How to choose a sampling method for customer research

Imagine your bank plans to introduce a new fee on credit cards. You want to know how many customers might stop using their cards and why. But asking every customer would take too much time. So, who should you ask?

The answer depends on what you want to learn. If you want to estimate how many customers might leave, you need a sample that represents all your customers. If you want to understand why they might leave, you need to hear from customers who can explain their reasons.

Picking who to ask is called sampling. To choose the right sampling method, start with three simple questions: What do you need to learn? Who can you reach? And do some groups need their own answer?

The short version

• If you need to know how many customers will leave, pick people at random from a full list of your customers.
• If you need to know why they will leave, pick people on purpose who can explain their reasons.
• Check your list first, because customers who are missing from it can never be picked.
• If one small group needs its own answer, pick extra people from that group.
• Check where survey answers come from, because fake answers can spoil even a good method.
• Big decisions often need both: interviews to find the reasons, then a survey to count them.

Three questions that point to the right method

Each question narrows the next one, so it helps to answer them in order.

1. What do you need to learn?

Go back to the bank. Your manager might ask, "What share of customers would leave?" That needs a number. Or your manager might ask, "Why would customers leave?" That needs reasons.

A number and a list of reasons need different samples.

For a number, everyone on your list needs a fair chance of being picked. Then the number you find stands for all your customers. This is called a probability sample. It also tells you how close your number is likely to be to the real one, a range called the margin of error.

For reasons, a random pick matters less. You want customers who have something to say, such as people who already cancelled their card or people who complained about fees.

2. Who can you reach?

You can only pick people from a list you have. That list is called your sample frame, and anyone who isn't on it can never be picked.

If your bank's customer database holds every cardholder, you can pick from all of them at random. Other lists leave people out. A list of customers who agreed to take surveys leaves out everyone who didn't agree. A list of customers your sales team knows leaves out everyone else.

So if your list leaves people out, your number only speaks for the people on the list.

3. Do some groups need their own answer?

Say only a few of your customers hold a premium card, and your manager wants a separate answer for them. If you pick at random from all customers, you might get only a handful of premium cardholders. That's too few to say anything about them.

So decide before you start how many people you need from each group, such as each card type or each country. Then pick extra people from the small groups.

Two kinds of sampling methods

Sampling methods come in two families. One helps you count, and the other helps you explain.

Probability methods give everyone on your list a fair, known chance of being picked. Use them when you need a number that stands for all your customers.

Non-probability methods let you choose who to ask. Use them when you need reasons. They can't give you a number that stands for all your customers.

Four ways to pick at random

If you have a full list of customers, there are four common ways to pick from it at random.

• Simple random sampling gives every customer the same chance, like drawing names from a hat.
• Stratified sampling splits your customers into groups, such as card types, and picks from each group.
• Systematic sampling picks at a fixed gap, such as every 20th customer on the list, after a random start.
• Cluster sampling picks whole groups first, such as bank branches, and then asks customers inside them.

Four ways to pick on purpose

When you choose who to ask, four methods are common.

• Purposive sampling picks people because they can tell you what you need to know, such as customers who just cancelled.
• Quota sampling sets how many people you need from each group, then takes whoever fits first.
• Snowball sampling asks each person to suggest the next one, which helps when people are hard to find.
• Convenience sampling asks whoever is easiest to reach, like asking only the people sitting near you.

Say only what your sample can show

The difference matters most when you write up what you found. Say you talked to customers who cancelled, and most of them mentioned fees. Here are two ways to report it.

Before: "Most customers who cancel do so because of fees."

After: "Among the customers we talked to who cancelled, fees came up most."

The first sentence talks about all customers, and your interviews can't show that. The second says what you found, and among whom.

Which method fits which question

This decision table matches each question to a sampling method. Each row starts with a question a team might ask, then shows the method that fits and the mistake to watch for.

Your question	Method that fits	Watch out for
How happy are all our customers?	Simple random, from a full customer list	Customers missing from the list, such as those who joined through a partner
How happy are customers with each card type?	Stratified, with extra people from small groups	Before you give one overall number, adjust each group back to its real size
How do customers rate our support?	Systematic, every 20th closed support request	A list sorted by support agent can put every pick on one agent
How do customers find our branches?	Cluster, whole branches first	Customers of one branch tend to be alike, so each answer tells you less
Why did customers leave last quarter?	Purposive interviews with people who left and people who stayed	Reporting percentages the interviews can't support
What do people think of a new idea?	Quota or convenience, as a first look	Treating it as the view of all customers
How does a company decide what to buy?	Purposive, then snowball	Every suggestion coming from one circle, so start from several people
How do people see our brand each month?	A panel picked at random	How good the panel is, and how it removes fake answers

Many of these questions work best as a pair: a survey to count, and interviews to explain. Our post on qualitative and quantitative research covers how the two fit together.

