Research design (study plan)

Imagine the situation: a company launches a large-scale customer survey, spends two weeks collecting responses, and then discovers that the data is impossible to interpret.

The questions turned out to be ambiguous, the sample was unrepresentative, and the research objectives were phrased so vaguely that it is unclear which question even needs to be answered. All of this is the consequence of having no well-thought-out research design.

What research design is

Research design (study plan) is a comprehensive plan that defines how a study will be conducted: from setting objectives and formulating hypotheses to choosing data collection methods, defining the sample, and selecting ways to analyze the results.

If we draw an analogy with construction, a research design is the architectural blueprint. Without drawings you can start building a house, but the walls will most likely be crooked, the windows will be in the wrong places, and the foundation will not bear the load. Research works the same way: without a clear plan you risk spending time and money on data that means nothing.

A research design answers five fundamental questions:

1. Why? — what problem are we solving and what decisions will we make based on the results
2. Who? — who are our respondents, how many do we need, and how will we find them
3. How? — which method will we use to collect data (survey, interview, observation)
4. What? — which specific questions will we ask and in what format
5. And then what? — how will we process and interpret the data we obtain

Why you need a research design

In practice many researchers — especially in small businesses — skip the planning stage and jump straight to building a questionnaire. This leads to characteristic problems:

The reliability problem. If the sample is formed incorrectly, the results cannot be generalized to the entire target audience. For example, you surveyed only those customers who are subscribed to your newsletter — but they are inherently more loyal than the average customer. Conclusions about overall satisfaction will be distorted.

The resources problem. Without a plan it is easy to "inflate" a study. Instead of 10 targeted questions you end up with a 50-item questionnaire that nobody wants to complete. Or the opposite — the survey turns out too shallow and fails to provide the needed depth of data.

The applicability problem. The most common case: the data is collected, beautiful charts are built, but management asks "and what are we supposed to do with this?". It means that at the planning stage the specific business decisions that depend on the study results were not defined.

A good research design is not an academic formality but a way to guarantee that every dollar spent and every minute of a respondent's time brings real value.

Main types of research design

Depending on the goal of the study, three classic types are distinguished. Each of them implies its own logic, its own methods, and its own depth of analysis.

Exploratory

Used when you are only beginning to study a topic and do not yet know which questions to ask. The goal is not to obtain precise figures but to understand the context, identify the key problems, and formulate hypotheses for further testing.

When to apply: launching a new product into an unfamiliar market, entering a new customer segment, investigating the causes of an unexpected drop in sales, studying a phenomenon for which there is little data.

Typical methods: in-depth interviews with 10-20 representatives of the target audience, focus groups of 6-8 people, analysis of open reviews and support tickets, expert surveys.

A real-world example. A chain of fitness clubs noticed that 40% of new members stop attending within the first three months. Before launching a mass survey with closed questions, they conducted 15 in-depth interviews with churned members. It turned out that the main reason was neither price nor the quality of the gyms, but the absence of personal attention during the first month. This hypothesis became the foundation for the next stage — a quantitative study.

An exploratory design does not provide answers — it helps you formulate the right questions. It is an investment in the quality of the research that follows.

Descriptive

The most common type of design for online surveys. The goal is to obtain an accurate picture: who your customers are, what they think, how they behave, how satisfied they are. A descriptive design answers the questions "who", "what", "where", "when", and "how many", but does not explain "why".

When to apply: regular NPS and CSAT measurements, researching the profile of the target audience, assessing employee engagement, monitoring satisfaction after a purchase.

A descriptive design comes in two subtypes:

Cross-sectional — data is collected once, at a specific point in time. It is a "snapshot" of the current situation. For example, an annual customer satisfaction survey. Fast and inexpensive, but it does not show the dynamics.

Longitudinal — data is collected repeatedly from the same or different groups of respondents over an extended period. For example, quarterly employee pulse surveys. More expensive and more complex, but it lets you track trends and assess the effect of the actions taken.

