
How to Do User Research Without a Research Team
On this page
- 1. Mine tickets, search terms and recordings
- 2. Watch five comparable people attempt the task
- 3. Interview recent customers about a real switch
- 4. Ask one open question on the site
- 5. Turn evidence into a findings table
- 6. Check consent and privacy before recording
- 7. Pay for a panel when recruitment is the constraint
UX
Say you run a 25-person local service company. Your booking page converts worse than you think it should, one manager blames the price, another blames the form, and you have no researcher or research budget. The task is to book a service appointment.
The answer to how to do user research is to start with evidence you already collect, watch five comparable people try one task, interview recent customers about a real decision, and then count what you find. Run the seven procedures in order: inspect existing evidence, test tasks, interview customers, ask one survey question, combine findings, check consent and privacy, and pay for a panel only if you need one.
1. Mine tickets, search terms and recordings
Run this: Choose the booking journey and pull either the last 90 days of support tickets or the last 100 tickets, whichever gives you a manageable set. Read sales-call notes from the same period. Assign each ticket one problem code, such as price_unclear, availability_unclear or form_failed, and count tickets by code. Keep an other code for issues that do not fit.
Use the customer's stated problem for the code, not the department that received the ticket. If a ticket contains several issues, code the main reason for contact and keep a note of the others. A canned reply or macro written for a question is also a useful clue that the question recurs; that is an inference about why the macro exists, so check the tickets before treating it as a count.
Next, test your own site search. Enter a distinctive term and inspect the address bar on the results page. GA4 enhanced measurement detects site searches from the URL query parameters q, s, search, query and keyword by default; add any other search parameter in the enhanced measurement settings. The resulting event is view_search_results, and its search_term parameter feeds the Search term dimension. Google's enhanced measurement documentation describes the configuration.
If an instant search filters the page without loading a results URL containing a search parameter, the URL-based enhanced measurement mechanism has nothing to detect. That is an inference from how the feature works; test your implementation before assuming search data is missing. Also inspect the terms for junk: a non-search URL parameter named s or q can be mistaken for a search.
The artefact:
GA4 site-search check and Exploration
1. Search the site for: appointmentavailability
2. Check the results URL for a query parameter, such as:
?q=appointmentavailability
3. In the web data stream's enhanced measurement settings, check that
Site search is enabled and that the actual search parameter is listed.
4. Open Explore > Free form.
5. Set the date range to the last 90 days.
6. Add dimension: Search term.
7. Add metric: Event count.
8. Put Search term in Rows and Event count in Values.
9. Add filter: Event name exactly matches view_search_results.
10. Inspect the terms and their event counts. Check unexpected values
against page URLs for non-search uses of q or s.
11. Record the date range, search parameter and any data-quality issue.
A search term count is a count of search events, not necessarily a count of different people. Group obvious variants for analysis, but retain the original terms so someone else can check your grouping. If no event appears, verify the search URL and configuration before concluding that nobody searches.
For recordings, use Microsoft Clarity to find sessions on the booking journey. Microsoft defines a dead click as a click on an element that receives no feedback in a reasonable time. A rage click is multiple clicks in a clustered area in rapid succession; a quick back is a return to the previous page after a short visit; excessive scrolling is more vertical scrolling than expected. The Clarity semantic metrics reference describes these signals, and the filter reference lists the available filters.
Clarity recording filter recipe
Date range: last 30 days
User info > Device: Mobile
User actions > Insights: Dead clicks
Path > Visited URL: select the booking page URL
Use RegEx if you need to match several relevant URLs.
Save the filter combination as a segment:
Mobile dead clicks on booking
Repeat the review with one Insights filter at a time:
Rage clicks
Quick backs
Excessive scrolling
For each recording, note:
Page and device:
Action immediately before the signal:
What feedback the visitor received:
Whether the visitor continued, went back or left:
Possible explanation to test:
Use the actual booking URL in the filter. If you use a regular expression, check that it matches the URL format Clarity displays for your site before saving the segment. Watch a selection of recordings rather than treating the filter count as a diagnosis.
