Anchor everything to the decision. 'Should we launch the ₹10 pack?' is researchable. 'Let's understand the youth' is a budget with no brakes. Working backwards from the decision tells you how much rigour to buy: a tagline test doesn't deserve six weeks, and a ₹50-crore launch doesn't deserve a Google Form. The cost of the research should scale with the cost of being wrong.
The three learning modes have different superpowers. Asking is the only way into meaning: why something matters, what it signifies, what buyers fear. But it's polluted by memory, politeness, and self-image. Watching (shop-alongs, home visits, analytics, sales data) captures true behaviour but not its reasons. Testing, the youngest mode, is the only one that proves causation: change one thing, hold the rest, watch what happens. Serious research questions usually need two modes triangulated: ask to generate hypotheses, watch or test to validate them.
Sequence qualitative before quantitative. Interviews and observation on a small scale surface what matters and the words customers use for it. Surveys then measure how much and how many, in those same words. Invert the order, and you launch a big survey before anyone has sat with ten real customers. That produces statistically precise answers to the wrong questions, which is worse than no research because it arrives wearing confidence.
Respect the say-do gap as a law, not a nuisance. Buyers overstate future virtue (healthy eating, higher spending, trying new things) and understate habit's grip. They can't reliably report their own price sensitivity or how an ad worked on them. So calibrate by question type. Behaviour and money questions deserve behavioural evidence, what people did or what they do in a test, while feelings, associations, and language genuinely need asking. New Coke is the eternal reminder: the taste tests answered the sip question correctly and the meaning question not at all.