Check where your answers come from

Say you design a careful survey and buy answers from an online panel, a group of people who signed up to take surveys. If many of them don't answer honestly, your careful design still gives you wrong results. It's like a good recipe made with stale ingredients.

Many quick surveys use opt-in panels, which anyone can join. Two things decide how far you can trust them: who is in the panel, and how it removes fake answers. Fake answers here means answers from people who aren't honest or aren't paying attention.

Pew Research Center tested this in an online opt-in survey of 11,114 US adults, run on Nov 14-19, 2024. It tried three ways to remove fake answers: trap questions, a paid screening service, and checking people against public voter records.

In its August 2026 report on that survey of 11,114 adults, Pew found that removing fake answers made most results more accurate. But checking people against voter records made results slightly less accurate, because it removed mostly honest people.

A March 2024 Pew Research Center analysis compared three online opt-in samples with three panels picked at random. Young and Hispanic people in the opt-in samples were much more likely to answer "yes" to 10 or more of 16 yes/no questions. Very few people in the random panels did that.

How a panel picks its people matters too. A 2024 review in the Oxford Research Encyclopedia of Criminology found that panels with more careful ways of picking people tend to give more accurate results.

Three questions to ask a panel company

Before your survey goes out, ask the panel company three questions. Each answer tells you how far to trust what comes back.

1. How do you find people for your panel? This tells you who could ever end up in your sample. A good answer names the method, such as letters sent to home addresses or open sign-up through ads, and says who it misses.
2. How do you catch fake answers? Pew's study shows that the way you remove people changes the result. A good answer covers checks before the survey and checks on the answers, such as a trap question that says "Choose 'somewhat agree' for this row."
3. How many finished surveys do you throw out? A good answer gives the share for a recent survey like yours, and the reason. If the company can't say, treat its results as a first look.

If you have a full list of your own customers, it's often the better place to start, because you already know who is on it.

One bank question, two samples

Back to the bank and the new credit card fee. The bank needs to know how many customers would leave, and why. Those are two questions, so it needs two samples.

The bank has four options. Each one can show some things about its customers, but not others.

Option	What it can show	What it can't show
A survey of an online opt-in panel	Quick reactions from people who look like customers	What the bank's own customers would do
A stratified survey picked from the customer database	An estimated share of each card type who say they would leave	Why they would leave
Interviews with customers who recently left and customers who stayed	Their reasons, in their own words	How common each reason is
Chats with customers in a branch	First impressions	Anything about customers who rarely visit a branch

The customer database holds every cardholder, so the bank can pick at random from it. The panel adds nothing here.

First, the bank interviews. It talks to customers who recently cancelled their card, and to similar customers who stayed. The interviews bring out reasons such as the size of the fee, a better card elsewhere, or rarely using the card at all.

Next, those reasons become the answer options in the survey.

Then the survey counts. The bank picks at random within each card type, with extra people from the small premium group. It adjusts the premium answers back to the group's real size before giving one overall number.

Now the bank can say what share of customers would leave, with a margin of error, and why. The interviews also show what would change customers' minds, such as ways to avoid the fee. So the bank can decide how to design the fee, as well as whether to charge it.

A simple rule for your next study

Start from what you need to learn. If you need a number and you have a full list of customers, pick people at random from that list.

If you need reasons, pick people who can explain them, and report what you found among those people only.

If you need both, interview first and count second.

Where Pulse fits

Once you've picked a method, you need the right people for both parts of the study. Pulse turns your question into a research plan, including who to talk to and a screener, a short quiz that checks each person fits the study.

It invites your own customers from a list you share, or recruits participants to match that profile. Every participant goes through fraud checks.

For the "why" part, an AI moderator runs voice interviews and asks a follow-up question when an answer is vague. Themes come back with quotes, each linked to its transcript.

For the "how many" part, surveys run alongside the interviews, so their answer options can come from what people said.

To see how both samples would run for your next study, book a demo.

Questions readers ask

Once you've picked a method, most questions are about numbers and shortcuts.

1. What are the four types of probability sampling?

They are simple random, stratified, systematic and cluster sampling. All four give everyone on your list a known chance of being picked, so you can report a number with a margin of error.

They differ in how they pick: from one pool, from each group, at a fixed gap, or whole groups at once.

2. Why not just ask every customer?

Asking every customer takes too long and costs too much, and many customers won't answer. So even if you ask everyone, you end up with a sample: the customers who replied.

A random sample from a full list is faster and cheaper, and you know how far off it might be.

3. How many people do you need in each group?

For a survey, it depends on how exact each group's number must be. A smaller margin of error needs more people, and the sample size calculator works it out for you.

For interviews, there's no set number. Keep going while new interviews still teach you new reasons, then say how many you ran.

4. Can a quota sample replace a random sample?

Only partly. A quota sample matches your customers on the traits you set, such as country or age. It can't match them on anything else, because each group fills with whoever comes first.

So treat its numbers as rough guides, and say so when you report them.