UX and research · Intermediate · 14 min
Turn research themes into insight statements
By Padding · Updated 18 Nov 2026
Any notes app and a spreadsheet · November 2026
You will have five insight statements, each saying what people do, why they do it and what it means for the design, with quotes from several participants and a confidence level beside it.
On this page
An affinity map tells you what came up. It does not tell you what to do about it. A cluster called "Delivery times" is a topic. "People skip weeks when they cannot be home for the delivery window, because the box is left outside and the chilled food spoils" is an insight. The second one gives a designer something to work on and a product manager something to argue with.
Most teams blur three different things when they write up research:
- Observation: what you saw or heard. "P3 said she skips any week she is on a late shift."
- Insight: your interpretation of a pattern across people, with the reason behind it. "People plan around the delivery, not the meals."
- Recommendation: what the team should do. "Let people choose a delivery day each week."
Keep them apart and each one can be judged on its own terms. An observation is either accurate or not. An insight is either supported by the evidence or not. A recommendation is either worth the cost or not. When the three are mixed into one sentence, a weak recommendation hides behind a strong observation and nobody notices.
The worked example in this tutorial comes from eight interviews about a meal-kit subscription: people who had paused or cancelled in the last three months. Swap in your own clusters as you go.
Step 1 · Set up an evidence sheet
Create a spreadsheet with one tab called Evidence. Give it these columns:
- Cluster: the name of the affinity group the note came from
- Participant: P1, P2 and so on
- Observation: one thing the person said or did, in plain words
- Quote: their exact words, if you have them, in quotation marks
- Source: where to find it (transcript file and timestamp, or notes page)
Copy the sticky notes from every named cluster on your map into this tab, one row per note. Do not pick favourites yet: the next step counts participants per cluster and sets the weak ones aside, so it needs all of them. Do not paraphrase into interpretation as you copy. "P5 cancelled after two boxes arrived while she was away" is an observation. "P5 found the service inflexible" is already an interpretation, so rewrite it back to what happened.
Add a second tab called Insights with the columns Insight, Evidence rows, Participants, Counter-evidence, Confidence and Recommendation. You fill it from step 3 onwards.
Check it worked
Your Evidence tab has at least one row per participant in every cluster you copied, and each row has a source you could open within a minute. Read down the Observation column: no row contains words like "frustrated", "confused" or "wants" unless the participant used that word themselves.
Step 2 · Choose which clusters deserve an insight
Not every cluster becomes an insight. Sort the Evidence tab by Cluster and count how many different participants appear in each one. A quick way is a pivot table, or a small summary block beside the data where you type the count by hand.
Keep a cluster as a candidate when:
- three or more participants contribute to it, or
- fewer participants contribute, but what they describe is serious (money lost, a safety issue, someone giving up entirely)
Set aside clusters that are one person talking at length. They are worth a note in your report's open questions, not a headline.
In the meal-kit study, "Delivery window" had six participants, "Recipe too long on weeknights" had five, "Packaging waste guilt" had four, "Price compared with supermarket" had seven, "Partner's dietary needs" had two but both cancelled because of it, "Missed the skip deadline" had two who were each charged for a box they had meant to skip, and "App colour scheme" had one. The last one was set aside, which left six candidates.
Carry at least six forward even if one looks weak. In the last step you will merge or drop some, and spares mean you still end with five.
Check it worked
You have a short list of six or seven candidate clusters, each with a participant count written beside it. You can say in one sentence why each low-count cluster you kept is on the list.
Step 3 · Write a draft insight for each cluster
An insight statement has three parts, and you can write it as one or two sentences:
- What people do: a behaviour, not an opinion. Start with the people: "People who…", "Customers…", "Parents…"
- Why they do it: the underlying reason, need or constraint
- What it means for the design: the tension or opportunity it creates, without naming a solution
Use this frame for the first draft, then rewrite it in natural language:
[People] [do this], because [reason]. This means [implication for the product].A draft from the "Recipe too long on weeknights" cluster:
People cook the quick recipes early in the week and leave the longer ones to rot, because after work they judge a recipe by its total time, not the stated cooking time. This means the time we show is not the time people experience.
Notice what it avoids. It does not say "add a filter for quick recipes". That is a recommendation, and it goes in its own column later.
Three tests for every draft:
- Could it be wrong? "People want convenience" cannot be wrong, so it says nothing. Make it specific enough that new evidence could disprove it.
- Is the reason in it? If you removed the "because" clause and the sentence still makes sense, you have a finding, not an insight. Go back to the quotes and find the reason.
- Would a team member act differently after reading it? If not, it is too general.
Check it worked
Every candidate cluster has one draft in the Insight column. Each draft contains a behaviour, the word "because" or an equivalent reason, and an implication. None of them names a feature, screen or button.
Step 4 · Attach evidence from several participants
For each insight, go back to the Evidence tab and list the row numbers that support it in the Evidence rows column. Write the participant codes in the Participants column, for example "P1, P2, P4, P6, P8".
