Skip to main content Scroll Top

Usability Testing Services: Methods, Sample Sizes and What a Round Hands Back

Updated November 2026 · Written and maintained by the Progression Agency strategy team

A design can pass every internal review and still lose people at the one step that matters. Usability testing finds that step by watching real people attempt real tasks while a researcher notes where they hesitate, misread or give up. The method decides what you learn: five people thinking aloud expose the problems, while a benchmark needs about forty before its numbers mean much. Below are the methods, who to recruit, what to measure, what a round hands back and what studies cost at market rates. Progression’s researchers and designers are based in New York and run studies for apps and websites across the US.

On this page · 13 sections
  1. Every participant called it easy. Two of five finished.
  2. Usability testing services, method by method
  3. How many participants does a round need?
  4. Writing tasks people can attempt without help
  5. Task success, time on task and the SUS score
  6. Benchmarking UX against a baseline
  7. Usability testing for mobile app releases
  8. UX audit services: an expert pass before participants
  9. What a round of testing hands back
  10. Recruiting, consent and incentives
  11. How testing fits a design and build schedule
  12. What usability testing costs
  13. Before the first session: what to send

Nothing built or drawn yet? With no screens drawn yet, research starts inside the design work itself, which our UI/UX design services page covers.

Examples marked “Illustrative example.” are hypothetical, and no real business, product or person is shown.

Every participant called it easy. Two of five finished.

Picture a hypothetical clinic that asks five patients to book an appointment on its new site. Afterward, all five call the flow easy. The recordings say otherwise: two booked, two stopped at an insurance field they thought was required, and one never found the later time slots.

Two drawn boxes. The first says all five participants rated a booking flow easy. The second says only two of five booked, two stopped at the insurance field and one missed the later time slots.Two drawn boxes. The first says all five participants rated a booking flow easy. The second says only two of five booked, two stopped at the insurance field and one missed the later time slots.
Illustrative example. Survey answers beside task results for a hypothetical booking flow.

That gap is why the method watches behavior instead of collecting opinions. In NN/g’s Usability Testing 101, a facilitator gives a participant tasks, then observes what the person does and listens for what they say along the way. A comment that the site looks clean tells you very little; the moment a person hovers over a field for twenty seconds tells you where to work.

  • Watch what people do before you weigh what they say.
  • Note the exact step where each person stops.
  • Treat a closing survey as a supplement, never the result.

Usability testing services, method by method

Each method answers a different question, and each one misleads when it is pointed at the wrong question. The table sets the seven common methods side by side, with the number of people each usually needs.

Seven study methods, the question each answers and where each misleads
MethodThe question it answersWho takes partWhat comes backWhere it misleads
Moderated, in personWhy people struggle, with follow-up questions on the spotAbout five per kind of userNotes, recordings, ranked problemsWhen the facilitator leads the participant
Moderated, remoteThe same as in person, for people spread across the countryAbout five per kind of userScreen and voice recordings, ranked problemsWhen silence is mistaken for confusion
Unmoderated, remoteWhether one element or one change worksFive or more, alone on their own scheduleRecordings and task successOn broad reviews, since nobody can ask why
Think-aloudWhat people believe the interface meansA handful, in any moderated sessionMisreadings in the user’s own wordsWhen task times are needed, since talking slows people
Guerrilla interceptA quick read on a label, layout or first stepPassers-by who fit the audienceA short list of obvious problemsWhen the audience is specialist or hard to reach
Quantitative benchmarkHow good the experience is, in numbersAbout 40 for most studiesSuccess rates, task times, SUS with margins of errorWith small samples, where numbers look precise and are not
Expert heuristic reviewWhat looks wrong before anyone is recruitedThree to five evaluators, no participantsProblems against named heuristicsWhen it is treated as a substitute for real users
Four bands. Why do people struggle: moderated think-aloud or a remote moderated session. Does one element work: an unmoderated remote task or a guerrilla intercept. How good is it in numbers: a quantitative benchmark or SUS after the session. What looks wrong already: a heuristic review by three to five evaluators.Four bands. Why do people struggle: moderated think-aloud or a remote moderated session. Does one element work: an unmoderated remote task or a guerrilla intercept. How good is it in numbers: a quantitative benchmark or SUS after the session. What looks wrong already: a heuristic review by three to five evaluators.
Study methods sorted by the question each one answers.

