Synthesis: Turning Research Into Insight
You did the interviews. You have a wall of notes, forty pages of them, a shared doc full of quotes, three teammates who each heard something a little different. It feels like progress.
It isn't. Not yet.
Notes aren't insight. The work is what you do after the interview.
This is the step most teams skip, and it's the one that decides whether all that research was worth doing. You talk to ten people, you fill a document, and then you go build whatever you walked in wanting to build. The notes sit there. Nobody clustered them. Nobody pulled a conclusion out of them. The research happened, and it changed nothing.
That's the most expensive kind of research there is. You paid for it in time and customer goodwill, and you threw the answer away because turning a pile of notes into a decision is hard, slow work that doesn't feel like building.
So let's do the hard part. Here's how you turn a wall of notes into something you can act on.
"We have forty pages of notes"
Great. And zero decisions, until you cluster them.
Raw notes are not the deliverable. A transcript is not insight. A folder of recordings is not insight. Those are the ingredients. Insight is what you cook out of them. The small number of true things you now know about your customer that you didn't know last week, stated clearly enough that someone could build on them.
The gap between a wall of notes and a clear insight is the entire job of this page. Most teams stand at one end of that gap, look across, and decide the notes are good enough. They're not. A note is a fact one person said one time. An insight is a pattern that held across the people you talked to, that points somewhere.
Getting from one to the other is a method. Four moves, in order.
The method: analyze, cluster, conclude, then find where it hurts most
I teach this at the University of Illinois, in the design thinking course at the Siebel Center for Design, and I run the same four moves on real product work. It's the same method either way. A pile of qualitative research goes in one end, the interviews and quotes, the messy human stuff, not numbers, and a short, defensible list of insights comes out the other.
Here's the shape, then we'll walk each step.
- Analyze. Break each interview down, question by question, and tag what you heard.
- Cluster. Group the tagged notes into themes.
- Conclude. Turn each theme into a stated conclusion.
- Find where it hurts most. Step back and see where the pain is concentrated.
Then you turn those conclusions into the artifacts that make them stick: a persona, a journey map, an empathy map. Stick with me, and once you've got conclusions I'll show you how to make them last.
Step 1: Analyze each interview
Start narrow. One interview at a time, broken down question by question.
For each interview, go through what the person said and pull out the moments that matter. You're hunting for three things specifically, and it helps to tag them as you go:
- Pains. Where it hurt. The friction, the workaround, the thing they complained about.
- Delights. Where it worked. The moment something clicked, the part they'd miss if it went away.
- Surprises. The thing you didn't expect to hear. These are gold, because a surprise is the sound of an assumption breaking.
Tag every note as one of those three. The cleanest way I've found is colored sticky notes, one color per type, so the color is the tag. Pains in one color, delights in another, surprises in a third. Pick a color, write the note on it, and now the note is labeled just by where it sits. It sounds like kindergarten. Stick with me, the color is doing real work, and it pays off in Step 4.
Do this for every interview before you do anything with all of them together. The discipline here is staying close to what people actually said. You're not interpreting yet. You're not deciding what it means. You're just pulling the real moments out of each conversation and labeling them, one at a time.
One rule while you do it. You analyze all of them. Not the two you remember best. Not the one that confirmed what you already believed. Every interview, every time. The pattern you're after only shows up when the whole set is on the table, and the interview you skip is usually the one that would have changed your mind.
Step 2: Cluster what you heard
Now zoom out. Take every tagged note from every interview, put them all in front of you, colors and all, and group the ones that are about the same thing. The colors come with them. Don't sort by color, sort by topic, and let each note keep the color it already has.
This is affinity mapping. The name sounds fancier than the move. You're sorting notes by what they have in common (that's the "affinity"), until related notes sit together and the groups start to name themselves. "These nine are all about how hard it is to find the right information." "These six are all about not trusting the data." Each group is a cluster, and the cluster gets a title that says what it's about.
Take the key points you heard and cluster them. That's the whole instruction. You'll know it's working when a note you weren't sure about suddenly has an obvious home, and when a cluster you thought was one thing splits into two because the notes inside it are actually pulling apart.
Don't force it clean. Some notes live in two clusters. Some clusters are tiny. Some are a mess until you move three notes and they snap into focus. That's normal. You're looking for the shape of what your customers told you, and the shape is in the data, not in your head.
Here's where a wall of notes stops being a wall. Forty pages become a handful of clusters, call it eight. Eight isn't a target, it's just what falls out when you stop splitting hairs. The number's right when you can hold them all in your head at once, argue about them, point at them, build on them. If you've got twenty, you over-split. If you've got two, you lumped things that don't belong together.
Step 3: Build conclusions
A cluster is a group of related notes. It's still not a decision. The next move is to look at each cluster and ask one thing. What's the one true thing this group of notes is telling me?
That sentence is your conclusion. One per cluster. A short, stated claim that the notes underneath it support.
So a cluster full of "I rely on this person, that person, the people I used to work with" becomes a conclusion. The relationships they already have do more for them than any tool we'd build. The cluster is the evidence. The conclusion is the claim, and you can defend the claim by pointing at the stickies under it.
