Plan a marketing research study

Marketing research exists to help someone make a decision: launch or not, which features, what price, which segment to target. A study that does not change a decision is wasted money. Good research starts by working backwards from the decision to the data you need.

From decision to research objectives

1. The management decision: “Should we redesign our mobile app, and which features should the new version include?”
2. The research problem: “What drives customer satisfaction with the app, which features customers value most, and how satisfied are app customers compared with store customers?”
3. Research objectives, each specific and measurable: measure NPS by channel; identify the drivers of overall satisfaction; rank twelve candidate features by preference; classify features as must-haves or delighters.
4. Information needs and methods: each objective maps to a question block and an analysis (NPS calculator, key driver analysis, MaxDiff, Kano).

Write a one-page brief with these four levels before writing a single survey question. It keeps the questionnaire short and focused.

Choosing the research design

• Exploratory research (interviews, focus groups, desk research) when you do not yet know the right questions or the language customers use. It generates hypotheses and the items for your questionnaire.
• Descriptive research (surveys) to measure how many, how much and how satisfied. Most of this course is descriptive.
• Causal research (experiments, A/B tests, conjoint) to measure the effect of changing something, such as price or a feature. Conjoint analysis (Lesson 7) is an experiment embedded in a survey; A/B testing is covered in Course 5.

Writing the questionnaire

• Start with screening questions that check whether the respondent belongs to the target group, then move from general to specific questions, and put demographics at the end.
• One idea per question. “How satisfied are you with the speed and friendliness of our staff?” asks two things.
• Neutral wording. Avoid leading questions such as “How much do you enjoy our new app?”.
• Balanced scales with labelled end points (and ideally all points): 1 = very dissatisfied to 5 = very satisfied. Use the same direction throughout, and reserve the 0–10 scale for NPS.
• Use validated scales where they exist (SERVQUAL for service quality, NPS for loyalty) so results are comparable.
• Pre-test with five to ten people from the target group, and watch where they hesitate.

The free questionnaire templates on this site include ready-made blocks for satisfaction, NPS, SERVQUAL, Kano and pricing questions.

Sampling

A probability sample (every customer has a known chance of selection, for example a random sample from the customer database) allows you to generalise results with a margin of error. A non-probability sample (online panels, social media links, convenience samples) is cheaper and faster, but the margin of error then describes only sampling noise, not selection bias. Use quotas (for example, by age, region and channel) to make non-probability samples resemble the population, and be honest about the limitation.

How precise must the results be? For a proportion such as “share of satisfied customers”, the survey sample size calculator gives 385 completed responses for a margin of error of ±5 percentage points at 95% confidence. For a population of 2,000 customers, the finite population correction reduces this to 323. If you plan to compare subgroups (for example, three channels), each subgroup needs enough respondents on its own: about 100 per group gives a margin of error of ±10 points for each group’s percentage.

Response rates and data quality

Divide the required completes by the expected response rate to get the number of invitations: 385 completes at a 15% response rate need about 2,570 invitations. Build in quality checks: an attention-check question, a minimum completion time, and screening for straight-lining (the data cleaning tool flags it). Plan the analyses in advance; the next eight lessons show which questions each method needs.