A rage click records clustered rapid clicks. Watch the recording before calling it frustration: the pattern tells you what happened mechanically, while the surrounding actions may suggest several explanations.
Expected output: You have a ticket count by problem code, a list of actual search terms with event counts, and a small set of recordings annotated with what visitors did. Each source has a date range, so you can judge whether the counts describe the same period.
Bad result: The search report contains image-quality values or tokens, or a recording is labeled “frustrated customer” solely because Clarity flagged a rage click. The first result suggests a parameter collision; the second assigns a motive that the signal cannot establish.
2. Watch five comparable people attempt the task
Run this: Recruit five people who resemble the visitors you want to help and who can attempt the booking task without your coaching. For the hypothetical service company, define the segment first, for example people who would be responsible for arranging this service for their household. Ask them to use their own phone if mobile booking is the journey you are testing.
Give each person the same goal and ask them to think aloud. A facilitator reads the script and keeps prompts neutral; a note-taker records actions, pauses and failures. Get permission before recording, and use the consent procedure in step 6 before collecting or retaining recordings.
Nielsen's model estimates the proportion of problems found as N(1 - (1 - L)^n), where N is the number of usability problems, L is the proportion of usability problems discovered while testing a single user, and n is the number of users. Nielsen gives 31 percent as a typical value of L averaged across many projects. With five users, the estimated proportion is:
1 - (1 - 0.31)^5 = 1 - 0.69^5 = 0.8436
That is about 84 to 85 percent under the model, assuming comparable users use the site in fairly similar ways. It is not a promise for your form. Nielsen's explanation says at least 15 users are needed to find all problems and recommends spending a 15-person budget on three iterative studies of five. For two distinct user groups, NN/g recommends three to four users per group; for three or more groups, it recommends three per group. The five-user recommendation does not apply to quantitative studies of usability metrics, where Nielsen recommends 20 users. The small-business UX guide shows the arithmetic and the journey a test can walk.

Notice the separate roles: the participant uses her own phone, the facilitator follows the script, and the note-taker captures what happens.
The artefact:
Moderated booking task script
Before starting
Facilitator: Thank you for helping us test the website. We are testing
the site, not you. You can stop at any time.
May we record your screen and voice for the purpose explained in the
consent line? If not, we can take notes without recording.
Warm-up
Tell me about the last time you arranged this kind of service.
What device would you normally use to do that?
Think-aloud request
Please say what you are looking for and what you expect to happen as
you go. I may stay quiet so I can see how the site works for you.
Task 1
Imagine you need this service at your home. Find out whether the
company offers the service you need.
Task 2
Find enough information to decide whether the service fits your
budget and schedule.
Task 3
Try to arrange an appointment for a time that would suit you.
Stop before submitting any real personal or payment information
unless we have agreed on a safe test arrangement.
Neutral prompts, only when needed
What are you looking for now?
What do you expect to happen next?
What are you thinking?
What would you do if I were not here?
After each task
How confident are you that you found what you needed?
What, if anything, made that task difficult?
Notes for each task
Started:
Completed, stopped or needed help:
Actions and exact point of difficulty:
Participant's words:
Facilitator intervention:
Do not name a button, menu or page label in the task. If a participant asks what to click, ask what they would do on their own and record the request for help. Stop them before a real submission unless the test arrangement makes that action safe.
Expected output: For each task, you can say how many of the five completed it, where each person paused, and whether anyone needed help. A finding such as “three of five could not determine availability before starting the form” points to a specific part of the journey for inspection.
Bad result: Every participant succeeds because the task text named the link they needed. That shows they could follow your instruction; it does not show they could find the route independently. Rewrite the task as a goal and test it with new participants.
3. Interview recent customers about a real switch
Run this: Contact the last 10 to 15 customers who bought and recruit people from one defined segment. Ask about one recent decision to choose the service, including what they did before choosing it. Keep the conversation on events they can recall: when the issue first appeared, what they noticed while passively looking, when they began actively comparing options, and how they decided.
