Design
UX Research Without a Lab Budget: What Actually Produces Results
Most teams do not need a research department, they need to stop guessing. Methods ranked by cost and payoff, the five-user rule and its important caveat, and how to turn findings into decisions without a sixty-page report nobody reads.
In short
The cheapest useful research costs nothing: four to eight hours reading session replays and funnels in your existing analytics. Next in value is a five-user usability test, which surfaces roughly 85 percent of the problems in a single segment and costs 30,000 to 120,000 RUB. Depth interviews with 8 to 12 people matter when the question is not where the flow breaks but why people do not buy.
You do not need a research department, you need to stop guessing
The conversation about UX research almost always starts from the wrong end. A company reads about ethnographic field visits, eye-tracking labs and respondent panels, prices it at several hundred thousand roubles and concludes that research is for big players. Then it goes back to deciding interface questions in meetings where the most senior opinion wins. That, not the missing lab, is what costs money.
Useful research starts an order of magnitude earlier and cheaper. Level one is the data you already collect: session recordings, funnels, form analytics, on-site search queries. That is zero roubles and four to eight hours of careful attention. Level two is showing the interface to five real people and watching quietly while they use it. That is half a working day and no tooling beyond a screen and a recorder. Those two levels already surface most of what there is to find.
The core difference between research and opinion is one thing: research answers a question defined in advance. "Let us see what users say" is not research, it is a chat. "Can a new customer complete a regional delivery order without contacting support once" is a question you can answer in one day with five people. Writing the question takes an hour and determines whether everything that follows is worth anything.
- Read the data you already have first. Commissioning research before that means paying for what is already in your analytics.
- Write one testable question. Without it, any method degrades into opinion collection.
- Watch behaviour, not statements. People predict their actions badly and demonstrate them well.
Methods ranked by cost versus insight
Every method has a question it answers well and a question it answers confidently and wrongly. The second is the dangerous one: a report with numbers looks equally convincing whether or not it measured anything meaningful. The table below is a working cheat sheet. Pick the row that matches your question, and read the "when it lies" column before you act on the results.
Note the order. The first two rows are nearly free and nearly always skipped, because they do not look like research. A five-user usability test costs a fraction of depth interviews and answers a different question: interviews explain motivation and the language customers use, a test shows where somebody gets stuck. Confusing the two is the most common and most expensive mistake. A team commissions interviews to fix a checkout and receives 40 pages about audience values.
The costs are market reference points for Russian agencies and research teams in 2026. In house, almost any of these is considerably cheaper, with one caveat: somebody who works on the product daily is far worse at not helping the participant. If the person running the session wrote the interface, expect some findings to be unconsciously softened.
| Method | Question it answers | Cost | When it lies to you |
|---|---|---|---|
| Analytics and session replays | Where exactly users drop off | 0 RUB, 4-8 hours of work | Shows where, never why |
| Hallway testing | Is the screen understandable at a glance | 0 RUB, 2-3 hours | Participants are not your audience and have no real goal |
| Usability test, 5 users | Can a person complete the task | 30,000-120,000 RUB | Five is too few when there are several segments |
| Depth interviews, 8-12 people | Why people buy and what words they use | 80,000-250,000 RUB | People report intentions, not behaviour |
| Quantitative survey | How widespread a known problem is | 10,000-60,000 RUB | Wording drives the answer, loyal users overrespond |
| A/B test | Which variant makes more money | Cost of building the variants | On low traffic it reports noise as a result |
| Heuristic UX audit | A fast list of obvious defects | 40,000-150,000 RUB | Expert opinion is not user behaviour |
| Customer journey map | Where the experience breaks between channels | 100,000-300,000 RUB | Becomes a poster with no owner and no tasks |
Why five users is enough, and when it is not
The five-user rule is industry canon, traceable to the work of the Nielsen Norman Group. The logic is that usability problems are unevenly distributed: the most frequent defects hit nearly every participant, so the first few people surface them immediately. The fifth participant adds little: together five uncover roughly 80 to 85 percent of the problems, while the next five add a few percent for double the cost.
The caveat people skip most often: the rule holds inside a single homogeneous segment. If your product serves a purchasing manager and an end user, a novice and an expert, a desktop office worker and a courier on a phone, that is not one audience but three or four. Then you need three to four people per segment and the total grows to 10 to 15. Five people averaged across everyone produce a mixture of incomparable observations that supports no conclusion about any segment.
The second caveat concerns the kind of conclusion you can draw. Five people find problems, they measure nothing. A sentence like "conversion will rise 12 percent because three out of five missed the button" is invention: that sample supports neither a proportion nor an effect size. If you need numbers, you need a different instrument: an A/B test, funnel analytics or a properly sized survey. Mixing qualitative findings with quantitative claims is how teams reach confidently wrong decisions.
The practical takeaway: do not plan one study with 15 people, plan three waves of five with fixes in between. Wave one finds blocking defects, you fix them, wave two finds the next layer that was hidden behind the first, wave three verifies the result. The total cost is comparable and the value is categorically different, because you test your own fixes twice instead of collecting a long list, half of which expires during a month of development.
