An opportunity solution tree is a one-page map that ties every solution a team is working on to a customer opportunity, and every opportunity to one outcome. Teresa Torres created it, and her book Continuous Discovery Habits is the full method. None of it is mine. As of November 2026, after 15 years of building products, I use the tree for one job: checking that the thing we're about to build still traces back to a reason. The outcome sits at the top. Opportunities are customer needs, pains, and wants found in research. Solutions hang under one opportunity each. Assumption tests hang under solutions. The tree goes wrong in two ways. Solutions get written in as opportunities, and nobody ever cuts a branch.
What is an opportunity solution tree?
An opportunity solution tree is a visual map with four levels. The desired outcome sits at the top. Under it are opportunities, which are customer needs, pain points, and desires. Under each opportunity are the solutions that might address it. Under each solution are the assumption tests that would show whether it works.
The method belongs to Teresa Torres. I'm describing it here as I understand it and use it. Where I add my own opinion, I'll say so.
Two details matter more than the diagram.
The outcome is a number the team can move. "More new accounts take a first booking in their first week" works. "Grow the business" doesn't, because nothing can be traced to it.
An opportunity is written in the customer's terms. It describes a need or a pain, and it leaves room for more than one answer. That second part is the one teams get wrong, and I'll come back to it.
What is the tree for?
The tree keeps solutions tied to an outcome through an opportunity. Every idea has to answer two questions before it earns a place: which customer need does this serve, and how does serving that need move the outcome? An idea that can't answer both has nowhere to hang. That is useful to know before anyone builds it.
A roadmap is a list. A list shows what and when. It doesn't show why, so the why lives in whoever argued hardest that quarter.
The tree puts the why on the page. That changes the conversation when someone brings a favourite feature. The question stops being "should we build this?" and becomes "where does this hang?" The second question is easier to answer without anyone losing face.
It also sets up a comparison. Torres's argument is that teams decide better when they compare a few solutions for one opportunity than when they judge one idea at a time. A single idea gets a yes or a no. Three ideas for the same need get a real choice.
My own opinion: the comparison is the part most worth protecting. A tree with one solution under each opportunity is a roadmap drawn sideways.
How do solutions end up disguised as opportunities?
It happens when a feature request gets written down in the customer's voice. A request for calendar sync sounds like a need, but it names a fix. The real opportunity is whatever made them ask. A quick check: if only one solution could ever satisfy the box, the box is a solution.
The disguise works because the words are real. A customer did say them. Writing them down feels like listening.
Here is how I'd rewrite four of them. These are examples I made up for a small-business scheduling tool, not quotes from real customers.
| What got written on the tree | Why it's a solution | The opportunity underneath |
|---|---|---|
| "I need calendar sync." | Only one thing satisfies it. | "I'm afraid of double-booking against my personal calendar." |
| "Send reminder texts." | It names a feature. | "My customers forget, and an empty slot costs me money." |
| "Let me export to a spreadsheet." | It names a file format. | "I can't answer my accountant's questions from inside the tool." |
| "Make onboarding shorter." | It's the team's voice, and it's a fix. | "I don't know which setup steps I can skip and still take a booking." |
Look at the right-hand column. Each line now allows several answers. Fear of double-booking could be solved by sync, by a blocked-time button, or by a warning before a booking is confirmed.
That is the test I trust. If I can't name a second solution, I haven't found the opportunity yet.
What does a worked tree look like?
It looks like an outline with reasons attached. The example below is a composite I made up for a small-business scheduling tool. The interview counts and details are invented to show the shape, and they come from no real project. Read the pruning notes first. They are where the thinking shows.
OUTCOME
More new accounts take a first real booking within 7 days of signing up.
├── OPPORTUNITY (TARGET)
│ "I can't tell if my booking page works until a real customer tries it."
│ Heard in 6 of 9 interviews. All 6 had delayed sharing their link.
│ │
│ ├── SOLUTION A: a test booking the owner can make on themselves
│ │ └── TEST: offered one at the end of setup, do owners take it?
