Product Discovery Framework: A Practical System for 2026 Teams
A product discovery framework is the loop that turns customer evidence into tested decisions. The models worth combining, five steps, and a copyable discovery record.
Share
A product discovery framework is the repeatable loop a team uses to decide what to build: set an outcome, collect customer evidence, frame the opportunities, test the riskiest assumptions, and record each decision with the evidence behind it. The strongest versions combine Teresa Torres's opportunity solution tree, Marty Cagan's four risks and Lean-style experiments, and keep the evidence where the next person can find it.
Key takeaways
Most ideas fail when they meet customers. At Microsoft, only about one-third of well-designed experiments improved the metric they were built to move (Kohavi, Crook & Longbotham, 2009).
Pendo's analysis of 615 product subscriptions found that 80% of features in the average product are rarely or never used (Pendo, 2019).
No single framework covers the whole job; combine Jobs to be Done, the opportunity solution tree, the four risks and experiments.
A framework without a shared evidence record decays. Teams spend over 25% of their working week searching for information (Atlassian State of Teams, 2025).
The practical loop is Capture, Connect, Act: get every signal into one place, link it to an outcome and an opportunity, then test and decide.
What is a product discovery framework?
Product discovery is the work of deciding what to build before you pay to build it. Marty Cagan separates it from delivery: discovery is about learning fast, while delivery is about building software a business can depend on (SVPG, 2015). A product discovery framework is the structure that makes that learning repeatable instead of heroic. For the end-to-end sequence, see the product discovery process guide.
Two models are in common use, and both are legitimate:
Discovery as a phase. The UK Government Digital Service runs discovery as a stage before anything is built, typically lasting around 4 to 8 weeks, and treats a decision to stop as a valid result rather than a failure (GOV.UK Service Manual).
Discovery as a habit. Teresa Torres defines continuous discovery as "weekly touch points with customers by the team building the product" (Product Talk). Here discovery never ends; see how to talk to customers every week.
Use the phase model for a new product or unfamiliar problem, and the habit model once you have customers and ship often. Running both tracks at once is what dual-track agile describes.
Why does a framework matter if you already talk to customers?
Because talking is not deciding. The numbers on idea failure are consistent across sources. Microsoft's experimentation team reported that about two-thirds of tested ideas did not improve their target metric. Pendo found that a mere 12% of features generate 80% of daily usage volume in the average product. And in CB Insights' March 2026 analysis of 431 venture-backed shutdowns, poor product-market fit was cited in 43% of the 385 cases where a reason could be identified (CB Insights, 2026).
A framework does not make ideas better. It makes bad ideas cheaper to find.
Which product discovery frameworks should you combine?
Teams argue about which framework to adopt. The well-known models answer different questions, so a working system borrows one layer from each.
Read the table as a stack. Customer development and Jobs to be Done find problems worth solving. The opportunity solution tree holds them in one picture tied to an outcome. The four risks say what evidence each idea still needs, and experiments produce it.
How does an opportunity solution tree work?
The opportunity solution tree has four levels (Product Talk):
Outcome at the root: the result the team is accountable for, ideally a product metric such as activation or retention.
Opportunities: customer needs, pain points and desires that, if addressed, would move the outcome.
Solutions: ideas for addressing one chosen opportunity.
Assumption tests: small experiments that check whether a solution creates value for the customer and the business.
Torres gives concrete starting rules. Collect three to four story-based interviews before drawing your first tree. Revise the opportunity space after every three to four interviews. And compare at least three solutions for the target opportunity, so you are choosing rather than confirming.
The tree is where the "connect" work happens. Every opportunity should point back to interview evidence, and every solution to the opportunity it serves. A branch with no evidence under it is a guess.
What are the steps in a product discovery framework?
The loop below follows the Capture, Connect, Act pattern. It works for a solo founder and for a team of ten, and it runs weekly once it is set up.
Step 1: Set one outcome
Pick the single result discovery should improve this quarter, with a number. "Raise week-one teammate invites from 30% to 40% of new workspaces" works. "Improve onboarding" does not. Without an outcome, every interesting quote looks like an opportunity.
Step 2: Capture evidence from every touchpoint
Pull interviews, support tickets, sales notes, analytics and churn reasons into one place, with source, date, segment, and whether each item is an observation or an interpretation. A customer interview template keeps notes comparable across conversations, and the customer interview questions guide covers what to ask.
Step 3: Connect evidence to opportunities
Group the evidence by the situation the customer was in, not by the feature they asked for. Map each group onto the opportunity solution tree under your outcome. Keep contradictions visible; a theme that fits only some segments is still useful once you name the segment. The synthesis template shows how to write findings that trace back to their source.
Step 4: Act by testing the riskiest assumption
For the chosen opportunity, sketch at least three solutions. For the leading one, list its assumptions under Cagan's four risks and test the one that would hurt most if wrong. Value risk usually comes first, because a usable, feasible feature nobody wants is still waste. Our guide to assumption mapping covers how to rank them.
Step 5: Decide and record
Write down what you decided, the evidence for it, what you still do not know, and what result would make you revisit. Link the decision to the ticket or spec it produces. This is the step most frameworks skip, and it is the one that stops the next person from repeating your research.
What counts as "validated" before an idea moves to delivery?
Use the four risks as a definition of ready. Cagan assigns value and business viability to the product manager, usability to the designer, and feasibility to the lead engineer (SVPG, 2017). An idea is ready to build when each owner can point to evidence, not opinion.
Risk
Question
Example evidence
Value
Will customers choose to use or buy it?
Repeated past behaviour in interviews; a fake-door or presale signal
Usability
Can people figure out how to use it?
