Do not start with the ten people who already love you. They will reply within the hour, give you all the time you ask for, and tell you your product is great. You will leave every call feeling good and learn almost nothing you can act on.
An interview project is won or lost before the first question is asked. It is won or lost at the invitation list. If you pick people because they are easy to reach, you will hear from people who are easy to reach. If you pick them because they can answer your decision, you will hear something you can use.
This guide shows you how to do the second. You will write eligibility rules from the decision you need to make, choose channels that reach the people those rules describe, and send an invitation that is easy to say yes to. It works the same way whether you are talking to customers, users, your own staff or partners. The kit is below: a screener, two invitation messages, a reminder, and a channel table. It starts once you know what you are studying; if you are still choosing, read which research study to run first.
Write the criteria from the decision, not from your customer list
Start with the choice you are trying to make. "Should we build online booking?" is a decision. "Learn about our customers" is a mood. The decision tells you who can help: for the booking question, it is people who recently arranged an appointment, whether they did it well or badly.
1 puts it plainly: your participants must be actual or likely users of your service. Its criteria can describe a demographic, a target user group, a particular experience, a problematic situation or the ways people access the service. You do not need a persona for any of them. You need people with the relevant experience.
Write your criteria in three buckets.
- Must have. The few things without which the person cannot answer your question. One recent behaviour and one role is a good start.
- Must not have. Things that make their answers unusable for this decision. A conflict is the obvious one: they work on this product or for a competitor. So is having no first-hand experience of the thing, or being an investor or adviser to the company.
- Nice to have. Things you want a spread of, such as size of business, length of time as a customer, or region. These shape who you pick among the eligible. They do not decide who gets in.
Keep the must-haves short. Nielsen Norman Group (NN/g) warns that2 produces a screener nobody can pass, and says to screen only for criteria that affect your research questions. If a criterion would not change what you do with the answer, delete it.
Recent behaviour beats demographics
"Women aged 35 to 54 who run small shops" tells you who someone is. It does not tell you whether they can describe what happened the last time they ordered stock. NN/g notes that2 for most studies.
Write your must-haves as things people did, and put a time window on them. "Booked a repair in the last three months" is a better criterion than "has used our repair service". Keep the window short so the occasion is still fresh. Your interview can then ask about that specific occasion, which is the foundation of good interview questions.
The eligibility screener
Copy this and fill the brackets. It has two parts: the rules you apply to your list, and the questions you ask when your records cannot tell you.
DECISION
We need to decide: [one sentence, e.g.
"whether to add online booking"]
MUST HAVE (all required)
1. In the last [time window], they [specific behaviour].
2. Their role: [who made or shared the decision / did the work].
3. [Anything else that lets them answer the decision.
Four items at most.]
MUST NOT HAVE (any one rules them out)
1. They work on this product or for a direct competitor.
2. They have no first-hand experience of [the thing],
only opinions about it.
3. They are close friends or relatives of the team,
or the interviewer manages them.
4. They are an investor in or adviser to the company.
NICE TO HAVE (use to build a spread, not to exclude)
1. [e.g. a one-person business, and a team of ten or more]
2. [e.g. customer for under 3 months, and over 2 years]
3. [e.g. two regions]
INCLUDE ON PURPOSE (at least [number] of your invitations)
- People who stopped, paused or downgraded
- People who started and did not finish
- People who looked at it and chose something else
SCREENING QUESTIONS (if your records cannot tell you,
and always for an open call)
Q1. Tell me about the last time you [behaviour].
When was it, and what happened?
Q2. What did you use, or who did you ask for help?
Q3. [One question about their situation that
separates fit from non-fit.]
PASS: Q1 describes a real, recent occasion in their own
words, and Q3 matches a must-have.
Two details matter. First, ask for the last occasion, not a yes or no. NN/g recommends3 that ask people to describe the last time they did something, and advises against relying on yes or no questions. The reason is simple: a story is hard to fake, while "do you do X?" is easy to answer yes to. Second, do not reveal what answer you hope to hear. Eager people2 and answer so they seem to match it, so keep the screening questions neutral. Naming the behaviour in an invitation is fine when you have already chosen the people from your own records. In an open call, such as a post on your site or social account, keep the post broad and let the screening questions do the sorting.