A real-world example. An online store surveys buyers 3 days after the order is delivered. The questionnaire contains 8 questions: rating delivery speed, packaging quality, whether the product matched its description, and the likelihood of a repeat purchase. Data is collected continuously, and once a month a summary report is generated for the quality department. This is a typical descriptive design with cross-sectional measurements.

Causal / Experimental

The most rigorous type of design. It allows you not just to describe a situation but to establish a cause-and-effect relationship: action A leads to outcome B. This requires controlled conditions — a control group and an experimental group, random assignment of participants, and isolation of variables.

When to apply: testing the effectiveness of a new support script, comparing two versions of a questionnaire (A/B testing), assessing the impact of a loyalty program on repeat purchases, testing hypotheses from the exploratory stage.

A real-world example. The HR department of a large company wants to understand whether the frequency of feedback from a manager affects employee engagement. Two departments with similar profiles are split into groups: in one, managers hold weekly 15-minute conversations with each employee; in the other, everything stays as before. After three months both groups take the same engagement survey. The difference in results allows a conclusion to be drawn about the impact of regular feedback.

A causal design is the only way to reliably answer the question "why". But it requires more time, resources, and methodological rigor than a descriptive one.

How to develop a research design: a step-by-step process

Developing a design is not a one-time action but an iterative process. Below is a sequence of steps, each of which affects the quality of the final result.

Step 1. Formulate the problem and the goal

Start with the business problem, not with the questionnaire. Ask yourself: "What decision will we make based on the results of the study?" If there is no answer, the study is premature.

Compare two ways of phrasing the goal:

Bad: "Find out what customers think about our service".

Good: "Identify the three main factors that cause customers not to renew their subscription, in order to adjust the onboarding program in Q3".

The second phrasing is specific, tied to an action and to a timeframe. It immediately sets the boundaries for the whole study: whom to survey (customers who did not renew), what to ask about (churn factors), and what to do with the results (change the onboarding).

Step 2. Choose the type of design and the method

Based on the goal, determine which type fits. For most business tasks a descriptive design with an online survey is enough. If you need to understand "why", add an exploratory stage (interviews). If you need to prove a cause-and-effect relationship, an experiment will be required.

For a more detailed overview of research methods, we recommend the article "Which research methods exist".

Step 3. Define the sample

The sample is the group of people you will survey. Two key parameters:

Who exactly. Define the criteria: age, gender, region, length of product use, customer segment. The more precisely the target group is defined, the more relevant the results.

How many. The sample size depends on the desired precision. As a reference point: to get results with a ±5% margin of error at a 95% confidence interval, you need about 370 responses for a population of 10,000 people. For 100,000 — about 383. For a million — about 384. Beyond a certain threshold the size of the population has almost no effect on the required sample size.

If you do not have your own respondent base or it is insufficient, you can use a respondent panel — services that give access to hundreds of thousands of profiles that can be filtered by dozens of criteria.

Step 4. Develop the data collection instrument

At this stage you design the questionnaire itself. A few principles that significantly improve data quality:

The funnel rule. Start with simple general questions and gradually move to more specific and complex ones. This reduces cognitive load and keeps the respondent's attention.

One question — one idea. Avoid questions like "How satisfied are you with the speed and quality of our delivery?". These are two different parameters, and an answer to a combined question cannot be interpreted unambiguously.

A balance of open and closed questions. Closed questions (scales, multiple choice) are easy to analyze, but they limit the respondent to predefined options. Open questions provide depth but are harder to process. The optimal ratio: 80% closed, 20% open. More about question types in the article "Open vs closed questions".

Optimal length. For most tasks — 8-15 questions, with a completion time of 3-7 minutes. Each additional question increases the share of incomplete responses. Studies show that after the 10th minute of filling out the form, the drop-off rate rises sharply.

Step 5. Run a pilot test

Before launching to the full audience, test the questionnaire on 10-20 people from the target group. Pay attention to:

• Clarity of wording — ask the pilot participants how they understood each question
• Completion time — if it exceeds 7-10 minutes, cut it down
• Technical issues — correct display on mobile devices, the operation of logic jumps
• Distribution of answers — if 95% of respondents answer a given question the same way, the question is useless

More about pilot studies in a separate glossary article.

