School Me MarketingSMM
FoundationsIntermediate10 min deep read · 30-sec skim available

Research Design & Method Choice

You know the three honest moves: ask, watch, test. Design is choosing which one, on whom, in what order, for this decision, because a brilliant answer from the wrong sample is still the wrong answer. Rigour isn't buying more research; it's matching the method to the mistake you can't afford.

⚡ Understand it in 30 seconds

  • Every research question is one of three species: exploratory (we don't know what matters yet), descriptive (we know what to measure, need how much and how many), or causal (we need to know what happens if we act). Each species has its own methods, and feeding a question to the wrong species produces confident nonsense.
  • Sampling is where studies quietly die: 1,200 of the right people beat 12,000 of the wrong ones, and a convenience sample of your existing fans will cheerfully validate whatever you show them.
  • Instruments leak bias by design: leading questions, order effects, scales that flatter, focus-group alphas who convert eight opinions into one. Craft is knowing where each method bleeds and bandaging in advance.
  • The Intermediate jewel is trade-off measurement: don't ask people what they'd pay. Make them choose between bundles (conjoint analysis) and derive what they value from the pattern of their choices. Behaviour, even simulated, out-testifies opinion.

Go deeper

The core idea

Run one decision through the taxonomy to see it work. The decision: should a cooking-oil brand launch a blended 'heart-healthy' variant at a 20% premium? Exploratory first: you don't yet know what 'healthy oil' even means to households, so depth interviews and kitchen visits surface the vocabulary, the anxieties (a cardiac event in the family reorders everything), and the hidden veto-holders (the cook and the payer may be different people). Descriptive second: now that you know what matters, a structured survey on a proper sample sizes the segments, quantifies stated importance, maps current oil repertoires. Causal last: a price-pack test in matched cities, or a conjoint exercise, measures what people do at ₹X versus ₹Y. Three species, in sequence, each building the next one's questionnaire. Skipping to the survey without the kitchens produces precise measurements of the wrong constructs.

Sampling determines whose truth you learn. Probability samples (everyone in the population has a known chance of selection) earn generalisation. Convenience samples (whoever was reachable, like your app users, mall visitors, that panel again) earn only hypotheses. The killer isn't size but frame and response bias: 500 respondents drawn properly beat 50,000 self-selected app raters, because the app raters are systematically your most engaged, research on survivors presented as research on the market. Every brief should state whose behaviour the decision depends on, and every sample should be audited against that sentence.

Instruments are machines with known failure modes. Surveys: leading wording ('how much do you love…'), order effects (the first brand named anchors the rest), scale games (a 1–10 with no anchor produces national politeness, not preference), and the say-do gap on anything involving money or virtue. Focus groups: brilliant for language, dangerous for verdicts, since one confident voice re-writes the room and groups perform social desirability live. Depth interviews trade group energy for honesty. Ethnography trades scale for truth (what people do in kitchens contradicts what they say in halls). Analytics observe everything and explain nothing. Design means pairing methods so one's bleeding is another's bandage: ethnography generates, surveys quantify, experiments confirm.

Causal questions deserve causal machinery. 'Did the campaign work?' compared against last quarter is astrology: seasonality, distribution changes, and competitor moves all confess to the same correlation. The honest instruments: matched-market tests (act in Indore, hold Nagpur), geo-split and holdout experiments, A/B tests where traffic allows, and conjoint analysis for the pricing and feature trade-offs where asking directly fails (customers want everything and claim to pay nothing). The discipline in one line: the strength of your causal claim can never exceed the strength of your control group.

The business case

Why marketers care

Research budgets are wasted at the design stage, not the fieldwork stage. A perfectly executed survey answering an exploratory-stage question, or a focus group asked to predict a price response, burns money and, worse, mints false confidence that outlives the study by years.

Method literacy is the marketer's defence against being sold. Agencies pitch what they run: panels pitch surveys, labs pitch neuro, consultancies pitch workshops. The buyer who can ask 'what species of question is this, and whose behaviour does the decision depend on?' turns vendors from oracles back into instruments.