For a small study, the screening can be a five-minute call or a two-line reply. You do not need a form.
Who your list leaves out
The people you can reach most easily are not a fair slice of the people you need. Three groups go missing.
The always-available superfans. They are on your newsletter, they answer every survey, and they are glad to help. One or two make a fine participant. As the whole list, they are a trap. Decide in advance how many invitations go to people with strong opinions and no recent behaviour to describe, and keep that number small.
People who stopped. The customer who quit last spring knows something your loyal customer cannot: what made leaving the easier choice. Add them on purpose, and write the invitation so they know their reasons are the point. The same goes for people who signed up and never started, and people who compared you and chose someone else. If your records cannot find them, ask frontline staff who remember them.
People unlike each other. When you want to learn what holds across very different situations, pick participants who differ on the traits that matter. In the research literature this is called7: you choose cases that differ, then look for patterns that appear across all of them. Your "nice to have" list is where you set those differences.
The principle behind all three is to choose people for what they can teach you about the decision.7 is the name for selecting people who are especially knowledgeable about, or experienced with, the thing you are studying, rather than whoever turns up.
Choose channels for who they reach and who they miss
There is no best channel. Each one reaches one kind of person and quietly skips another. Pick channels to cover your must-haves and your include-on-purpose groups, and use at least two.
| Channel | Who it reaches | Who it misses | Effort |
|---|---|---|---|
| Personal email from someone they know (owner, account manager) | Current customers with a relationship | Quiet, lapsed or new customers who do not know the sender | Low |
| Message inside your product or on a receipt | Active users | People who stopped, people who never started | Low |
| Frontline staff ask in person or by phone | People who visit or call | Online-only customers, anyone who avoids staff | Medium |
| Email or text to lapsed and cancelled customers | People your active channels cannot find | Anyone whose contact details are out of date | Medium |
| Introduction through a partner, reseller or franchisee | People you have no direct line to | Anyone the partner does not choose to introduce | High |
| Internal call for your own people (branches, reps, agents) | Staff with first-hand experience of the work | Staff who doubt their answers will stay private | Medium |
| A short post or form on your site or social account | Anyone willing to volunteer | Busy people; expect the keenest to dominate | Low |
Channel choice changes the answers, too. When you recruit from your own network,4 that people already have a relationship with you and may feel reluctant to give honest negative feedback. So have someone other than the account owner send the invitation, and say plainly that criticism is the most useful thing they can give.
The same applies, with more force, when you interview your own team. Assume people who report to the person running the study will be careful about what they say, and do not try to coax candour out of them yourself. Have a neutral colleague invite them and run those conversations, and promise only what you can deliver about who sees what.
Before you contact anyone, check you are allowed to. GOV.UK says to make sure you have users'1 to take part in specific research. Follow your organisation's consent and privacy rules on how you use contact details.
An invitation that is easy to say yes to
A yes needs four things. The person must know why you are asking, how long it will take, what will happen, and what they get for it. NN/g's consent guidance asks for5. Write for someone reading on a phone between jobs.
Be specific about the purpose, but do not give away the answer. "We are deciding how to change the booking process" is fine. "We think our booking process is confusing, do you agree?" is not.
Subject: 30 minutes on how you [did the thing]?
Hi [first name],
I'm [your name] from [organisation]. We are deciding [the decision in one plain sentence], and we want to learn from people who [recent behaviour, e.g. "booked a repair this year"].
Could you give us 30 minutes by [phone or video] in the next [two weeks]? We will ask about the last time you [behaviour]. There is nothing to prepare and there are no right answers. Criticism is the most useful thing you can tell us. [If they stopped: You do not need to be a current customer. Your reasons are exactly what we want to hear.]
What happens: we talk, I take notes [and record only if you say yes]. We use what we learn to inform [decision]. [One sentence on how you will handle their details, following your organisation's rules.] You can skip any question or stop whenever you like.
As a thank-you, [your thank-you].
Reply with two times that suit you, or pick one here: [booking link]. If it is not for you, just say so and we will not chase.
[Name, role, contact]
Short text message
Hi [first name], it's [name] from [organisation]. We're deciding [decision, short] and would value 30 minutes of your time to hear how you [behaviour]. No sales, no prep. [Thank-you.] Interested? Reply YES and I'll send times. Not for you? Reply NO and I won't ask again.