Bob Moesta and Chris Spiek's four forces give you a way to organize the account. The push of the existing situation plus the pull of the new solution must outweigh anxiety about the new solution plus habit with the present one. Listen for each force in the customer's timeline without putting those labels in their mouth.
Rob Fitzpatrick's The Mom Test offers three useful principles: talk about the person's life instead of your idea, ask about specific past events instead of generalities or future opinions, and talk less while listening more. Do not ask whether a customer “would use” a feature and count a polite yes as evidence of a purchase. Ask what they actually did when they last faced the problem.
The artefact:
Recent customer interview guide
Opening
Thank you for speaking with me. I want to understand a decision you
already made. There are no right answers. May I record this conversation
for the purpose and retention period I have explained?
Use the right-hand question. The left-hand version is a prompt to avoid.
Avoid: Would you use online booking?
Ask: Tell me about the last time you booked this kind of service.
Avoid: Was our service a good idea?
Ask: When did you first think you needed a different solution?
What happened that day?
Avoid: Do you usually compare providers carefully?
Ask: What did you do after that first thought? What did you notice
before you actively started looking?
Avoid: Would a comparison page help you?
Ask: When did you begin actively looking? Where did you look, and
what did you compare?
Avoid: Is price your main concern?
Ask: What costs did you check? What else did you need to know before
deciding?
Avoid: Would you trust an online form?
Ask: What, if anything, made you hesitate about the option you chose?
What almost stopped you?
Avoid: Was the old way inconvenient?
Ask: What were you doing before? What made it easy to keep doing that?
Avoid: Did our best feature persuade you?
Ask: What made the new option appealing at the time? Who else was
involved in the decision?
Avoid: Would you recommend us in future?
Ask: What happened after you decided? What did you do next, and what
surprised you?
Close
Is there anything about that decision I have not asked about?
Confirm who may see the recording and when it will be deleted.
Follow an interesting answer with “What happened next?” or “Can you give me an example from that time?” Leave room for silence and record uncertainty when the person cannot remember. One person's explanation is a lead to investigate, not a percentage of your market.
How many interviews are enough depends on the segment and question. In their systematic review, Hennink and Kaiser reviewed 23 articles and found that empirical studies reached saturation within about nine to 17 interviews, particularly in studies with relatively homogeneous populations and narrowly defined objectives. That range is not a universal rule. Review themes as you go and stop adding interviews when new conversations stop changing the explanation; recruit another segment separately if its decision context differs.
Expected output: You can reconstruct several decision timelines and identify recurring pushes, attractions, anxieties and habits in the customer's own words. You can point to the interviews that support each theme and identify where accounts differ.
Bad result: Your notes say “customers want online booking” because interviewees answered yes to a hypothetical question. You have a stated preference without evidence of the circumstances, tradeoffs or actions that produced a real decision.
4. Ask one open question on the site
Run this: Choose one question based on the decision you need to understand. Show it once per visitor on the confirmation page after a completed booking, or on the pricing page after the visitor has spent some time there. As a working rule, wait at least 30 seconds on the pricing page and do not interrupt people in the first seconds after they land.
Use an open text field and make the question optional. Avoid collecting names or other details you do not need in the answer. Record the page and trigger with your results so you remember whose views the question could capture.
The artefact:
Choose ONE question for the survey:
Confirmation page, after a completed booking:
"What, if anything, almost stopped you booking today?"
Use for: obstacles experienced by people who completed a booking.
Pricing page, after at least 30 seconds:
"What are you looking for that you can't find?"
Use for: missing information among visitors viewing pricing.
Confirmation page, after a completed booking:
"What was unclear while you were arranging your appointment?"
Use for: wording and steps that remained confusing after completion.
Display rules:
- Show once per visitor.
- Use one optional open text field.
- Do not show in the first seconds of a landing.
- Record the page, trigger, dates and number of answers.