Writing tasks and questions that do not lead the participant
The main risk in cheap research is not the small sample, it is the prompt. One badly chosen word turns an interface test into a compliance test, and the result still looks completely normal. The base rule: a task describes the goal in the participant own words and never names interface elements. Bad: "Find the Checkout button at the bottom of the card and click it." Good: "You need this item delivered by Friday. Do what you would do at home."
The same applies to questions. "Is the filter convenient to use?" is a question a polite person answers yes to about 80 percent of the time. "Tell me about the last time you looked for something similar. What happened next?" asks about past behaviour, which is hard to falsify without noticing. The general principle: ask about what already happened, not about what somebody would do. Respondents describe hypothetical futures optimistically and almost always inaccurately.
Three techniques work during a session. First, before a click ask "what do you expect to see here". The gap between expectation and reality is the finding. Second, hold the silence. Thirty seconds of a participant searching is unbearable for the moderator and invaluable for the result, and rescuing them erases the exact moment you came for. Third, when asked "how is this supposed to work", answer "how does it look to you" and stay quiet.
One last point on recruiting: a participant has to be somebody who genuinely has your problem. Testing a building materials store checkout on designers from the next room is pointless, because they do not know what crane delivery is and will not notice the option is missing. Five suitable people are easier to find than it seems: customers who recently contacted support, lapsed leads, members of relevant communities. An incentive of 1,500 to 3,000 RUB for 40 minutes covers most situations.
- A task describes the goal, not the route. Naming buttons in the task is a hint.
- Ask about past behaviour, never about a hypothetical future.
- Hold a 30-second pause. Helping the participant destroys the finding.
- The participant must genuinely have your problem, otherwise the test measures politeness.
From findings to decisions: severity ranking instead of a sixty-page report
Research that ends in a document has not ended. A sixty-page report with elegant quotes and participant photos gets read by two people, the author and the person who paid, and both skim. The working format is different: one page listing findings, each with a severity, a frequency, a specific recommendation, an owner and a link to a 20-second clip. The clip matters more than the prose: half a minute of a customer failing to find a price convinces better than any paragraph.
Severity is scored on three axes: how many participants hit it, how much it obstructs the task, and whether a workaround exists. A blocker means the task cannot be completed at all or the person leaves. Serious means it can be completed but with lost time and irritation. Moderate means the workaround is obvious. Cosmetic means only a specialist notices. The practical rule: fix blockers on the money path first, meaning forms, cart, payment and lead capture, then everything else by frequency.
A separate discipline is not fixing everything at once. Of ten findings, two or three usually make the next sprint; the rest go to the backlog with their severity attached, and within a quarter half of them stop being relevant on their own. That is fine. The worse scenario is a team trying to close the whole list, development stalling for a month, and the effect becoming unmeasurable because twenty things changed simultaneously.
And a process honesty check: every finding needs an owner and a date, otherwise the list becomes an archive. A healthy research cycle shows, a month after the test, three concrete product changes and a before and after number on at least one of them. If there is nothing to show, the money bought reassurance rather than a decision.
The quantitative signals you already have
Before commissioning anything, spend one working day on what is already being collected. Yandex Metrica session replay shows real sessions: where the cursor hunts, where somebody rereads, where the tab closes. Twenty replays of sessions that ended on the checkout page produce a sharper hypothesis list than most heuristic audits. Do not watch them at random, filter first: one page, one segment, sessions without the target action.
Click heatmaps and scroll maps answer two questions that usually get settled by argument. First, are people clicking things that are not clickable, which is a direct signal of misleading visual affordances. Second, does the audience even reach the block the team considers critical. If 70 percent never scroll to the form, the button wording is not the issue. Form analytics answers a third: which field people start filling and abandon, and how many times they return to a field with an error.
On-site search deserves its own pass. It is the only place where users write, in their own words, what they need: their phrasing, their synonyms, their typos. Zero-result queries are simultaneously catalogue gaps and raw material for copy and keyword work. Same category: split the funnel by device. A mobile conversion rate half the desktop one almost always means one specific broken thing on mobile, not that mobile users buy less.
One important limitation: all of the above answers where, never why. Analytics will show that 60 percent abandon the form on the tax ID field, and will not say whether the person does not know their number, does not want to give it, or does not understand why it is required. Three different causes need three different fixes. That seam is exactly where qualitative research starts, and exactly why it works best when commissioned after the data review rather than instead of it.
- Twenty replays filtered to "payment page without purchase" is the cheapest hypothesis list there is.
- A scroll map settles half the arguments about nobody seeing the key block.
- Zero-result on-site searches are catalogue gaps and ready-made keyword data.
- A twofold mobile versus desktop conversion gap is a defect, not an audience trait.
When research is genuinely a waste of money
Case one: the decision is already made. If a director has settled on the redesign and research is commissioned to confirm it, the outcome is known in advance, because inconvenient findings will be declared unrepresentative. That is not research, it is a legitimisation budget. It is more honest to save the money and admit the decision was a judgement call. Sometimes that is a perfectly reasonable way to run a company, but you do not need to pay for scenery.