Step 6. Collect the data

Choose distribution channels based on where your audience is. An email campaign gives a good response rate when working with existing customers. Embedding the survey on a website (an iframe or a pop-up window) is suitable for collecting feedback at the moment of interaction. QR codes work in offline locations: stores, clinics, events.

While collecting data, keep an eye on two metrics: the share of those who responded and the share of those who abandoned the questionnaire halfway. If the second metric exceeds 30-40%, it is a signal to reconsider the length or structure of the survey.

Step 7. Analyze and act

Collected data without analysis is just a table of numbers. The minimum set of analytics for any study includes:

• Descriptive statistics — averages, medians, and percentage distributions for each question
• Segmentation — comparing answers across different groups (new vs. old customers, men vs. women, regions)
• Identifying relationships — how answers to one question correlate with answers to others

But the most important thing is translating the data into specific actions. Every conclusion of the study should be accompanied by a recommendation: what to do, by whom, and within what timeframe.

Common mistakes and how to avoid them

Over years of conducting research, an entire catalog of mistakes that recur again and again has accumulated. Here are the six most destructive ones:

1. "Research for the sake of research". A survey is launched because "we haven't done one in a while" or "competitors are doing it". Without a clear business goal the results will end up in a "for the future" folder and never be used. Before you start, answer: which specific decision will change because of this data?

2. Selection bias. You survey only those who are convenient to reach: newsletter subscribers, website visitors, active users. But these people are not your "average" customers. They are more loyal, more engaged, more digitally savvy. Conclusions based on such a sample will be overly optimistic.

3. Leading questions. "Do you agree that our new interface has become more convenient?" is not a question but a hint. The respondent will be inclined to agree out of mere politeness. A neutral phrasing: "How would you rate the usability of the new interface compared to the previous version?" More about typical distortions in answers.

4. A questionnaire that is too long. Each additional question is a trade-off. On the one hand, more data. On the other — lower answer quality and a higher rate of incomplete questionnaires. The golden rule: if a question is not directly related to the goal of the study, remove it.

5. The absence of a pilot. It often seems that the questions are "clear as they are". But what is obvious to the questionnaire's author may be completely opaque to the respondent. A pilot test on 10-15 people from the target group reveals 80% of the problems.

6. Analysis without context. An average score of 4.2 out of 5 — is that good or bad? Without a point of comparison (the previous period, competitors, industry benchmarks) any figure is meaningless. Always plan what you will compare the results against, before you even start collecting data.

How to choose the type of research design: practical recommendations

The choice of design depends on three parameters: the goal, the resources, and the timeframe.

If you need to collect feedback quickly — a descriptive design with a short online survey. A questionnaire of 5-8 questions, distributed via email or embedded on a website. Results can be obtained in 3-5 days.

If you need to get to the bottom of a complex problem — start with an exploratory stage (5-10 interviews), then move on to a mass survey. This will take 3-4 weeks but will significantly improve the quality of the results.

If you need to prove the effectiveness of a change — a causal design with a control group. The most resource-intensive option, but the only way to reliably answer the question "does this work?".

If resources are limited — use ready-made survey templates. For typical tasks (NPS, satisfaction, exit interviews) there is no need to invent a questionnaire from scratch — proven templates already contain an optimal set of questions and a structure.

Research design and online surveys

Modern survey-building platforms significantly simplify the implementation of a research design. The SurveyNinja builder lets you bring almost any research idea to life:

For an exploratory design — open text fields and elaborate questions. You can set a minimum of constraints and let respondents speak freely.

For a descriptive design — rating scales, numeric scales, matrix questions, and logic jumps to adapt the questionnaire to the respondent's profile. Built-in analytics with filtering and data export.

For a causal design — hidden variables that let you pass information about a respondent's membership in the control or experimental group through URL parameters, without showing it in the questionnaire. More about this in the guide to creating a high-quality survey.

A research design is not a bureaucratic step that can be skipped for the sake of speed. It is the foundation that determines whether the collected data will be useful or useless. The more complex the task and the higher the stakes, the more attention you should pay to planning before you create the first question.