Then pick two or three quotes from different participants to carry the insight. Choose quotes that show the reason, not only the behaviour. For the recipe insight:
"It says 25 minutes. It is 25 minutes after you have chopped everything, and nobody tells you that." (P2)
"Thursday's box is always the one that goes in the bin." (P6)
"I look at the photo and think, that is a Sunday meal." (P4)
Now do the part most teams skip: look for counter-evidence. Search the Evidence tab for people who behaved the opposite way or gave a different reason. In the meal-kit study, P7 cooked every recipe on the day it was planned. P7 also worked from home. Write that in the Counter-evidence column. It does not weaken the insight; it marks its edge ("this applies to people cooking after a commute").
If the counter-evidence outweighs the support, rewrite the insight or drop it.
Check it worked
Every insight has at least two quotes from two different participants, a participant list, and either a counter-evidence note or the words "None found" after you searched. You can click from any quote back to its source row.
Step 5 · Give each insight a confidence level
Confidence tells readers how much weight to put on an insight before they spend money on it. Use three levels and write the reason next to each:
- High: most participants show the pattern, it appears in what they did as well as what they said, and counter-evidence is rare or explained
- Medium: several participants show it, but the evidence is mostly what people said, or the sample covers only part of your users
- Low: a few participants, or a strong signal from a narrow group. Worth testing, not worth building on yet
For the meal-kit insights:
- Delivery shapes the week (6 of 8): High. Four described skipped or spoiled boxes they had actually had.
- Total time, not cooking time (5 of 8): High. The behaviour showed in order histories participants shared during the calls.
- Price is compared per meal, not per box (7 of 8): Medium. Mostly stated; nobody showed a real comparison.
- Waste guilt drives pausing (4 of 8): Medium. A consistent reason, but all four lived in one city.
- Mixed diets break the plan (2 of 8): Low. Both cancelled, but only two households.
- The skip deadline arrives before people know their week (2 of 8): Low. Both were charged and showed the emails, but only two people.
Be honest about the ones you want to be true. Researchers tend to rate their favourite insight one level too high.
Check it worked
The Confidence column is filled for every insight, and each level has a reason a colleague could check against the Evidence tab. At least one insight is rated below High. If they are all High, look again at whether the evidence is behaviour or opinion.
Step 6 · Write recommendations in their own column
Only now write what the team should do. Put it in the Recommendation column, one or two per insight, and keep it separate from the insight wording.
A recommendation can be a design direction, a further question to research, or an experiment. Match it to the confidence level:
- High confidence: a design change or a decision. "Show total time, including preparation, on every recipe card."
- Medium confidence: a test. "Run a two-week trial of a per-meal price display with new sign-ups."
- Low confidence: more research. "Interview five more households with mixed diets before changing the plan builder."
If you find yourself editing the insight so that it points neatly at your favourite recommendation, stop. The insight comes from the evidence; the recommendation comes from the insight, and there may be several good ones.
Check it worked
Read across each row. The Insight cell would still be true if the team rejected the Recommendation. Every Low insight has a research or test recommendation, not a build recommendation.
Step 7 · Choose five and pressure-test them with a colleague
Pick the five insights you will take forward. Favour the ones with higher confidence and the biggest effect on the product question you started with. Move the others to an "Also noted" list at the bottom of the Insights tab.
Ask a colleague who did not run the sessions to spend twenty minutes with the sheet. Give them only the Insights tab first and ask them to say, for each one, what they think the evidence was. Then show them the Evidence tab. Note where their guess and the evidence differ: those insights are either unclear or overreaching.
Ask them two more questions:
- Which insight would you challenge in a planning meeting?
- Is any pair of insights really the same thing said twice?
Rewrite based on what you hear. Merging two overlapping insights often produces a sharper one. In the meal-kit study, "Delivery shapes the week" and "Waste guilt drives pausing" turned out to share a reason (food arriving when people could not use it) and were combined into "Food arrives when people cannot use it" (7 of 8, High). That took the six candidates down to five, so "Mixed diets break the plan" and "The skip deadline arrives before people know their week" both stayed in as Low-confidence insights with research recommendations.
If merging or dropping leaves you with fewer than five and nothing in "Also noted" is strong enough to promote, stop at four and add one line under the table saying why. A thin fifth insight does more harm than a short list.
Check it worked
Your Insights tab has five rows above the "Also noted" list, or four with a one-line note explaining why. Each has a statement, at least two quotes from different participants, a participant count, a counter-evidence note, a confidence level with a reason, and a separate recommendation. Your colleague can explain each insight back to you in their own words without looking at the sheet.
Next steps
Your five insights are the core of what you share with the team. Turn them into a short report and playback that leads with the decisions they support in Report research findings so the team acts on them.
If your clusters felt thin while you wrote the evidence sheet, go back a stage and rebuild them with Build an affinity map from interview notes. Insights that are rated Low often need more sessions; Run your first user interview covers how to run them, and Build a persona from research data shows another way to use the same evidence.
Part of these paths
Related tutorials
- Build an affinity map from interview notesYou will turn the notes from five or six interviews into an affinity map of five to eight groups, each named as a finding and backed by notes from more than one participant.
- Report research findings so the team acts on themYou will have a short findings report that opens with the decisions it supports and backs each finding with evidence, plus a 15-minute playback plan that ends with a named owner for each recommendation.
Need this done for you?
Padding does this work for growing companies, from the first diagnosis to a system that keeps working.
Get new tutorials