Case two: the question is not about the interface. If the product does not sell because it costs 40 percent more than competitors, or because nobody has heard of it, a usability test will find minor friction in the lead form and will not answer the question at all. The tell is a hypothesis phrased as "people do not understand the value". Value is positioning, price and channel, not button placement. Talking to the sales team and reading loss reasons in the CRM is far cheaper here.
Case three: the research costs more than the fix. If the disputed element can be rebuilt in four hours while a test costs two days and 60,000 RUB, it is cheaper to rebuild it and watch the numbers. The same goes for A/B tests on low traffic: detecting a 10 percent relative lift on a 2 percent conversion rate takes tens of thousands of visits per variant. At 200 visitors a month an A/B test measures nothing while looking like measurement, which is the worst possible outcome.
And the fourth, most common case: nobody is available to fix the findings. Research pays off when there is a team, a sprint and a budget for change. If development is booked for a quarter and there is no designer, the report becomes a file somebody opens a year later to discover half the problems are still there. In that situation the same money is better spent fixing the two most obvious defects, which are visible without any research at all.
Frequently asked questions
How much does UX research on a website cost?
Reviewing session replays, funnels and form analytics costs nothing beyond four to eight hours of attention, and that is always the place to start. A five-user usability test runs 30,000 to 120,000 RUB on the Russian market depending on how hard recruiting is. Depth interviews with 8 to 12 people cost 80,000 to 250,000 RUB. A heuristic UX audit runs 40,000 to 150,000 RUB. Participant incentives for a 40-minute session are usually 1,500 to 3,000 RUB each and are billed separately.
Is five users really enough for a usability test?
Yes, with a caveat. Five people surface roughly 80 to 85 percent of usability problems inside one homogeneous segment, a classic result traceable to Nielsen Norman Group research. If the product has genuinely different user types, such as novice and expert, purchaser and end consumer, or mobile and desktop scenarios, you need three to four people per segment. Separately: five people find problems, they do not measure anything. You cannot forecast conversion from that sample.
What is the difference between a UX audit and usability testing?
A UX audit is an expert evaluation against a set of principles: a specialist walks the flows and lists violations. It is cheaper, faster and good at catching obvious defects such as unclear errors, missing states and accessibility problems. Usability testing observes real users attempting a task. It finds what an expert cannot predict: wrong expectations, unfamiliar terminology, misleading visual affordances. In practice the two are combined, with the audit run first so participants do not spend a session on defects you already knew about.
Where do you find participants without a research panel?
Four sources cover most situations. First, customers who recently contacted support: their experience is fresh and they are willing to talk. Second, lapsed leads from the CRM, who will explain why they did not buy better than any survey. Third, the communities and chats where your audience actually lives. Fourth, an on-site invitation with a two-question screener. An incentive of 1,500 to 3,000 RUB for 40 minutes settles consent in most consumer niches; in B2B the offer of sharing the results usually works better than payment.
Can a social media poll replace research?
No, and this is one of the more expensive substitutions. A social poll is answered by the loyal, active part of the audience, people who already understand the product and never hit the beginner problems. On top of that, wording almost always drives the answer, and hypothetical questions like "would you use this feature" are judged optimistically and inaccurately. A survey is useful in exactly one scenario: sizing a problem you already know about, on an adequate sample. It cannot be used to find problems.
Need a hand with this?
We do this work, not just write about it. Describe the task and we will scope it and send a staged estimate.
Related services
- UX/UI design for websites and appsWe design interfaces that survive contact with production: research, structure, UI kit and a clean handoff to the developers. Everything lives in Figma as components with real states and real breakpoints, not as a folder of flat mockups. The same process covers landing pages, corporate sites, e-commerce, web apps and mobile.
- IT consulting and product auditThe most expensive mistakes in software happen before the first line of code: a misread problem, a stack chosen out of a contractor’s habit, and a specification that does not exist. Consulting exists to settle all of that before the development meter starts running: what to build, out of what, at what cost and in which order. The result is a document, not an opinion on a call.
- Website support and improvementNot every project needs a rewrite. It is often cheaper to fix the speed, clear the technical SEO errors, add the missing sections and integrations, and end up with a working site without paying for a new build.
Read next
- Design Systems: When They Pay Off and When They Are an Expensive ToyMost teams asking for a design system need a component library and naming discipline instead. Where the line sits, how to calculate the payback before you start, and what actually belongs in a minimum viable version.
- Redesigning a Site Without Losing Traffic: Stages, Risks and the SEO Migration ChecklistHalf of all redesign requests are really content, speed or offer problems. How to tell them apart, why a phased rollout beats a big bang launch, and what belongs in the migration checklist so the organic traffic survives.
- A Slow Site Loses Money: Core Web Vitals and ConversionSite speed is not a technical metric, it is a line in the revenue report. Core Web Vitals explained in plain language, how to find the real cause of a slow page, and a fix list sorted by return per hour of work.