enough to answer you, and you will not see the ones who left or the ones who never arrived. Use these results to generate hypotheses and rank priorities, not to publish numbers.

Local Reality: Who You Reach, and Who You Miss

Most recruitment guides are written for a market with ready-made participant panels. The picture here is different, and two numbers are worth stating.

In Egypt, internet users numbered 98.2 million in October 2025, or 82.7% of the population, which means 20.6 million people remained offline, and 56.5% of the population lives in rural areas. In Saudi Arabia, access is close to universal: 34.4 million users at 99%. That gap alone decides whether digital recruitment can represent your market.

Then comes the bias of the platform itself. According to the ad tools' own figures, Facebook's adult ad audience in Egypt was 60.1% male against 38.8% female, and TikTok's 64.2% against 35.8%. If you recruit participants through an ad on either platform, you inherit that skew in full before you ask your first question.

Recruitment Channels That Work Here

Your existing customer list beats any ad: a personal WhatsApp message from the number they recognise gets read and answered. Next comes the branch or the shop itself, where ten minutes with a customer standing at the till is worth a whole survey. A sales rep can arrange an interview with an angry customer, who is the most valuable participant you have. And when the segment is a specific profession, such as pharmacists, teachers or workshop owners, professional WhatsApp and Telegram groups are faster than any paid panel.

Always make it clear that you are testing the design and not the person, and that the session is not a sales call. If you offer an incentive, make it payment for time rather than for a positive opinion, and say so at the start of the session.

Dialect or Standard Arabic?

Write the question guide in Modern Standard Arabic so every facilitator asks the same thing, then ask in dialect. A question phrased in formal Arabic invites a formal, polished answer, and the participant shifts automatically into survey-answering mode instead of telling you what happened to them.

Record the answer in dialect, exactly as it was said. When a customer says "that button isn't visible", do not convert it in the report into "weak visual hierarchy in the element": the second phrasing hides the information you will need later. In the Gulf, mind the differences between dialects; a facilitator whose accent is unfamiliar to the participant adds a barrier for no reason.

One last point belongs to culture more than to language: participants are people who want to be helpful and to please the person in front of them, and holding back direct criticism in front of the business owner is expected behaviour here. Handle it two ways: have someone other than the decision-maker run the session, and ask about the past rather than about opinion. "Walk me through the last time you ordered" is more useful than "is the app easy?".

From Observation to Design Decision

This step is what separates useful research from research done for show. Research that ends in a handsome slide deck and changes not one line of the design was a waste of time.

A Decision Log Instead of a Report

After each round, turn every observation into a single line in this shape:

The "what would prove us wrong" column matters most, because it stops every change from becoming a success in hindsight.

A hypothetical example: a clothing store in Mansoura ran five sessions on an interactive prototype. Three participants stalled at the "governorate" field because its long list opens in an unexpected order, and a fourth closed the form entirely when asked to register an account before paying. The decision: turn the governorate field into a text search, and allow guest checkout. The metric: completion rate of the address step. This is an illustrative example, not an actual project.

Frequency Is Not Severity

A counterintuitive point: there is no correlation between how severe a problem is and how likely it is to be discovered, a conclusion reported in the CHI 2001 panel paper cited above, drawing on James Lewis's work. It means a problem seen by one participant may be more dangerous than one seen by all five.

So rank by two criteria together: the severity of the impact on the user, and how many people will actually hit it according to your measurement data. When you are torn between two observations, compare what you heard in the session with what analytics and support logs say; an observation confirmed by three different sources deserves immediate action. And if what you came out with is a list of features rather than a list of fixes, you are facing a scope question rather than a design question, and its place is the minimum viable product guide.

Four Mistakes That Drain Research of Its Value

Leading Questions

Their tell is that they begin with "did", "was" or "is". "Was that step easy?" invites agreement, while "how did you find that step?" leaves the participant free to answer. The rule Nielsen Norman Group recommends is to start with "how" and "what", and to keep closed questions for clarification only.

Researching to Confirm Rather Than to Discover

If you walk into the session with a hypothesis you love, you will hear what confirms it. The methodical fix is simple: gather data early, before you invest months in a design you will then defend, and have a colleague who was not part of the design review the question plan and sit in on the findings.

The Wrong Sample, Not Just the Small One

Five participants from one coherent group beat twenty mixed ones. The more segments you have and the more distinct they are, the less each participant uncovers and the larger the numbers you need. Ask yourself before recruiting: do these five use the product the same way? If the answer is no, you are running five small studies, not one.

A Single Observer

When researchers reviewed 11 studies on the evaluator effect, they found that agreement between any two people evaluating the same system ranged from 5% to 65%. Two people watching the same session come out with worryingly different problem lists. Record the sessions, have two people review them, and compare the two lists before you decide.

Consistency after the decision is a separate matter: scattered fixes on scattered screens produce an incoherent interface, which is what a design system addresses. For the general principles the interface itself is built on, see the UX/UI design principles guide.

If you are planning a digital product or reworking an existing interface and want a discovery phase built on real research, send your idea to the Snaabble team through our UI/UX design services and get an initial plan with a time and cost estimate within 24 hours, free and with no obligation.

Sources

• The modelling paper the whole debate rests on: A Mathematical Model of the Finding of Usability Problems, by Jakob Nielsen and Thomas Landauer, ACM INTERCHI'93, pp. 206–213 (1993).
• The article that published the 85% figure: Why You Only Need to Test with 5 Users, Jakob Nielsen, Nielsen Norman Group (2000).
• A critical review of sample size, co-authored by Nielsen with Carl Turner and James Lewis: Current Issues in the Determination of Usability Test Sample Size, a CHI 2001 panel paper.
• The independent empirical check: Beyond the Five-User Assumption, Laura Faulkner, Behavior Research Methods, vol. 35, pp. 379–383 (2003).
• Nielsen Norman Group pages cited here: choosing a method, how many users in a qualitative test, participants for quantitative studies, card sorting, diary studies, open-ended and closed questions, and confirmation bias.
• Market figures: DataReportal's 2026 reports for Egypt and Saudi Arabia.

All links reviewed on 22 September 2026.