Two kinds of question call for two different kinds of research. Qualitative research — focus groups, depth interviews, ethnography — is unstructured and exploratory, used on small, deliberately chosen samples to build insight and understanding rather than to generalise to a population (Malhotra, Nunan and Birks, 2020). It is the right tool when the question is "why": why do health-conscious professionals hesitate over price, what do they actually mean by "healthy", what story would make them switch. The focus group itself has a long history outside marketing — Merton and Kendall (1946) developed the "focused interview" to study US Army morale films, and it was Calder (1977) who set out how a focus group's open, group-dynamic format suits exploratory marketing questions but is poorly suited to anything that depends on counting or generalising.

Quantitative research — surveys, experiments, behavioural data — is structured, uses large and ideally representative samples, and answers "how many" and "how much": how many buyers would switch, how much more would they pay, how does willingness to pay vary by segment (Malhotra, Nunan and Birks, 2020). Its strength is generalisability with a stated level of confidence; its weakness is that it can only measure what the researcher thought to ask.

The common mistake is reaching for the wrong tool: running a 600-person survey to find out why people feel hesitant, a question a dozen good depth interviews would answer more cheaply and more richly, or running three focus groups to estimate what proportion of the market would pay a premium, a question no qualitative sample, however insightful, can answer with any precision. Many real decisions need both, in sequence — qualitative work to find the right questions, quantitative work to size the answer (Creswell and Creswell, 2018) — which is the logic Birchwater follows in this module's worked example. The method choice should follow from the decision in front of you, not from habit or budget.

A piece of market research is only as good as the brief that commissioned it. Hague (2021) defines a research brief as the sponsor's statement of objectives and background, detailed enough for a researcher to design an appropriate study, and argues that market research exists to reduce the risk in a specific business decision, not to satisfy general curiosity. The discipline that follows is simple to state and easy to skip: before anything else, name the decision the research will inform. "Which of our three candidate segments should we target, and what do they value most?" is a decision. "What do customers think of us?" is not; it has no action attached, and a researcher asked to answer it will produce an expensive summary that nobody can use.

Once the decision is named, a workable brief sets out: the objectives, the specific questions that, answered, let you make the decision, which Malhotra, Nunan and Birks (2020) frame as translating a management problem into a research problem; the method, qualitative, quantitative or both, and why, following the logic of the previous lesson; the sample, who, how many, and how they'll be selected, developed further in the next lesson; and a timeline and budget realistic enough that the answer arrives before the decision has already been made by default.

Two failure modes are worth naming. A brief with no decision attached produces research for its own sake. A brief with the method decided before the objectives are clear, "let's run a survey", often turns out to be asking a quantitative instrument to answer a qualitative question, or the other way round. Writing the decision first, and the method last, guards against both.

Research produces two very different kinds of output, and conflating them is a common way a study fails to earn its budget. A fact restates something already observable: "42% of our survey respondents are aged 25–34" or "health-conscious professionals buy flavoured water more often than other segments." Both are true, and neither tells you what to do next. An insight reframes the problem or reveals something not previously obvious — a reason behind the pattern that changes the decision. Zaltman (2003) argues that most market research stops at the level of what customers say they think, because surveys, questionnaires and even focus groups depend on people being able to articulate reasons they often don't consciously have access to; genuine insight usually requires pushing past the first, comfortable answer.

The distinction becomes concrete with Birchwater's own data, developed fully in this module's worked example. The fact is that health-conscious professionals buy flavoured water frequently — useful, but it only tells you who already buys. The insight is that this same segment over-indexes on willingness to pay a premium and under-indexes on price sensitivity: they are not just frequent buyers, they are the least price-driven buyers in the market, which is a reason to build a premium strategy around them rather than compete on price. The same dataset supports the fact and the insight; the difference is whether the finding changes the decision in front of you.

Combining qualitative and quantitative evidence makes a weak insight harder to fake. Triangulating a quantitative pattern against the qualitative reasons that explain it, the logic behind sequencing methods in Lesson 2.1, is what lets you distinguish a real insight from a number dressed up as one (Creswell and Creswell, 2018).

Before acting on a piece of research, your own or someone else's, check three things: the sample, the method and the limits the researchers themselves would admit to. Representativeness matters more than raw size: a survey of 600 respondents who are not representative of the population you care about is weaker evidence than a smaller sample drawn and weighted correctly. Groves et al. (2009) frame the risk through total survey error: coverage error, the parts of the population that had no chance of being sampled; sampling error, the natural variability of drawing a sample rather than measuring everyone; and non-response error, the people who didn't answer may differ systematically from those who did. All three combine, and a large sample size only controls one of them. A sample "representative by age and region", as Birchwater's survey is, addresses coverage on those two dimensions specifically; it says nothing about representativeness on income, usage frequency or any dimension nobody thought to quota against.

Method matters as much as sample. A leading question, "how much do you agree that our healthier recipe deserves a premium price?", will inflate agreement regardless of what respondents actually think; Fowler (2014) sets out the standard that a good question produces the same stimulus for every respondent and precludes alternative readings. Self-reported measures carry a related risk: people systematically over-report healthy, high-status or generous behaviour and under-report the opposite, whether consciously, through impression management, or not, through self-deceptive enhancement (Paulhus, 1984); a serious concern for a survey asking people how much they'd pay for a "healthy" product.

Finally, resist treating any single study as definitive. A 95% confidence level means that if you ran the same survey 100 times, about 95 of the resulting intervals (each result plus or minus its margin of error) would contain the true population value, not that this one result is certain. State the confidence level and margin of error alongside any headline finding, and treat a result that would change a major decision as a reason to replicate it, not a reason to stop looking.