And design thinking scales down honestly: the same taxonomy governs a founder's ten interviews and an FMCG giant's national U&A study. Small budgets don't excuse bad design. They make it costlier, because one study is all the evidence the decision will get.

See it

The visual model

Choosing the instrument

Locate the question first: the method falls out of the quadrant, not the vendor's pitch deck.

Read this diagram as text

A two-by-two map for method choice. The horizontal axis runs from exploring what matters to measuring how much. The vertical axis runs from what people say to what people do. Bottom-left, exploring through words: depth interviews and ethnography. Bottom-right, measuring through words: surveys and usage-and-attitude studies. Top-left, exploring through behaviour: analytics and observation. Top-right, measuring through behaviour: experiments, test markets, and conjoint trade-off exercises, the strongest evidence for the costliest decisions.

The receipts

Where it comes from

The exploratory/descriptive/causal taxonomy comes from the marketing-research canon (Green & Tull; Malhotra's standard texts); conjoint analysis enters marketing through Paul Green's work at Wharton (Green & Srinivasan's reviews, 1978/1990).

The taxonomy earns its place by mapping questions to evidence standards: exploration tolerates small, rich, biased samples because it only claims hypotheses; description demands representative sampling because it claims measurement; causation demands controls because it claims prediction. Most research scandals, in business as in science, are category errors: exploratory evidence retailed at causal confidence.

Conjoint's insight was that preference is revealed by trade-offs, not testimony. Force respondents to choose between full profiles (this brand, this feature set, this price) repeatedly, and regression recovers the weight each attribute actually carries, including the rupee value of a brand name, which no direct question can extract. It remains the closest thing research has to reading willingness-to-pay honestly.

The modern layer bolts experimentation onto the old canon: always-on digital A/B testing and geo-holdouts have made causal evidence cheaper than ever, which changes the design calculus. Questions that once justified a survey ('which claim converts?') are now answered better by a week of live traffic. The surveys can retreat to what only asking can reach: meaning, language, and why.

Brands you know

Seen in India

Educational readings of familiar brands — how the concept helps you see what they do, not claims about their current campaigns.

Marico

The FMCG house behind Parachute and Saffola, known for insight-led innovation.

Its innovation pattern reads as sequenced design: in-home and kitchen observation surfacing how oils, oats, and hair care actually get used (exploratory), followed by structured studies sizing the opportunity, followed by regional pilots before national rollouts (causal-by-geography). Format innovations like easy-pour and small packs read as answers to observed friction, not surveyed wishes, the kind of insight kitchens yield and questionnaires don't.

What to steal: Sequence the species: watch to find the question, ask to size it, test to confirm it — and let each stage write the next one's instrument.

Paper Boat

The beverage brand that bottled childhood — aam panna, jaljeera, aamras.

The proposition reads as exploratory research converted directly into strategy: the discovery wasn't a flavour preference (surveys would rank mango forever) but a meaning, nostalgia for drinks tied to homes and childhoods, the kind of construct only depth conversation surfaces. Everything downstream (naming, pack softness, storytelling tone) reads as executions of one qualitative insight.

What to steal: Qualitative work doesn't just feed questionnaires — sometimes the meaning it uncovers is the strategy, and quantification's job is merely to size the bet.

Meesho

The e-commerce platform built for Bharat's first-time online shoppers.

Its operating style is commonly read as experimentation-as-design: features, pricing surfaces, and category bets trialled against holdouts and staged geographies, with the tier-2/3 user's behaviour, not metro proxies' opinions, as the evidence base. The design insight is the sampling one: decisions about Bharat measured on Bharat, where convenience samples of metro smartphone power-users would mislead by construction.

What to steal: Your most available sample is usually your least representative — pay the cost of reaching whose behaviour the decision actually depends on.

Beyond India

The global lens

IKEA

The furniture giant famous for researching homes, not showrooms.