Moderated or unmoderated: who asks the follow-up question?

In a moderated session a facilitator is present, in the room or on a call, and can ask why a person paused. In an unmoderated session a testing tool delivers written tasks, the participant works alone, and the researcher reviews recordings and task success afterward. NN/g’s guidance on remote usability tests prefers in person sessions where possible, keeps unmoderated work for a few specific elements, and notes that a remote study is still better than skipping the test altogether.

  • Moderated: deeper answers, fewer sessions a day.
  • Unmoderated: many sessions at once, no chance to probe.
  • Unmoderated participants often go quiet, with nobody to prompt them.
  • Practice the remote tools before the first real session.

When is a guerrilla session enough?

A guerrilla session stops people who fit the audience, in a café, a lobby or a trade show aisle, and asks for five minutes on one task. It answers narrow questions fast: does this label make sense, can someone find the first step. It suits consumer products with broad audiences and fails for specialist software, where a passer-by is not a real user.

How many participants does a round need?

Fewer than most people expect for finding problems, and more than most people expect for measuring them. NN/g’s model in Why You Only Need to Test with 5 Users puts the average share of problems one participant reveals at 31%, so five people uncover roughly 85% and fifteen are needed to see nearly all of them.

Horizontal bars rising from 31% with one participant to 52% with two, 67% with three, 77% with four, 84% with five, 95% with eight, 98% with ten and 100% with fifteen.Horizontal bars rising from 31% with one participant to 52% with two, 67% with three, 77% with four, 84% with five, 95% with eight, 98% with ten and 100% with fifteen.
Problems found by number of participants, computed from NN/g’s model of 31% per person.
Participants by kind of study
Kind of studyParticipantsWhy that number
Qualitative round, one kind of user5Most of the common problems appear by the fifth person
Qualitative, three kinds of userAbout 3 to 5 for each kindDifferent people use the product in different ways
Fifteen people availableThree rounds of 5Fixes go in between, and each round checks the last
Quantitative study, 95% confidenceAbout 40A 15% margin of error on a success rate
Quantitative study, 90% confidenceAbout 30The same margin, with more risk accepted
Satisfaction questionnaire aloneAt least 20 to 30Fewer rarely generalize
Expert review3 to 5 evaluatorsEach catches problems the others miss

Numbers change once the goal is a metric rather than a list of problems. NN/g’s summary of quantitative sample sizes recommends 40 participants for most quantitative studies, which gives a 15% margin of error at 95% confidence; at 90% confidence the figure falls to about 30.

5: Per kind of user. A qualitative round with one kind of user, as NN/g advises.
31%: Found by one person. The average share of problems a single participant reveals in NN/g's model.
3 x 5: Rounds, not one big study. How NN/g would spend a budget of fifteen participants.
40: For a quantitative study. Enough for a 15% margin of error at 95% confidence.
68: Average SUS score. Across 500 studies; 80 or more reaches the top tenth.
3 to 5: Expert evaluators. Each reviews the interface alone before the findings merge.

Three rounds of five, with fixes in between

Given fifteen people, NN/g would run three studies of five rather than one of fifteen. The first round finds most problems, the second checks whether the fixes worked and digs deeper into structure, and the third catches what the first two missed. A single participant is not enough on its own, since one person may do something odd by accident.

  1. Round one: five people, the main tasks, the obvious problems.
  2. Fix the problems that blocked people.
  3. Round two: five new people, the same tasks, the fixes checked.
  4. Fix again, now at the level of navigation and flow.
  5. Round three: five more, to confirm and catch the rest.