│ │
│ ├── SOLUTION B: a "ready to share" check that lists what's missing
│ │ └── TEST: do owners who see the check share their link sooner?
│ │
│ └── SOLUTION C: a live setup call with our team
│ PRUNED. Would work. We can't staff it, and it teaches us
│ nothing about the product.
│
├── OPPORTUNITY (NEXT)
│ "My regulars still text me to book, so the tool sits empty."
│ Heard in 4 of 9. No solutions yet. We'll add them when this
│ becomes the target, not before.
│
├── OPPORTUNITY
│ "I need my old appointments in here before I can switch."
│ PRUNED. Heard in 2 of 9, and both had already taken a first
│ booking. Real need, wrong outcome. Parked for the retention tree.
│
└── "Add calendar sync."
REJECTED AS WRITTEN. It's a solution. Asked what happened last
time: one owner was double-booked on a Saturday.
Rewritten as an opportunity: "I'm afraid of double-booking
against my personal calendar." Heard once. Parked until we
hear it again.
A few things I did on purpose.
Only one opportunity has solutions under it. The rest wait. Filling in every branch feels thorough and spreads the team across everything at once.
Solution C was pruned even though it would probably work. A solution can be right and still be the wrong thing for this team to learn from.
The import opportunity is the one that hurts to cut. It's a real need from real interviews. It doesn't move this outcome, so it goes on a different tree.
Every cut has a sentence beside it. Six months from now, somebody will ask why import isn't being built. The answer is already written down.
Why does a tree need pruning?
Because a tree that only grows turns back into a backlog. Every interview adds opportunities and every meeting adds solutions. If nothing is ever cut, the tree stops showing a choice and starts showing everything anyone has said. Pruning is the team stating, in writing, what it has decided not to pursue and why.
Torres recommends working one target opportunity at a time. Pruning is how the rest of the tree stays honest while you do.
These are the reasons I'd accept for cutting a branch. They're my rules, not hers.
- Wrong outcome. The need is real, but meeting it wouldn't move the number at the top.
- Too thin. One person said it once. Park it and see if it comes back.
- Failed its test. The assumption test came back against it. This is the easiest cut and the one teams resist most.
- Out of reach. The team can't change it from where it sits.
Pruned doesn't mean false. "Real need, wrong outcome" is the note I'd expect to write most often.
The opposite failure is a tree built once in a workshop and never touched. It looks like discovery. It's a poster.
How does the tree connect to the say-do gap?
The tree is only as good as the evidence in its opportunity boxes. If those boxes hold what customers said they want, the tree inherits the say-do gap: the distance between what people tell you and what they do. Opportunities drawn from specific past behaviour hold up better than ones drawn from stated wishes.
I wrote about the say-do gap from the losing side of an argument. Customers told us they didn't want advertising in a product I worked on. The business shipped it anyway. People stayed, and usage held.
Here is my inference from that, and it is an inference. On a tree, "I don't want advertising here" would have been a well-supported opportunity box. Many people said it, and they meant it. It still didn't predict what they did.
Torres's interviewing advice points at the same problem. Ask for a specific story about the last time something happened. Don't ask what someone would do.
The calendar sync row in the example shows the difference. "Add calendar sync" is what the owner said. "I was double-booked on a Saturday" is what happened.
The bottom level of the tree is the other defence. An assumption test measures what people do with a small version of the idea. It turns a belief on the tree into behaviour you can count.
This is the same order of work I argued for in AI product discovery: find the pain before you pick the model. A model is a solution. On a tree it belongs on the third level, under a need somebody has.
When is the tree the wrong tool?
Skip it when there is no outcome to put at the top, when the work is a fixed commitment with no real choice left, or when nobody is talking to customers. In each case the tree becomes decoration. It organizes discovery. It can't stand in for doing the discovery.
This part is opinion. A team with a contract deadline and a signed scope doesn't need a tree. It needs a plan.
A team with no customer conversations can still draw one. Every box will hold a guess, and the diagram will make the guesses look like findings. I'd rather see an honest list of guesses.
A tree that has never lost a branch hasn't been used to decide anything.