Two ways of knowing
Qualitative research is any method where the evidence is words, behaviour and meaning rather than counts: one-to-one interviews, group discussions, observation of people using a product, diary studiesA qualitative method in which a small set of people record, at the time it happens, what they do during an episode that unfolds over days or weeks., open-ended feedback read and coded by a person. Its unit of analysis is the individual case, and its output is themes, mechanisms and vocabulary. It answers why, how and what else is going on here. A good interview, as the customer interviews tutorial shows, is built to let the participant lead, which is how the unexpected surfaces. Group discussions add the influence participants have on each other, and the tutorial on running a focus group shows how private written reactions and analysis by group keep that influence from becoming the finding.
Quantitative research is any method where the evidence is numbers from a defined population: surveys with closed questions, behavioural data from analytics and transactions, panel measurements, and controlled experiments such as an A/B TestControlled comparison of two variants.. Its unit of analysis is the sample, and its output is estimates, distributions and comparisons. It answers how many, how much, how often and is the difference real. The Market Research Society's guidance on both families rests on the same professional standards (informed consent, honest reporting, methods stated with their limits), but the standards of evidence differ: a qualitative finding is credible when it is grounded in what participants said and did; a quantitative finding is credible when the sample stands for the population and the instrument measured what it claims to.
The two are often described as a choice. They are better described as a division of labour, and the research design guide places the choice where it belongs, after the decision and the questions are written.
Compare table
Qualitative research vs Quantitative research
| Row label | Qualitative research | Quantitative research |
|---|---|---|
| Core question | Why, how, what matters, what did we miss | How many, how much, how often, which is larger |
| Typical methods | Interviews, group discussions, observation, usability sessions, diary studies, open-text coding | Surveys with closed questions, analytics and transaction data, panels, controlled experiments |
| Sample | Small and purposive: chosen to cover the cases that matter | Larger and structured: drawn to stand for a defined population |
| Output | Themes, mechanisms, vocabulary, hypotheses, verbatim quotes | Estimates, proportions, distributions, comparisons, confidence intervals |
| Strength | Surfaces the unexpected; explains behaviour; captures language | Measures prevalence and size; supports comparison across groups and over time |
| Weakness | Cannot say how common anything is; interviewer and selection effects are strong | Can only measure what was asked; precise about the wrong thing if the questions are wrong |
| Cost and speed | Cheap to start, slow to analyse; a dozen interviews take weeks to run and code | Expensive to field well, fast to analyse once the instrument exists |
| Skill that matters most | Asking without leading; listening; disciplined synthesis | Instrument design; sampling; reading statistics without over-reading them |
| Failure mode | Anecdotes treated as findings; the loudest participant sets the story | Bias dressed as precision; a survey that confirms the sponsor's assumptions |
| Typical place in a programme | Early, to find the questions; late, to explain surprising numbers | Middle, to measure what the early stage found; tracked over time for a KPIA metric tied directly to an objective. |
| Best for | Proposition development, message testing, discovering jobs and obstacles, understanding churn | Sizing segments, pricing research, brand tracking, testing a change on a FunnelModel of the customer journey stages. |
Where qualitative wins
When the team does not yet know what to ask. A survey can only measure the options its author imagined. Interviews find the options. Before any proposition, pricing or segment study, a qualitative stage reveals the categories people actually think in, the alternatives they actually weigh and the words they actually use, and the survey that follows is built from them rather than from the team's assumptions. Where the behaviour unfolds over days or weeks, the tutorial on running a diary study has people record each moment as it happens, before memory tidies it away.
When the question is about meaning or mechanism. Why people churn, why a message falls flat, why a product is used for something it was not designed for: these are stories with causes, and only a conversation can follow a cause back to its source. The jobs-to-be-done lens is the clearest example. Asking someone what they were trying to accomplish when they bought, what they tried before and what nearly stopped them produces an account of the purchase that no closed question can. The same holds for a checkout or sign-up form that people abandon: watching people attempt it, as the tutorial on running a usability test on a conversion journey sets out, shows where and why they stall in a way no funnel report can.
When the audience is small, hard to reach or expert. Senior buyers in a niche category, specialist clinicians, procurement leads: there may be only a few dozen people whose view matters, and a survey would reach ten of them. A dozen well-chosen interviews cover the population better than any sample could.
When the budget is small and the decision is reversible. A handful of interviews costs a few days and a few incentives. If the decision can be revisited, that is often enough evidence to act on, and the measurement stage can wait until the stakes rise.