When the product serves very different people

The five person figure holds only for people who use a product in similar ways. A marketplace with buyers and sellers, or a school app used by parents and teachers, needs a small round for each kind of user, because their tasks and habits barely overlap.

Writing tasks people can attempt without help

The task is the instrument. A task names a realistic goal, gives just enough context to start, and never uses the words printed on the button the person needs to find.

  • Start from a goal: book a cleaning for next Tuesday afternoon.
  • Leave out interface words that point to the answer.
  • Give the details a real person would have, such as a date or a budget.
  • Ask the participant to read each task aloud before starting.
  • Keep the same order for everyone in a quantitative study.
  • Plan one or two spare tasks in case a session runs fast.

What makes a task scenario fair?

A fair task describes what the person wants, not how to get it. NN/g warns that small wording errors can prime participants, so a task that says click Schedule has already done half the work. Reading the task aloud also confirms the person read all of it.

The think-aloud protocol in three steps

NN/g calls thinking aloud the single most valuable usability method, and the recipe is short: recruit representative users, give them representative tasks, then stay quiet and let them talk. It works from paper sketches to finished systems, and about a day of sessions with a handful of people produces the most important findings.

  • Ask people to say what they are looking for and what they expect.
  • Prompt gently when someone falls silent.
  • Never explain the interface during a task.
  • Save questions about preferences for the end.

Task success, time on task and the SUS score

Three measures carry most quantitative work. Task success is the share of participants who complete a task, which NN/g calls the simplest usability metric. Time on task shows effort. The System Usability Scale captures how usable the whole product felt.

Scoring task success in four levels
LevelWhat happenedHow it is reported
Complete successThe task finished exactly as askedShare of participants who finished with no error
Success, minor issueFinished, with a small slip such as a skipped optional fieldShare who finished with a minor issue
Success, major issueFinished, but with an error that changes the outcome, such as a wrong dateShare who finished with a major issue
FailureCould not finish, or gave upShare who failed, with the step where each one stopped

Partial credit is a judgment the team agrees before the sessions, not after. Decide which slips count as minor and which change the outcome, write it down, and score every session against the same definitions.

The SUS is a ten question survey given once, after all the tasks. As MeasuringU’s SUS guide explains, each item has five response options; odd items score the answer minus one, even items score five minus the answer, and the total times 2.5 gives a number from 0 to 100. John Brooke released the scale in 1986.

Is a SUS score of 70 good?

It is slightly above average. NN/g’s article on measuring perceived usability puts the average at 68 across 500 studies, and notes that a score of 73 sits only in the top 30%, while the top 10% needs 80 or more. The score is not a percentage, and changing the order or wording of the ten questions breaks comparisons with published data.

One question after each task

A short rating after each task shows which parts of a flow feel hardest while the task is fresh. NN/g advises keeping these to one to three questions, often a single ease rating, on a scale of no more than seven points, and pairing any rating with a quick why.

  • SUS: once per session, after the last task.
  • Single ease rating: after every task.
  • At least 20 to 30 people before an average score means much.
  • Always read ratings beside success and time, never alone.

A flow that loses people?Tell us the task and where it lives. You get a free proposal with the method, the participants and the price.

Get a free proposalTell us about a project you’d like help with.

Benchmarking UX against a baseline

A benchmark answers a different question from a findings round: not what is wrong, but how good the experience is and whether it improved. NN/g describes benchmarking as a summative evaluation, a snapshot of one version measured against a meaningful standard, using analytics, surveys or quantitative testing.

Four reference points for a UX benchmark
Compare againstIllustrative example.When it fits
An earlier versionCheckout took a median of 90 seconds before the redesign and 60 afterEvery redesign after the first baseline
A competitorOur sign-up success rate beside a rival’s on the same taskThe first study, when no earlier figure exists
An industry standardA SUS score beside the average of 68Any product, since SUS has wide published data
A goal set by the teamBooking success of at least 90% by the next releaseWhen leadership has set a target

The first measurement becomes the baseline. Until a second one exists, a competitor or a published standard gives the number some meaning; after that, each release is compared with the last.