A confirmation-page question reaches people who completed the booking, so it cannot by itself explain why other visitors left. A pricing-page answer can surface missing information, but a handful of responses cannot tell you how common each concern is across all visitors.
For a proportion at 95 percent confidence, assuming p = 0.5, the calculation 1.96² × 0.25 / e² gives 385 completed responses for a margin of plus or minus five percentage points and 97 for plus or minus ten points, rounded up. These figures do not repair a biased sample. A few dozen answers are directional: read the wording and themes, not percentages presented as precise estimates. Counts also carry noise of roughly the square root of the count, so a count of 16 has noise of about four as a rough guide. The customer research guide helps match a method to your question and explains the survey calculation.
Expected output: You have a dated set of answers tied to one page and one trigger, with recurring phrases grouped into themes. You can distinguish a question asked after successful booking from one asked while someone was still considering the price.
Bad result: You report that “40 percent of customers are blocked by price” because four of ten optional responses mention cost. The respondents are a small, self-selected group, and the question may never have reached people who left earlier.
5. Turn evidence into a findings table
Run this: Put the observations from the first four procedures in one table. State the source and count beside each finding, including the denominator when you have one. Keep a link or reference to the underlying tickets, recordings or notes in your working copy so the owner can verify it.
Use three severity levels: blocks means the person cannot complete the task; slows means they can continue but must spend extra effort or time; annoys means the issue causes friction without preventing or materially delaying the task. Assign severity from the observed effect, not from how strongly a manager dislikes the page.
As a rule of thumb, give a finding a fix slot when it has support from at least two independent sources or one observed failure. For example, tickets and search terms can point to the same missing information, while a participant unable to submit a form is an observed failure on its own. Treat sources that repeat the same underlying incident as one source, not independent confirmation.

Notice how the existing sources feed one table. The count and source travel with each proposed fix.
The artefact:
| Problem | Evidence: source and count | Severity | Fix | Owner |
|---|---|---|---|---|
| Appointment availability is unclear before the form | Hypothetical example: 14 tickets and 3 of 5 test participants | Slows | Show the availability information people need before they start | Website owner |
| Visitors cannot submit the booking form | Hypothetical example: 1 observed task failure | Blocks | Reproduce the failure and repair the form | Developer |
| Visitors search for a service term that is hard to find | Hypothetical example: 22 search events and 2 sales notes | Slows | Check the service page wording and route from search results | Content owner |
The counts in the table are examples, not results from a real company. Replace them with your own evidence and make each fix specific enough for its owner to check. If the finding points to a broken form or a slow template that needs more extensive work, a site rebuild is one possible fix.
Expected output: Each row connects a problem to an observable effect, evidence, a proposed change and one person responsible for it. A reviewer can tell which issues block booking and which merely add friction.
Bad result: The table contains opinions such as “the page feels old” with no source, count or observed failure. There is no way to decide whether the proposed work addresses the booking problem. After a fix ships, use UX measurement and repair to check whether the relevant task improved.
6. Check consent and privacy before recording
Run this: Check the recording setup before recruiting participants or reviewing new sessions. Under GDPR Article 4(1), personal data is information relating to an identified or identifiable person; Recital 30 identifies IP addresses and cookie identifiers as online identifiers that may be associated with people. Under ePrivacy Article 5(3), storing or accessing information on a device generally requires consent unless an exception applies. Arrange a data processing agreement with the recording vendor under GDPR Article 28(3).
Clarity enforces consent signal requirements for visits originating from the EEA, the UK and Switzerland beginning 31 October 2025. Without consent, Clarity uses limited tracking, assigns a unique ID per page view and does not use cookies to persist session data. We infer that recordings of one visit across several pages are limited.
Send the signal that reflects the visitor's actual choice through your consent setup. The Clarity Consent API reference specifies the consentv2 call and its fields. The following shows the syntax for a visitor who has granted both categories; it is not a default to send before that choice.
