Generative vs. Evaluative Research: Define Problems vs. Judge Solutions

Why it exists

Product development requires answering two different questions: "What is the problem?" and "Is our solution working?" Generative and evaluative research provide distinct toolkits for each. Using the wrong toolkit for the job leads to building products nobody wants or failing to improve things that are broken. They formalize the process of exploring a problem versus inspecting a solution.

The mental model

Think of it as an explorer versus a building inspector. Generative research is the explorer, mapping unknown territory to find opportunities and define the problem space (e.g., "Where should we build a city and why?"). Evaluative research is the inspector, checking if a specific house is built to code and meets its residents' needs (e.g., "Is this kitchen functional?"). One finds the "what," the other validates the "how well."

How it works

Generative research, also called discovery research, uses open-ended, qualitative methods like in-depth interviews, field studies, and observations. The goal is to uncover latent needs and pain points without a preconceived solution. You are listening for problems. Evaluative research uses more constrained methods, often quantitative, like usability testing, A/B tests, or surveys. The goal is to measure a specific design against criteria like task success rate or user satisfaction. You are testing a hypothesis about a solution.

When to use it

Use generative research at the very beginning of a project, before you have a clear solution. It's for finding new opportunities, understanding a new user segment, or exploring a broad problem space. Use evaluative research once you have a prototype, feature, or live product that you want to assess. It's for validating design choices and identifying areas for improvement.

When not to use it

Do not use evaluative methods when you don't understand the problem. Asking users to rate two versions of a bad idea is a waste of time. Conversely, do not use generative methods to get a simple "yes/no" on a design decision; open-ended interviews won't give you statistically significant data to choose between two button colors.

One canonical example

A team wants to build a new app for home cooks. Generative research would involve interviewing people about how they plan meals and shop for groceries, looking for unsolved problems. They might discover a huge pain point is food waste. Based on that, they design a feature to track pantry inventory. Evaluative research would then involve giving users a prototype of that feature and observing if they can successfully add items and understand the expiration alerts.