
Customer Research Methods
On this page
- Match the question to the method
- If you are launching a new product or offer: interview people about the problem
- If growth has stalled: reconstruct recent switches
- If you are losing deals: ask buyers about the decision
- If you are entering a new segment: sample that segment separately
- Interview counts: start small and check for new themes
- Survey sample size: count responses, then check the population
- Ask about past behavior without leading the witness
- Mine existing language, with consent for recordings
- Use Search Console queries to find buyer questions
- Sample what AI assistants say about your category
- Keep an insight log that connects evidence to decisions
Marketing
You have a hunch about why people buy, your CRM says lost deals came down to price, and the buyer may have built a shortlist in a chat window and a search box you never see. Customer research methods answer different questions: why people buy or leave, why deals go, how many share a problem, what words they use, and what they hear before contacting you. For interviews, the most useful starting fact is that a systematic review found saturation at about 9 to 17 interviews for narrowly defined questions in relatively homogeneous groups.
Match the question to the method
Start with the decision you need to make. A company of 10 to 200 people rarely needs every method at once, and no single method answers all of these questions.
| Method | What it can answer | What it cannot answer on its own |
|---|---|---|
| Customer interviews | How people handle a problem and what makes it difficult | How common a finding is across a population |
| Jobs to be done switch interviews | What prompted a recent buyer to change solutions | What people who have not switched will do |
| Win/loss interviews | Why a buyer chose you, a competitor, or no purchase | The needs of buyers who never entered a deal |
| Surveys | How many respondents report a known problem or preference | Why they behave as they do |
| Support tickets and reviews | The words existing customers use, often without prompting | What prospects who never bought think |
| Search Console queries | Which visible search terms bring people to your site | Every query people use or the buyer's full journey |
| Sales call recordings | What buyers ask and challenge during a call | Research that happened before the call |
| Repeated AI assistant prompts | Which names and criteria appear in sampled answers | What every buyer saw or whether an answer caused a purchase |

Notice how the first method changes with the decision: a new offer starts with problem interviews, while lost deals call for conversations about actual decisions.
The thresholds in the four branches are our rules of thumb, drawn from the research cited here. They are planning aids, not published standards or guarantees of saturation.
If you are launching a new product or offer: interview people about the problem
If: You have an idea but few or no customers for it, you need to know whether the problem occurs, what people do now, and what that workaround costs them.
Then: Start with open interviews about recent behavior. Recruit people who fit one intended buyer segment, ask about the last time the problem arose, and trace what they did before presenting an idea. Use the question rewrites below to keep the conversation grounded in events.
Threshold: Plan for 8 to 12 conversations in that segment as a rule of thumb. Start with 5 to 6, review the notes, and continue until new conversations stop adding themes. If the offer serves distinct groups, treat each group as a separate sample.
First step: Write one question you need to answer, such as whether operations managers already spend time reconciling supplier records. Ask a potential participant to describe the last time they handled that task, including the people involved and any money or hours spent.
If growth has stalled: reconstruct recent switches
If: You have customers but cannot explain why similar prospects choose you, delay, or stay with their current approach, examine actual changes in behavior.
Then: Interview recent customers who switched to you and, where possible, people who recently left for another solution. A jobs to be done switch interview follows the decision from the first thought of changing through purchase. Ask what was happening before they looked, which alternatives appeared, and what nearly stopped the move. Pair these interviews with Search Console query mining to find the language people use before contact.
Threshold: As a rule of thumb, contact the last 10 to 15 customers who bought and any recent departures. Review themes as responses come in and stop adding interviews for a segment when new conversations no longer change the explanation. If buyers in different roles describe different decisions, recruit those roles separately.
First step: Pull a list of recent purchases and departures. Invite people who actually switched, then begin with the circumstances before they started looking, not a request to rate your service.
Bob Moesta and Chris Spiek's four forces model frames the move as Push from the current situation plus Pull toward a new solution exceeding Anxiety about the new solution plus Habit around the present one. Ask about all four: the frustration, the attraction, the risk, and what made staying easy. In Demand-Side Sales 101, Moesta describes stages from first thought through passive looking, active looking, deciding, onboarding, and ongoing use; use those stages to reconstruct the buyer's account without assuming every purchase followed an identical path.
If you are losing deals: ask buyers about the decision
If: The CRM repeatedly says price caused the loss, you have a useful prompt for an interview, not a complete diagnosis. Clozd, a win/loss vendor, reports that sales representatives often record price as the reason, while buyer interviews can reveal other factors or a more specific role for price.
Then: Run short win/loss interviews with buyers who recently chose you, chose someone else, or decided against buying. Ask what started the search, which alternatives made the shortlist, how the buyer compared them, and what decided the outcome. Include wins so you can compare the circumstances on both sides.
Threshold: As a rule of thumb, begin once you have at least five recent decisions to approach, aiming for both wins and losses. Contact each buyer within two weeks of the decision when possible, and keep reviewing answers as they arrive. If five decisions all come from one buyer type, do not treat them as an explanation for every segment.
First step: Pull the five most recent decisions from the CRM and note the recorded reason without showing it to the interviewer as the expected answer. Invite buyers to discuss their decision, including buyers who chose a competitor.