Its home-visit programme reads as institutional ethnography: researchers observing morning routines and storage chaos across markets, with product lines answering observed life (small-space living, bedroom clutter) rather than stated preference. The method bet is explicit: people can't accurately report frictions they've normalised, so you go watch the frictions.

What to steal: For unarticulated needs, observation isn't a supplement to asking — it's the only instrument that registers the signal at all.

Netflix

The streaming service that A/B tests nearly every surface.

Its artwork and interface experimentation is commonly cited as causal design at industrial scale: competing thumbnails and layouts served to randomised audiences, with viewing behaviour as the verdict. The instructive part is what it implies about opinion: the winning artwork frequently isn't what panels say they prefer, the say-do gap measured daily at scale.

What to steal: Where behaviour can be measured directly, let it testify — surveys retreat to the questions behaviour can't answer: why, and what it means.

From theory to Monday morning

How to use it

  1. Write the decision-first brief

    One page: the decision, the options on the table, the cost of choosing wrong, whose behaviour the outcome depends on, and what evidence would change the current default. If the brief can't name the decision, the study is curiosity with a budget. Postpone it.

  2. Classify the question's species before choosing any method

    Exploratory, descriptive, or causal? Then check the match: no surveys before the vocabulary exists, no focus groups asked to predict behaviour, no causal claims without a control. Most bad studies are good methods on the wrong species.

  3. Audit the sample against the decision

    Write one sentence ('this decision depends on the behaviour of X'), then interrogate the sample frame: who's included, who's systematically missing, who self-selected in? Existing customers, app users, and panels each exclude exactly the people growth decisions are about.

  4. Pilot the instrument on ten people

    Before fielding anything: cognitive-test the questionnaire (what did respondents think each question meant?), rotate orders, check for leading wording and scale bunching. An afternoon of piloting catches the biases that would otherwise be discovered in the board meeting.

  5. Buy causal evidence for expensive decisions

    Pricing, claims, and launch bets deserve trade-off or field evidence: a conjoint exercise for willingness-to-pay, a two-city matched test for the launch, a holdout for the campaign. Match the rigour to the cost of being wrong: the Beginner rule, now with the machinery to obey it.

Watch out

Common mistakes

✕ Fielding the big survey before any exploratory work.

Fix: You'll measure your own assumptions in the customer's absence. Ten interviews first: the survey's job is to size what exploration surfaced, in the words respondents actually use.

✕ Asking focus groups causal questions ('would you buy at ₹99?').

Fix: Groups produce performed consensus, not purchase prediction. Take pricing and behaviour questions to conjoint, test markets, or live experiments, and keep groups for language, reactions, and the reasons behind held views.

✕ Trusting big convenient samples over small correct ones.

Fix: Size cures noise, not bias: 50,000 self-selected reviews still describe only reviewers. State the population the decision concerns, then sample it deliberately, even if that means hundreds instead of thousands.

✕ Reading pre-post comparisons as campaign proof.

Fix: Seasons, distribution, and competitors move too. No control, no causal claim: hold out a region or an audience cell, or report 'sales rose' without the verb 'because'.

Don't just read it

Practice task — 10 minutes

Take a live decision from your work: a price change, a feature, a campaign claim. Classify its species, write the one-sentence population it depends on, and design the minimum honest study: method, sample, instrument risk, and the result that would change your mind. Then price it, and compare that number to the cost of deciding wrong.

If you remember five things

  • Three species of question (exploratory, descriptive, causal), and each demands its own methods, samples, and confidence limits.
  • Sequence them: watch/ask to find what matters, survey to size it, experiment to confirm it.
  • Sampling decides whose truth you learn: frame and self-selection bias beat sample size every time.
  • Every instrument bleeds somewhere: pair methods so one's weakness is another's strength, and pilot before fielding.
  • Causal claims need controls. Willingness-to-pay needs trade-offs (conjoint), not testimony. Match rigour to the cost of being wrong.