Prototype test where participants finish the task unaided
Feasibility
Can we build it with our time, skills and technology?
A time-boxed technical spike
Viability
Does it work for the rest of the business?
Pricing, legal, support and sales sign-off
Set the thresholds before testing, so a weak result cannot be reinterpreted as a pass.
Why does discovery need infrastructure, not just rituals?
Most discovery fails in storage, not in the interview. Notes live in personal documents, recordings in a video tool, and the reason a feature exists in one person's head. When that person leaves, the reasoning leaves too.
Atlassian's State of Teams 2025 survey of 12,000 knowledge workers found teams spend over a quarter of their week searching for information, an estimated 2.4 billion hours a year across the Fortune 500 (Atlassian, 2025). Discovery evidence is among the hardest to find, because it is unstructured and scattered.
Treat the evidence as a shared layer with four rules:
One home. Every observation lands in the same place, whatever its source.
Traceable. Each item keeps a link to the original call, ticket or recording.
Linked. Evidence connects to opportunities, opportunities to decisions, decisions to tickets.
Dated. Old evidence is marked as old, so a two-year-old complaint does not outrank last week's.
A worked example: one discovery cycle in four weeks
The team and figures below are illustrative, not a customer case.
A six-person B2B SaaS team sets an outcome: raise week-one teammate invites from 30% to 40% of new workspaces.
Weeks 1-2, capture. The PM and designer run 8 story-based interviews with recent sign-ups (4 who invited a teammate, 4 who did not) and tag 40 support tickets from the same cohort.
Week 2, connect. Three opportunities emerge. Five of the 8 interviewees did not know which teammate to invite first; 3 worried about exposing unfinished work; 2 said their teammates use a different tool. The team picks the first opportunity because it appears in both segments.
Week 3, act. They sketch three solutions: a suggested-invite list, a "share one page" link, and a role-based invite prompt. The riskiest assumption for the leading idea is value: will people send an invite if the product suggests who? They show a clickable prototype to 5 new users; 4 complete the invite flow without help.
Week 4, decide. The team ships the smallest version to 20% of new workspaces, records the decision with links to the interviews and tests, and sets a review date four weeks out.
Result: one tested bet, two parked opportunities with evidence, and an auditable record.
Copyable template: the discovery record
Paste this into your workspace and keep one record per opportunity:
Outcome (metric, baseline, target, date):
Opportunity (customer situation in their words):
Evidence
- Source link / date / segment:
- Observation (what they did or said):
- Interpretation (labelled as such):
Solutions considered (at least three):
Riskiest assumption and which of the four risks it sits under:
Test (method, sample, pass threshold set in advance):
Result:
Decision (build / change / park / stop) and owner:
What we still do not know:
Review trigger (date or signal):
Linked spec or ticket:
Common mistakes with product discovery frameworks
Treating discovery as a phase that ends. If discovery stops when delivery starts, the next cycle begins with no validated ideas. That is the problem dual-track agile was named to solve.
Counting activity. Ten interviews that change nothing are not discovery. Track how often discovery changes or kills a planned item.
Solutions on the tree without evidence. Every node should link to something a customer did or said.
Treating a small sample as prevalence. A systematic review of 23 studies found interview saturation within 9 to 17 interviews, but mainly for relatively homogeneous groups with narrow questions (Hennink & Kaiser, 2022). Saturation tells you that you have heard the main themes, not how common they are. Use analytics or a survey for frequency.
Where Specky fits
Specky covers the infrastructure part of this framework. It reads customer calls, feedback, Slack, Jira and analytics into a Product Graph, organises evidence under outcomes in an Opportunity Tree, and drafts specs with inline citations back to the customer evidence. Decisions go into a Decision Log with their reasons attached. You still make the calls; Specky keeps the trail.
Frequently asked questions
What is the difference between product discovery and product delivery?
Discovery decides what to build; delivery builds it well. Cagan describes discovery as learning fast and testing ideas before committing engineering time, and delivery as producing software customers can depend on. Most teams run both at once: discovery works on the next set of bets while delivery ships the current one. The handover point is an idea that has evidence against all four risks.
How long should product discovery take?
For a new product or unfamiliar problem, a bounded discovery phase of around 4 to 8 weeks is typical in the UK government model. For an existing product, discovery is continuous: Torres recommends at least weekly customer contact by the team building the product. A single feature question can often be answered in one to two weeks of interviews and a prototype test.
What is the best product discovery framework for a small team?
For a team of two to ten, the lightest complete system is an opportunity solution tree tied to one outcome, weekly story-based interviews, and the four risks as a definition of ready. Add a design sprint only for a single high-stakes question. Founders before product-market fit should start with Blank's customer discovery questions instead.
How many customer interviews do you need for product discovery?
Start with three to four interviews before mapping opportunities, as Torres suggests, and keep going weekly. Research on saturation suggests main themes appear within 9 to 17 interviews for a fairly homogeneous group. If you interview several distinct segments, you need more. Interviews show what happens and why; use data or surveys to learn how often.
Is the opportunity solution tree the same as a product discovery framework?
No. The opportunity solution tree is a visual map that connects an outcome to opportunities, solutions and assumption tests. A full product discovery framework also covers how you capture evidence, what counts as validated, and how decisions get recorded and handed to delivery. The tree is the centre of the framework, not the whole of it.
GDPR-aligned. EU workspace data is stored in Frankfurt, Germany. GDPR and EU data Built in Vienna, Austria. A product of influence solutions fm GmbH (FN 675374y). Imprint 14-day trial. Card required, and nothing is charged until day 14. 30-day refund after subscribing. See pricing