- →Invite promptly
Clozd recommends contacting buyers within two weeks of the decision and keeping the interview to 30 minutes or less. Offer a thank-you incentive if appropriate; its guidance gives examples starting at $25.
- →Follow up
Send a second email when the first receives no answer. Clozd reports that most interview appointments are booked on or after the second email.
- →Reconstruct the choice
Ask which options were considered, what each needed to prove, and what happened before the decision. Record the buyer's account alongside the CRM reason.
- →Compare outcomes
Group findings by buyer type and outcome. Check whether price means an unaffordable total, an unclear return, or a competitor's offer that made a different tradeoff attractive.
These scheduling details come from Clozd's win/loss interview guidance, published by a vendor in the field. When interviews show which alternatives buyers compare and which buyers fit your offer best, use the small business positioning guide to turn those findings into a position.
If you are entering a new segment: sample that segment separately
If: You know your current customers well but are selling to a different industry, role, or company size, the old interview count does not automatically describe the new group.
Then: Interview people in the new segment about their recent buying and problem-solving behavior. After you have their words and alternatives, sample AI assistants with questions a buyer in that segment might ask. Compare the assistants' criteria with the criteria interviewees actually used.
Threshold: Plan for 8 to 12 interviews in the new segment as a rule of thumb, starting with 5 to 6 and continuing until new themes stop appearing. For the assistant sample, use five buyer-style prompts, three assistants, and ten fresh runs per prompt and assistant. Review the results after the first interview batch so the prompts reflect real buyer language.
First step: Define the segment by a role and buying context, such as facilities managers evaluating a maintenance service. Recruit people who have faced that decision recently, and ask what triggered it and where they looked.
Interview counts: start small and check for new themes
The familiar advice to test with five users comes from usability testing. Nielsen and Landauer estimated that five testers could find about 85% of usability issues in that context. Exploratory interviews seek a different kind of evidence; Maria Rosala of NN/g wrote on October 31, 2021 that five participants are too few for many such studies. Use UX research without a research team when the question is whether people can use an interface.
The Hennink and Kaiser systematic review found saturation within 9 to 17 interviews in empirical studies, primarily with homogeneous populations and narrowly defined objectives. That range is evidence for planning, not a promise that interview number 17 will settle every question.
NN/g reports that Guest, Bunce and Johnson analyzed 60 interviews with women in two West African countries in sets of six. Of 36 high-frequency codes, 34 appeared after six interviews and 35 after twelve, in a group the authors describe as fairly homogeneous. NN/g also cites Griffin and Hauser's finding that 20 to 30 interviews uncovered 90 to 95% of needs for portable food-carrying devices, and Mason's median of 31 interviews across more than 2,000 qualitative PhD theses. Those are different studies and populations, so their counts provide context rather than a universal target. See NN/g's interview sample-size analysis and the Guest, Bunce and Johnson paper.
Recruit more people when the question is broad, the population is diverse, the interviewer is less experienced, participants know less about the subject, or interviews vary substantially in structure. Keep a common set of open questions, analyze after the first 5 to 6 interviews, and add participants while new themes keep appearing. Record the themes as you go so the decision to stop has an observable basis.
Survey sample size: count responses, then check the population
A survey can estimate how widely a problem appears after interviews have identified it. For a proportion at 95% confidence, use z = 1.96; p = 0.5 gives the conservative sample size when the proportion is unknown.
n = z² × p × (1 - p) / e² n_adjusted = n / (1 + (n - 1) / N)
Here, e is the desired margin of error as a decimal and N is the size of the population you are sampling. With p = 0.5, a margin of error of plus or minus 5 percentage points requires 385 completed responses after rounding up; plus or minus 10 points requires 97. Applying the finite population correction to an unrounded calculation for a base of 300 customers gives 169 and 73 completed responses, respectively.
Those are completions, not invitations. In a hypothetical 300-customer base with a 20% response rate, inviting everyone would yield about 60 responses, short of either target. Use the result as a directional count of the problems interviews found, and report the response count and who was invited. The margin-of-error calculation also assumes a suitable random sample; a low or selective response rate can introduce bias the formula does not fix.
Ask about past behavior without leading the witness
Rob Fitzpatrick's The Mom Test sets out three rules: talk about the person's life rather than your idea, ask about specific past events rather than general views or future opinions, and talk less while listening more. The book addresses a practical problem: people can be polite about an idea, so compliments and hypothetical promises to buy do not establish that a problem is real or that a purchase will happen.
| Leading or hypothetical question | Question grounded in behavior |
|---|---|
| Would you use a tool that solves this? | When did this last come up, and what did you do? |
| Do you find supplier reporting frustrating? | Tell me about the last supplier report you prepared. What took the most work? |
| How much would you pay for a fix? | What do you spend now dealing with it, in money or hours? |
| Would faster onboarding make you choose us? | How did onboarding affect your last choice, and what happened after purchase? |
| Is our competitor too expensive? | Which options did you compare, and how did their costs affect your decision? |
Follow the answer with a request for a concrete example. If someone says a task is frequent, ask when it last happened; if they mention cost, ask how they calculated it. Leave enough silence for them to describe a constraint you did not anticipate.
