Most organisations describe their customers as they imagine them. Real decisions are messier, and the gap between the two is where marketing money gets wasted. Buyer behaviour is how people and organisations recognise a need, search, compare, buy and judge what they bought.

The classic way to describe it is the consumer decision process: a need (my coffee has run out), a search (what are my options?), evaluation (which suits me best?), the purchase, and what happens after purchase (was it worth it, will I buy again, will I tell anyone?). Treat it as a simplification, not a law. People skip stages, loop back, and make many buying decisions with almost no search at all.

That's because a lot of buying runs on habits and mental shortcuts. We buy the same brand because it worked last time, choose the option already ticked, or read a higher price as a sign of quality. Daniel Kahneman's Thinking, Fast and Slow (2011) describes fast, intuitive thinking alongside slower, deliberate thinking. You don't need a figure for how often each happens; you need to notice which one your customers are probably using when they buy from you.

Business buying adds people. A business purchase usually involves a decision-making unit: the user who will use the product, the buyer who places the order, the approver who signs it off, and influencers whose opinions count. In Larkfield Coffee Roasters, the fictional business this course follows, a café's decision to buy coffee involves the owner and the head barista. Market only to one of them and you can lose the sale to the other.

UK example: Tesco Clubcard. Tesco launched Clubcard in February 1995. It worked with dunnhumby, the data science firm founded by Edwina Dunn and Clive Humby, to analyse what members actually bought. The lesson for you is the principle, not the scale: what customers do (what they buy, when, how often) is often better evidence than what they say they'll do. Most organisations hold some behaviour data already, such as order histories, bookings or attendance, and asking questions works best alongside it.

When you sketch how your organisation's customers decide, mark each step as "known" (you have evidence) or "assumed" (you're guessing). The assumed steps are your research questions for the next lesson.

Good research starts with a question, not a survey. A useful research question is narrow and tied to a decision: "Why do customers cancel in their first three months, so we can decide what to change?" is far better than "What do customers think of us?". If no decision would change whatever the answer, it isn't worth researching. Choosing which question is worth your time is a first piece of marketing leadership.

Secondary first. Before you ask anyone anything, check what's already known. In Module 1 you built a source log; now go back to it. Public reviews, social comments, complaints, sales records and published reports often answer half the question for free. Write down what they show and, just as important, the gaps they leave. Those gaps are what your own research is for.

Primary research is research you carry out yourself for your question. It comes in two broad kinds, qualitative and quantitative:

• Qualitative research explores reasons and meanings through words, for example interviews or open comments. It tells you why.
• Quantitative research measures how many or how much, for example a survey with closed questions. It tells you how common.

Most small projects use both: a few interviews to find the reasons, and a short survey to see how widespread they are.

Samples. You won't ask everyone, so you ask a sample. A small sample can still be useful if you say honestly what it can't show. Five interviews can reveal reasons you hadn't thought of, but they can't tell you what share of customers feel the same. A survey of your own customers tells you about them, not about the whole market. Write that limitation into your plan before you start, not after.

Consent and anonymity. People should know who is asking, why, what will happen to their answers, and that they can say no or stop at any time. Ask for no more personal data than you need, and report results so no one can be identified. The Market Research Society's Code of Conduct (revised 2023) sets the standard UK researchers follow. The legal rules on personal data return in depth in Module 5.

The wording of a question can decide the answer. Small changes make a big difference, so this lesson is about writing questions that let people tell you the truth.

Closed questions give a fixed set of answers (yes or no, a list, a rating scale). They are quick to answer and easy to count. Open questions ("What nearly stopped you ordering?") let people answer in their own words. They are slower to analyse but often surprise you. A good survey is mostly closed questions with one or two open ones. Rating scales need balanced ends, such as "very poor" to "very good", with a middle option.

Order matters. Start with easy, general questions, then move to the specific and the sensitive. Put personal questions, like age, at the end and only if you need them. Earlier questions can colour later ones, so ask about overall satisfaction before you ask about one feature.

Three faults spoil most first drafts: leading, double and vague questions.

• A leading question pushes people towards an answer: "How much do you love our new range?"
• A double question asks two things at once: "Was delivery fast and well packed?" Someone who got a fast, damaged parcel can't answer.
• A vague question means different things to different people: "Do you buy coffee regularly?" Weekly? Daily?

Interviews use an interview guide: five or six open questions in a sensible order, plus prompts such as "Tell me more about that" or "What happened next?". Ask about real past behaviour ("Tell me about the last time you...") rather than guesses about the future.

Piloting means trying your questions on one or two people before the real thing. Watch where they hesitate, re-read or ask what you mean. Every pause is a question to fix.

AI activity. Ask an AI assistant to draft a short survey for your organisation's research question from Lesson 2.2. Then go through it line by line and find every leading, double or vague question it wrote, rewriting each one. Finally, note what it missed about your customers: the AI doesn't know who they are, how they buy or what your source log showed. Keep both versions; the comparison is useful evidence of your judgement.

Once the answers are in, resist the urge to report everything. Your job is to find what matters and what your organisation could do about it.

Start simple. For closed questions, count: how many people chose each answer, and out of how many? Always give the base ("39 of 96", not just a percentage), so readers can see how big the group was. For open comments and interviews, read everything once, then group similar comments into themes and count how many fall under each. Pick one or two short quotes that capture each theme in customers' own words, with no names attached.

A finding is not an insight. A finding is what the data says: "Most cancellations mention having too much coffee." Customer insight is a fresh, evidence-based understanding of why customers act as they do that the organisation can act on. It explains the finding and points to a decision.

A simple way to write one is the insight statement: "We noticed... (the finding), because... (the reason, from your evidence), so we could... (an action)." If you can't fill in the "because" from your evidence, you have a finding and a guess, and the guess is your next research question. If you can't fill in the "so we could", the insight isn't yet useful.

Say what it can't show. Every insight comes with limits. Who did you hear from, and who didn't you hear from? Did happy customers reply more than unhappy ones? Is it your customers or the whole market? Writing one honest line of limitations makes your insights more credible, not less.

Read the worked example later in this topic before you start the task: it takes Larkfield's survey all the way from a count to an insight statement. Then use the same steps on your own results or, if you can't survey anyone, on published reviews and comments about your organisation.