How often should a benchmark run?

Quantitative testing is costly enough that many teams benchmark once a year, while analytics-based measures can be checked after every major redesign. Keep the tasks, wording and recruiting the same between rounds, or the comparison stops meaning anything.

  • Pick three to five tasks that matter to revenue or support load.
  • Record success, time and a satisfaction score for each.
  • Recruit the same kind of people every time.
  • Report every figure with its margin of error.

Usability testing for mobile app releases

A phone study follows the same plan as any other study, with extra care for the device, the network and the recording. NN/g’s guide to mobile usability testing recommends recruiting people who have used their own kind of device for at least three months, unless learnability of a new device is the point.

A drawn card with five rows: participants used to their device for three months, a screener naming the exact device, cellular and WiFi passes, no light above the screen, and a camera that records screen and fingers.A drawn card with five rows: participants used to their device for three months, a screener naming the exact device, cellular and WiFi passes, no light above the screen, and a camera that records screen and fingers.
A drawn card with five checks for a mobile session.
  • Screen for the exact device, since many people own more than one.
  • Run tasks on cellular data and on WiFi; each shows different problems.
  • Control the light so the screen recording stays readable.
  • Record with a camera that captures the fingers as well as the screen.
  • Confirm participants know their store password if an install is part of the task.

Mobile app usability testing before a store submission

A round on the release candidate catches what design reviews miss: a step that needs two hands, a field the keyboard hides, a permission request with no explanation. Our mobile app design company page covers the store assets and handover that follow.

UX audit services: an expert pass before participants

An expert review checks an interface against a set of usability principles before anyone is recruited. NN/g’s guide to running a heuristic evaluation recommends Jakob Nielsen’s ten heuristics for general usability, with three to five evaluators working independently, because any one reviewer misses problems the others catch.

  1. Agree the heuristics and the scope of screens.
  2. Each evaluator reviews alone and logs every issue.
  3. Merge the logs and remove duplicates.
  4. Rate each issue by severity.
  5. Turn the top issues into tasks for the next round with participants.

Can an audit stand in for sessions with users?

An audit complements sessions rather than replacing them. It finds glaring problems cheaply and shows a study where to aim, but NN/g is plain that a heuristic review cannot replace research with real users, because usability depends on who is using the product and why.

What a round of testing hands back

A useful round ends in decisions, not a slide deck. Every finding is tied to evidence, ranked by how badly it hurts the task, and paired with a proposed change.

A drawn card with eight rows: test plan, screener, task scenarios, session recordings, highlight clips, findings by severity, metrics sheet and retest plan. Findings by severity is highlighted.A drawn card with eight rows: test plan, screener, task scenarios, session recordings, highlight clips, findings by severity, metrics sheet and retest plan. Findings by severity is highlighted.
A drawn card listing eight items a round of testing hands back.
  • A one page summary of the top problems.
  • A findings table ranked by severity.
  • A clip for each problem, under a minute each.
  • Success and time by task where measured.
  • The SUS score with its margin of error, for benchmark studies.
  • A retest plan for the changed screens.

Who should watch the sessions?

The people who will make the fixes. NN/g observes that even skeptical developers and executives soften once they watch customers struggle, and that sitting in on a few sessions takes little of their time. A shared live link or a same-day clip reel gets them there.

The right five people matter more than the method. A participant should be a realistic user: someone who already uses the product, or someone with the same needs and background as the people it is built for.

  • Write a screener that filters on behavior, not job title alone.
  • Exclude people who work in design, research or for a competitor.
  • Ask for consent to record before the session starts.
  • Explain how recordings are stored and who will see them.
  • Pay an incentive that respects the participant’s time.
  • Recruit one spare in case someone does not show.