Mine existing language, with consent for recordings
Support tickets and reviews contain language customers use without being prompted by an interview question. Joanna Wiebe calls the copywriting practice of extracting that language message mining. Capture the original wording and its context, then group observations by the problem described; a ticket about a difficult handoff may be more useful than a generic satisfaction score.
These sources over-represent existing customers and problems that generated a complaint. They cannot tell you why an unknown prospect never contacted you. Sales call recordings add questions and objections from active buyers, though they begin after some research has already happened.
Announce the recording at the start of the call and obtain the required consent. For example, California Penal Code section 632 addresses recording confidential communications without the consent of all parties.
Use Search Console queries to find buyer questions
Open the Performance report in Search Console and filter queries with a custom regular expression. This RE2 pattern finds queries beginning with common question words, plus queries containing comparison, cost, and review terms:
(?i)^(how|what|why|which|who|when|where|can|does|do|is|are|should)\b|\b(vs|versus|alternative|alternatives|best|cost|price|review|reviews)\b
A query such as how much does calibration cost suggests a cost question; acme vs zeta suggests a comparison. Neither tells you who searched or whether they purchased. Use the phrases as prompts for interviews and as evidence of questions visible in your site's search data.
The Search Console Performance report documentation says the interface table shows at most 1,000 rows and omits anonymized queries, including some rare or sensitive queries. An API can provide more rows, but the report is still not a complete record of every search. Keep the date range and filter with each export so someone can reproduce the observation.
Sample what AI assistants say about your category
Buyers may encounter a list of suppliers before your sales team knows they are looking. In a Responsive vendor survey of 350 B2B buyers reported in October 2025, about a quarter said generative AI had overtaken traditional search for vendor research, and nearly two-thirds said they used it as much as or more than search. Gartner's widely cited 2018 B2B buying research reported that buyers spent about 17% of buying time meeting potential suppliers, shared among the suppliers under consideration.
One answer from ChatGPT is an anecdote. In a SparkToro and Gumshoe study, ChatGPT and Google's AI returned identical brand lists in under 1% of repeated runs. The study was not peer-reviewed, and Gumshoe sells AI tracking, so treat the result as a reason to sample variation, not as a universal rate.

The lists change across runs. Count the runs containing a name to calculate its presence rate for that prompt and assistant.
Write five prompts using the words and situations you heard from buyers. Run each prompt ten times in fresh sessions on each of three assistants: 5 × 3 × 10 gives 150 answers. Record the prompt, assistant, run, brands named, whether your company appeared, and the criteria listed. For each prompt and assistant, divide the runs naming your company by ten; six appearances give a 60% presence rate for that small sample.
Treat the criteria as a draft of the checklist a buyer might encounter, then compare it with interview evidence. Assistants draw on information across the web, so the sample can also reveal how the category is described beyond your own site. Iron Goo's AI search visibility service can help track and address how assistants describe your company.
Keep an insight log that connects evidence to decisions
Use one shared spreadsheet with one row per observation. Keep an evidence link to the source material, with access appropriate to the sensitivity of the interview or recording. The following rows are openly hypothetical examples.
| Date | Source | Segment | Exact quote or data point | Job or problem | Times seen | Decision affected | Evidence link |
|---|---|---|---|---|---|---|---|
| Hypothetical: 2026-10-05 | Interview | Operations manager | “I have to chase three suppliers for the same report.” | Collect supplier reports | 1 | Which problem to investigate | Link to interview notes |
| Hypothetical: 2026-10-12 | Support ticket | Existing customer | “I cannot tell which report is current.” | Identify the current report | 1 | Whether to clarify report status | Link to ticket |
Increase a tally when separate observations support the same theme, while keeping the individual evidence available. A tally is a prompt to investigate, not a population percentage: ten tickets from existing customers do not represent ten randomly selected buyers. Repeated buyer quotes can become raw material for the claim in your value proposition.
| Situation | First method | Planning sample | Time to a first pattern (our estimate) | What it will not tell you |
|---|---|---|---|---|
| Launching a new offer | Problem interviews | Start with 5 to 6; plan 8 to 12 per segment | 2 to 4 weeks recruiting strangers; a first pattern after 5 to 6 interviews | How many people in the market share the problem |
| Growth has stalled | Switch interviews | Contact the last 10 to 15 buyers and recent departures | 1 to 3 weeks; recent buyers are in the CRM | Why people who never switched stayed put |
| Losing deals | Win/loss interviews | Approach at least 5 recent decisions, including wins and losses | 1 to 2 weeks for the first interview, often 1 to 2 months until 5 or more decisions accumulate | Why prospects who never entered a deal stayed away |
| Entering a new segment | Interviews in that segment | Start with 5 to 6; plan 8 to 12, then sample assistant answers | About a day for the assistant sample, 2 to 4 weeks for interviews | Whether the segment's findings apply to existing buyers |
Teresa Torres describes continuous discovery as including weekly customer touchpoints by the team building the product. Give someone ownership of a weekly log review: add the new observation, update its theme tally, and name the decision it could affect.
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.