The artefact:
window.clarity("consentv2", {
ad_Storage: "granted",
analytics_Storage: "granted"
});
Check masking before recordings contain customer information. Clarity offers Strict, which masks the entire content; Balanced, the default, which masks numbers and email addresses; and Relaxed, which leaves content unmasked while input boxes and dropdowns remain masked. Input-box content is masked in all modes. You can mask a specific element with data-clarity-mask="True"; masking changes apply only to new recordings and can take up to an hour, according to Clarity's masking documentation.
For interviews, ask permission to record at the start and again on the recording. Explain who will see it and when it will be deleted, and remove names from notes before sharing them. California Penal Code section 632 is one example of an all-party consent rule for confidential communications. This is not legal advice; use the privacy and consent reference to work through the applicable requirements with appropriate counsel.
Interview recording consent line
Before recording:
"I would like to record this conversation so [named people or roles]
can review it for [specific research purpose]. We will delete the
recording by [date or stated retention period]. You may decline
recording and still take part. Do I have your permission to record?"
After starting the recording:
"Recording has started. For the recording, do I have your permission
to record this conversation for the purpose and retention period I
just explained?"
If permission is not given:
Do not record. Ask whether the person is comfortable continuing
with written notes.
Fill in the audience, purpose and deletion date or period before the call. Keep the recording and notes accessible only to the people who need them for that stated purpose, and follow the deletion schedule you gave the participant.
Expected output: Your consent system sends the appropriate Clarity signal for the visitor's choice, the vendor agreement is in place, and a new test recording shows the intended masking. Interview recordings contain the participant's permission and have a known deletion date.
Bad result: A team assumes recordings are anonymous because names are hidden, sends granted for every visitor, or changes masking and assumes old recordings have been altered. Identifiers and existing recordings still need to be assessed; a masking change does not rewrite prior captures.
7. Pay for a panel when recruitment is the constraint
Run this: First try to recruit comparable users from the people you can already reach, without requiring them to be employees or people who know the site. Consider a paid panel when you cannot reach the segment yourself or need responses quickly enough that waiting would hold up the decision. Write the segment, task length, screener and recording requirement before comparing prices.
Prices and documentation were read on 29 October 2026. Lyssna's help center lists Growth at about $165 to $199 per month, depending on billing, with the annual figure listed as $1,990 per year. For unmoderated tests, its panel pricing is one credit per minute of test length per response; a credit generally costs about $1, with possible bulk discounts. Interview recruitment is priced separately. A screener adds two credits per participant and recordings add two credits per response.
The artefact:
Panel cost check
Recruitment need:
Comparable people I need:
Why I cannot reach them myself, or why volume is time-sensitive:
Lyssna example:
Responses ordered: 5
Test length: 20 minutes
Base: 20 credits per response
Screener: 2 credits per participant
Recording: 2 credits per response
5 x (20 + 2 + 2) = at least 120 credits
At about $1 per credit: at least about $120 before any bulk discount
Before ordering:
Check the current plan price and billing term.
Check the test length, screening and recording charges.
Confirm the credit total and price on the order screen.
Confirm the panel can reach the defined segment.
The at least $120 figure is arithmetic using the stated assumptions, not a vendor quote, and it excludes the plan subscription. Screeners and recordings can lengthen the estimated test, increasing the credit total. Check the order screen for the actual charge. A longer test or a different recruitment requirement changes the calculation.
Maze's pricing page showed plan features but no readable prices on the date checked; it listed interview studies as Enterprise only. UserTesting's plans page asks you to request pricing and does not publish prices. Ask each vendor for a quote against the same segment and study requirements if a panel is necessary.
Expected output: You can explain which people you need, why your own recruitment is insufficient, and the estimated cost per completed response. The order screen confirms the terms before you commit.
Bad result: You buy five panel responses before defining who should take the test, or present the at least 120-credit example as a guaranteed quote. A fast response from the wrong participant does not answer whether your intended customer can book.
Put your next steps in order
Build on what you have just read with a free ebook on semantic SEO and AI visibility. Get worked examples, decision rules, checklists and a do-this-week action in every chapter, plus a six-week plan for your site.



