# Before conclusions Check what the data supports and where uncertainty remains. 1. Perfect accuracy does not exist. Results are affected by the method, sample, and data-collection approach. But with a careful approach, distortions are minimal — and the level of reliability will be more than enough to make a confident business decision. 2. Do not treat research as a source of a ready-made answer. It will provide validated facts that help the team make a balanced decision. But to make that decision, you still need to check against business goals and think carefully about the best way to meet user needs. Figuratively speaking, sometimes when a user asks for a blue button, their real problem may be solved by a green banner. Example: A marketplace team was choosing a name for a new delivery type and decided to test several options: describe the delivery format and ask people to come up with a name for it, then choose the most suitable one from a list of ready options. To expand the list of options, alongside other, more creative names, the survey included “Convenient delivery.” Because of its cognitive simplicity, this option received the most votes, and similar phrases often appeared in answers where respondents were asked to invent a name themselves. However, from the perspective of branding and product positioning — if we call one delivery type convenient, does that mean the others are inconvenient? — the team could not choose the most popular survey option and considered the names that placed second and third. In an ideal world, “Convenient delivery” should have been excluded from testing. 3. One study is often not enough. To make a balanced and well-founded decision, you usually need different methods and several research rounds. ## How to analyze unmoderated research results 1. Look at comments or detailed respondent answers if you have them. Remove answers that are clearly random or meaningless. Count how many comments point to problems with the question or layout. If there are many such answers, more than 10%, fix the issue and run the test again. Analyze open comments and create a classification of the most frequent opinions or complaints. Pathway’s AI summary will help you quickly orient yourself around the main themes and sentiment in open comments. 1. Evaluate numerical metrics if you received them from the study. 1. If you compared variants, report the observed difference and its confidence interval. Check whether the interval includes effects that would change your decision. ![A comparison calculator with sample sizes and observed percentages for groups A and B.](https://wynde.evgeniichekov.com/guide/illustrations/4ce8abfa-8808-4f0d-aaa5-94521bb2f891.png) 1. If you need to generalize the result, check who was recruited and how. The interval captures uncertainty under the method’s assumptions. Assess recruitment bias separately. ![A margin-of-error calculator asks for the number of participants and the observed percentage.](https://wynde.evgeniichekov.com/guide/illustrations/229b40f6-eac9-45f0-8dcf-4f487ff16def.png) 1. If the study used a prototype, analyze how participants interacted with it. Session recordings show where people struggled. Transition maps reveal common paths. Heatmaps highlight where participants interacted with the interface. ![A Pathway heatmap highlights areas of the prototype where participants interacted more often.](https://wynde.evgeniichekov.com/guide/illustrations/5cbae630-0064-40a7-a3a2-6069d16d445d.png)