NN/g notes that even a simple study usually pays a few hundred dollars in participant incentives. Specialist audiences, such as clinicians or finance staff, cost more to reach and take longer to schedule.

Should staff ever take part?

Only for a dry run. Colleagues know the product, the jargon and the shortcuts, so they cannot show where a newcomer stumbles. Use them to rehearse tasks and timing, then recruit real users for the study itself.

How testing fits a design and build schedule

Studies cost least when they come early. A clickable prototype can be tested before a line of code exists, which turns a costly rebuild into an afternoon of design changes.

A timeline of four points: sketch or wireframe, clickable prototype, coded build and live product, with the kind of study that fits each.A timeline of four points: sketch or wireframe, clickable prototype, coded build and live product, with the kind of study that fits each.
Four points in a build where a study fits.
A six step flow: set the question, choose the method, recruit, run the sessions, rank the findings, fix and retest.A six step flow: set the question, choose the method, recruit, run the sessions, rank the findings, fix and retest.
Six steps in one round of testing.

How is a usability test different from QA?

QA checks whether the software works as specified; a study checks whether people can use it to reach their goal. A release can pass every functional test and still lose half its users at an unclear step. Our quality assurance testing services cover the functional side, and conversion rate optimization carries findings into live experiments once a product has traffic.

  • Wireframe: think-aloud with five people.
  • Prototype: moderated sessions on the main tasks.
  • Coded build: task tests on the release candidate.
  • Live product: a benchmark repeated each year.

What usability testing costs

Price follows the shape of the study. NN/g sketches the floor as about three days of work, plan, test five people and analyze, plus incentives, and notes that the most elaborate studies can reach several hundred thousand dollars.

UX research and design work: published US price bands (market ranges only)
WorkTypical low to published highNotes
UX audit and fix list$3,000 to $8,000Expert review with a ranked list of changes
Fixed scope project (audit, plan, launch)$2,500 to $40,000A defined deliverable with a start and an end
App UI/UX design, one platform$8,000 to $25,000Research and testing folded into design
App UI/UX design, iOS, Android and web$20,000 to $40,000Three platforms, more sessions
UI/UX design project$10,000 to $100,000Wide spread, set by screens, platforms and research depth
Horizontal range bars for a UX audit, a fixed scope project, app design on one platform, app design across three platforms and a UI/UX design project.Horizontal range bars for a UX audit, a fixed scope project, app design on one platform, app design across three platforms and a UI/UX design project.
Published US price bands (market ranges only) for five kinds of UX research and design work, typical low to published high.

Progression’s price for a study is set in the written proposal, after scope, and is subject to change. Studies inside a wider design project are often folded into that project’s price, covered on our UI/UX design page.

What moves the price of a study?

The drivers NN/g lists are the ones that matter in practice: testing several designs against each other, testing in more than one country, testing several kinds of user, running a quantitative study, specialist equipment such as eyetrackers, a room where others can observe, and a detailed written report.

  • Number of kinds of user, each with its own round.
  • Moderated or unmoderated sessions.
  • Qualitative findings or a quantitative benchmark.
  • Specialist recruiting and incentives.
  • Countries and languages.
  • Depth of the report and the clip reel.

Pricing depends on scope and is confirmed in a written proposal. Published prices are subject to change.

Before the first session: what to send

A short brief saves a week of back and forth. Send these and the plan can be drafted before the kickoff call.

  • The product, prototype or link to be tested.
  • The two or three tasks that matter most to the business.
  • Who the real users are, and any kinds of user that differ.
  • Analytics or support tickets that point to trouble spots.
  • Any earlier study, survey or SUS score.
  • The release date the findings must meet.

Digital.gov’s usability topic page frames the goal well: how easily and effectively people reach their goals with a product, measured through success rates and satisfaction. See also our product design agency and landing page design pages for the design work around a study.