Where quantitative wins
When the decision depends on how many. Whether a segment is large enough to serve, what share of customers would pay more, how far a proposition travels beyond the enthusiasts who turned up to interviews: these are prevalence questions, and prevalence can only be measured. A qualitative finding that "customers want flexibility" says nothing about whether that is a majority or a vocal minority.
When the business needs to compare or track. Brand health across markets, satisfaction across segments, awareness quarter on quarter: comparison needs the same instrument applied to comparable samples, and that is a quantitative design by definition. Conversion rate (CVRShare of visits or clicks that convert.) by source, or the step-by-step losses in a funnel, are quantitative evidence the business already holds, and the conversion research methods guide shows how behavioural data of that kind sits alongside interviews and surveys when the question is why a page loses people.
When the question is causal. Does the new pricing page lift sign-ups? Does the campaign raise enquiries? Only a controlled comparison answers that, and the design, sample size and reading of a controlled test are a discipline of their own, covered in the sample size and significance guide. The marketing research point is simply that interviews cannot establish cause, however vivid.
When the finding has to persuade. Rightly or wrongly, a proportion from a defined sample carries more weight in a board paper than a set of quotes. Qualitative evidence persuades the people who heard it; quantitative evidence travels.
Sequencing: explore, then measure, then explain
The strongest programmes treat the two families as stages rather than rivals, and the order is consistent.
Explore. Start with secondary evidence and a small qualitative stage. Interviews, observation or a read of open-text feedback establish what the decision really turns on, which options exist, what language the audience uses and which assumptions the team held that do not survive contact with customers. The output is a set of hypotheses and a vocabulary.
Measure. Build the quantitative instrument from that output. Every closed question in the survey has a reason to exist that traces back to the exploratory stage; every answer option is phrased in words a participant used. Field it to a sample designed to stand for the population the decision concerns. The output is the size of each finding and how it differs across groups. The tutorial on turning interview findings into survey questions walks through the hand-off: choosing which findings a survey should measure, writing the options in participants' words and tracing each question back to the interviews.
Explain. Numbers produce surprises: a segment that behaves against expectation, a proposition that polled well with one group and badly with another, a satisfaction score that moved without obvious cause. A second, short qualitative stage aimed at the surprise explains it, and that explanation is often the most valuable finding in the programme. It also closes the loop for the next cycle, because the explanation becomes the next hypothesis to measure.
The sequence can compress. A small team with two weeks can run six interviews, write a short survey and field it to its own customer base inside that window. What it should not do is run the survey first and the interviews afterwards to decorate the deck, which is the most common inversion and the least useful.
Mixed methods in one study
The sequence is easiest to see in a single study where all three stages serve one decision.
A language-learning app decides what to build next
Lingbury, a language-learning app with about 140,000 monthly active users, has to choose between building live tutor sessions and building a structured exam-preparation track. The product marketer runs an exploratory stage first: fourteen interviews, split between long-term subscribers and people who cancelled in the past three months. Interviews surface a job the team had not considered: a large group of users are preparing for a citizenship or visa language test and cancel the moment they pass. The measurement stage is an in-app survey built around that finding, answered by 3,900 users. It shows that 41% of respondents are learning for a formal test, that this group rates exam preparation far above live tutoring, and that they churn at roughly twice the rate of hobby learners. An explaining stage of eight further interviews with test-takers finds that most would stay subscribed after passing if the app offered a maintenance mode at a lower price. Lingbury builds the exam track first and adds a reduced-price maintenance plan, and the whole programme runs in seven weeks for under £6,000 in incentives and survey tooling.
The same study run as a survey alone would have asked users to choose between tutoring and exam preparation, measured a preference, and missed both the test-taker segment and the maintenance plan that turned out to be the larger opportunity. Run as interviews alone, it would have found the segment and been unable to say how large it was. The methods section collects the design guidance that makes the sequence routine rather than exceptional.
Common mistakes
- Choosing a method before writing the question. "We need focus groups" or "let's run a survey" as the first sentence of the brief.
- Counting interviews. Reporting that "seven out of twelve participants" wanted something, as if a purposive sample of twelve could estimate a proportion.
- Surveying before exploring. Closed questions written from the team's assumptions, measured precisely, wrong.
- Treating open-text survey answers as interviews. A few hundred one-line comments lack the follow-up that gives interviews their depth.
- Letting one vivid participant set the story. Qualitative synthesis is a discipline, not a memory of the most quotable session.
- Reading a survey as if the sample were the population. Bias does not disappear because the response count is large.
- Skipping the explain stage. Surprising numbers reported without the conversations that would have made them actionable.
- Expecting interviews to establish cause. A participant's account of why they bought is evidence of their reasoning, not of what would happen if the business changed something.