This is the move that separates synthesis from note-taking. You're going on record. You're saying, out loud, here is what I now believe about the customer, and here is the pile of real quotes that backs it up. A conclusion you can't trace to specific things people said isn't a conclusion.
Build a conclusion for every cluster that earns one. Some clusters are too thin, two notes from the same person, and they don't get to graduate into a claim. Be honest about which ones held and which ones were noise.
Step 4: Find where it hurts most
Now the colors pay off.
Step back from the whole board and just look at it. Where is one color stacked up? A cluster that's mostly pain-colored is a place where it hurts the most. A cluster that's mostly delight-colored is a place where there's something to protect or extend. You're not reading the words anymore. You're reading the colors.
This is the cheapest, fastest read in the whole method. Which area of the board is the most pain? That's probably where your biggest opportunity is. Which is the most delight? That's the part you'd be a fool to break. You tagged every note by color in Step 1 exactly so that, three steps later, you could see the answer at a glance instead of re-reading forty pages to find it.
That concentration is a finger pointing at where to look first. It doesn't make the decision for you. It just makes the stack of pain impossible to miss, the thing your gut would otherwise talk you out of.
Use AI to get through the wall faster
Here's where AI earns its place in this work, and it's a real one.
Clustering a wall of notes by hand is slow. AI is genuinely good at the first pass. Feed it your raw, tagged notes and it'll group them into themes, surface the patterns that repeat, and draft candidate conclusions in the time it'd take you to lay out the sticky notes. For a job that used to eat an afternoon, that's worth a lot.
Two guardrails, because this is where people get lazy.
First, the input has to be real. AI doesn't invent insight out of nothing. It reflects the quality of what you feed it, so feed it the actual interview notes, all of them, not your summary of what you think they said. A summary is already your bias. Hand the model the receipts.
Second, you own the conclusion. The model is a fast intern doing the first cut, not the senior person signing off. Read what it grouped. Argue with it. Move the notes it got wrong, kill the conclusions it overreached on, keep the patterns you can trace back to real quotes. The judgment is yours. The grunt work is its.
AI doesn't replace synthesis. It just gets you to the part that needs a human faster.
Turn the insight into something that sticks
You've got conclusions now. Clear, defensible, traceable to real quotes. The last move is to package them so they survive contact with the rest of your team, because an insight that lives only in your head dies the first time someone with a louder opinion walks into the room.
Three artifacts do this. Here's what each one is and when to reach for it. These are the quick version. The Toolbox goes deeper on each, with templates you can copy.
The persona. A persona is a single, named, made-up character who stands in for a real pattern of customers. Not a demographic, a behavior. Say you give yours a name and call her Dana. You build Dana out of the conclusions. What she's trying to get done, where it hurts, what she's already doing about it, what she believes. The point of a persona isn't to be cute. It's to give the team one specific person to build for, so that "the user" stops being an abstraction everyone fills in with themselves. When your engineer hears "would this work for Dana?" and can actually picture her, the persona did its job.
The journey map. A journey map lays out, step by step, what the customer goes through to get a job done, start to finish, with the highs and lows marked along the way. It's a timeline, not a snapshot. You walk the whole path: what they do first, what they do next, where they hit the wall, where they give up or break through. The value is that it shows you problems in context. A pain that looks small on its own often turns out to sit right at the step that decides everything, and you only see that when you map the whole journey instead of one moment.
The empathy map. An empathy map captures what a customer says, thinks, does, and feels about a situation. Four quadrants, one person, one moment. The reason it's worth doing is the gap between the quadrants. What someone says and what they actually do are often different things, and that gap is exactly where the real insight hides. Picture a car buyer who tells you he negotiates everything, then admits he paid the full sticker price on his last car without blinking. That gap, between what he says he does and what he actually did, is what an empathy map is built to catch and hold up where you can't look away from it.
You don't need all three on every project. You need the one that makes your insight undeniable to the people who have to act on it. Reach for the persona when the team keeps building for a vague "user" and you need one face to aim at. Reach for the journey map when the problem is a sequence and you need to show where in the flow it breaks. Reach for the empathy map when someone's words and actions don't line up and that gap is the point. Pick the artifact that turns your conclusion into something a teammate can see.
What this is all for
Synthesis isn't paperwork. It's the bridge between hearing things and knowing things.
A wall of notes is what you have before synthesis. A short list of conclusions you can defend, wrapped in an artifact your team can see, is what you have after. One of those changes what you build. The other gets filed and forgotten.
And it points somewhere specific. The whole reason you run the method, analyze, cluster, conclude, find where it hurts most, is to answer one question. Of everything you just learned, what do you do something about first? The board gave you a hint. The next step makes it a decision.
Finding where it hurts most pointed you at the worst spot. The thing sitting under that spot is usually a bet you're making about your customer, and you haven't tested it yet. Every solution you'll build rides on a stack of those bets, those assumptions, and some of them, if they're wrong, kill the whole idea. Your synthesis just told you where to point that question. So before you build, figure out which assumption to test first, and how to test it cheap. That's what Assumptions & Risk is for.