What this page leaves out: the Progression Method, proprietary and reserved for clients. Reach out to learn more

Frequently asked questions

Which pieces of work come with usability testing services?
A test plan, recruiting, task scenarios, the sessions themselves, recordings, findings ranked by severity with clips, and a retest plan. Benchmark studies add success rates, task times and a SUS score with margins of error.
How many participants does a usability study need?
About five for a qualitative round with one kind of user, and about 40 for most quantitative studies. Products with very different kinds of user need a small round for each.
What is the difference between moderated and unmoderated sessions?
A facilitator is present in a moderated session and can ask follow-up questions. In an unmoderated session a tool delivers the tasks, the participant works alone and the researcher reviews recordings afterward.
Is remote testing as reliable as testing in person?
It is reliable enough for most studies, and far better than skipping the test. NN/g still prefers in person sessions where possible, because body language is easier to read in the room.
What is the think-aloud method?
Participants say what they are thinking as they use the product, so the researcher hears their expectations and misreadings in their own words. It works from paper sketches to finished software.
When do teams use guerrilla intercepts?
Teams use them for narrow questions about a broad consumer audience: a quick, informal session with people who fit it, stopped in a public place for a few minutes on one task.
How is task success scored?
As the share of participants who complete the task. Many teams add levels such as complete success, success with a minor issue, success with a major issue and failure, defined before the sessions.
Which SUS result counts as above average?
Anything above 68 is above average, since 68 is the average across 500 studies. A score of 80 or more places a product in the top tenth.
How many people does a quantitative benchmark need?
Around 40 for most studies, which gives a 15% margin of error at 95% confidence. About 30 is enough at 90% confidence.
What does benchmarking tell a product team?
Whether the experience is improving, measured against an earlier version, a competitor, an industry standard or a goal the team has set.
How do you test a mobile app with participants?
On the kind of device each person already uses, on both cellular data and WiFi, with a camera that records the screen and the fingers. The tasks and analysis follow the same plan as any other study.
How do UX audit services review an interface?
Three to five evaluators each check it alone against named usability heuristics, then merge their issues and rank them by severity.
Can an expert review replace sessions with real users?
It cannot replace them, though it makes them sharper. An audit finds obvious problems cheaply and points the next study at the screens most likely to fail.
What does a findings report contain?
A short summary, the problems ranked by severity, a clip for each, success and time by task where measured, and the changes proposed for each problem.
What price range should a buyer expect for usability testing?
Published US bands run from $3,000 to $8,000 for a UX audit and fix list and from $2,500 to $40,000 for a fixed scope project. Progression sets its price in the written proposal.
From planning to findings, how many days does a round need?
About three days of work for the simplest round, once participants are booked: a day to plan, a day of sessions and a day of analysis. Recruiting specialist audiences adds time.
Who should be recruited?
Realistic users: people who already use the product or share the needs and background of those who will. Colleagues and people in design or research roles are left out.
Can you test a prototype before anything is coded?
Yes, and that is when changes are cheapest. Clickable prototypes and even paper sketches work well with think-aloud sessions.
Is a usability study the same as QA testing?
They answer different questions. QA checks that the software works as specified; a study checks that people can use it to reach their goal.
Are participants paid?
Yes, usually. Even a simple study typically pays a few hundred dollars in incentives, and specialist audiences cost more.
Can Progression run usability studies for our app or website?
Yes. Progression’s researchers and designers plan and run moderated, remote and unmoderated studies, benchmarks and expert reviews, and the team is based in New York with on-site sessions available on request.

A flow that loses people?Tell us the task and where it lives. You get a free proposal with the method, the participants and the price.

Get a free proposalTell us about a project you’d like help with.

Get a free marketing proposal

Tell us what you are trying to grow and we will come back with a plan, not a pitch deck. Same-day reply on weekdays.

Privacy Preferences
When you visit our website, it may store information through your browser from specific services, usually in form of cookies. Here you can change your privacy preferences. Please note that blocking some types of cookies may impact your experience on our website and the services we offer.
Contact Us