Follow-up reminder (send once, a few days after the first message)
Subject: Re: 30 minutes on how you [did the thing]?
Hi [first name], a quick nudge in case this got buried. I'm still looking for 30 minutes to hear how you [behaviour], and I can fit around your day. Reply with any time that suits, or "no thanks" and I'll close this out.
Every choice in these messages is deliberate. The ask is small and has an end. The person is told which experience matters, so they can judge whether they fit. There is an easy no, which makes the yes more honest. The thank-you is stated up front rather than sprung at the end. Choose a form of thanks your organisation allows and can deliver reliably, and do not promise one you cannot.
Plan for no-shows, and keep a log
People cancel. Plan for it. NN/g's advice is to6 against the same screening criteria as the main group, so a late cancellation does not leave a hole in your schedule. Its recruiting guide adds that4 who just missed the criteria can serve as backups.
For a study of eight conversations, invite more than eight and hold one or two reserve slots. Do not go further. Do not send a hundred invitations and then turn willing people away; those are the customers you will want to ask again. Send in small waves, check who you have, and stop when your spread is covered.
Keep a simple log. One row per person, and nothing you do not need:
| Name or ID | Group (current, stopped, never started, partner, staff) | Channel | Invited on | Reply | Date booked | Outcome (done, no-show, declined) | Notes |
|---|
The log does three jobs. It shows whether your spread holds: after eight interviews you can see that all of them came from one channel. It stops you inviting the same person twice. And it records who declined or never replied, which is a result in itself. If nobody who stopped will talk to you, say so when you report what you found. Follow your organisation's rules on how long to keep contact details.
Worked example: a physiotherapy software firm
This example is invented, to show the method. A small company sells scheduling software to physiotherapy clinics. The owner must decide whether to build automatic appointment reminders. She writes the decision at the top of the page and fills in the screener.
- Must have: managed the clinic's booking in the last six months; dealt with a missed appointment in the last three months.
- Must not have: works on this product or for a direct competitor; has no first-hand experience of booking, only opinions about it; is a close friend or relative of the team, or managed by the interviewer; is an investor in or adviser to the company.
- Nice to have: a one-room clinic and a multi-site practice; customers for under a year and for over three years.
- Include on purpose: at least two clinics that cancelled, and one that signed up and never finished setting up.
Her instinct is to email her ten happiest customers. Instead she leaves them out on purpose and builds a list of fifteen. Six are current customers, picked for the spread. Four are lapsed customers pulled from the cancellation records. Two never finished setup. Three are clinic managers introduced by a reseller she has met once.
The support lead sends the invitations, not the owner, so nobody has to be polite to the boss and the lapsed clinics are not writing back to the person they chose to leave. The current customers get the email. The lapsed and never-started clinics get the text message. The reseller sends his introductions himself.
Nine people reply. Seven book. Two cancel, and her reserve slots cover one. She ends with six conversations: three current, two lapsed, one who never started. Both lapsed clinics describe the same trouble in the first week: they could not import their existing diary, and they gave up before any reminder could matter. The current customers never mention it, because they worked around it long ago. Two clinics are a lead, not a count, so she checks it with the clinic that never finished setup, who confirms it. She would not have heard this from the happy ten. She moves fixing the first week of setup ahead of reminders, and plans to ask more of the next wave about it.
What this will not tell you
A list built this way is a purposeful sample, not a representative one. It gives you good accounts from people who can speak to your decision. It does not tell you how many of your customers share an experience. If you need counts, you need a survey on top, and a different kind of sample.
Assume the people who say yes are not typical of the people who do not, however carefully you choose them. Expect a lean towards the helpful, and read what you hear with that in mind. Every group on your list will have members who never reply, so say who is missing when you share your findings.
Recruitment also cannot prove that the people you spoke to remember accurately. A screener confirms they did the thing recently. It does not make their memory perfect.
Finally, the rules about how you may contact people, store their details and thank them belong to your organisation, your country and your sector. This guide does not cover them. Check your own before you send the first message.
Your next step
Today, write the decision and fill in the must-have section of the screener, then check it against your list. Mark each name with a group, count how many of the people who left or never started are on it, and send the first wave of invitations through two channels. Once the calls are booked, see customer interviews.
