Product & Project Delivery
Build an Opportunity Solution Tree from Research Evidence
A completed fictional dispatch-planning tree links four story records to two customer needs and three possible solutions. Two branch corrections remove an unsupported feature requirement and a motive claim while preserving the next open evidence decisions.
Build an opportunity solution tree by connecting one defined outcome to customer needs supported by research, then to several possible solutions and their unresolved assumptions. Keep the evidence behind each opportunity visible. A requested feature belongs among the solution candidates until you can explain the underlying need; a branch without supporting research should not become a delivery commitment.
This LandOffer.ai guide shows a completed fictional tree, its evidence links and two branch corrections. The customer stories, product, metric and decisions are authored examples. No customer interviews, product analytics, experiments or production changes were performed for this article. The result is a reviewable discovery record, not proof that a feature will improve a business outcome.

Set the outcome before drawing branches
Teresa Torres’s Product Talk guide to opportunity solution trees, dated December 6, 2023, connects an outcome, customer opportunities, possible solutions and assumption tests. It distinguishes a customer need from the feature offered to address it, recommends starting from customer stories, and treats the tree as an evolving team document. Its suggested starting inputs include a defined audience, an outcome and three or four story-based interviews; that is practitioner guidance, not a statistical sample-size guarantee.
Product Talk’s outcome guide, dated July 17, 2024, distinguishes what a team produces from the effect it wants that work to have. Use that distinction when naming the root.
Start your own record with a metric definition and an accountable owner. “Launch an acknowledgment dashboard” commits to an output before comparing alternatives. “Make scheduling easier” leaves the team unable to judge which branch matters. A useful outcome names the behavior, population and measurement window while leaving several ways to help the customer.
For the fictional ReedHarbor dispatch product, product manager Nessa, designer Arun and engineer Jo focus on depot supervisors preparing next week’s driver shifts. The chosen product outcome is the percentage of eligible weekly plans confirmed by Thursday at 5 p.m., using each depot’s local time. A confirmed plan has every required shift assigned and every required driver acknowledgment recorded against the current revision.
The scenario’s invented baseline is 52 confirmed plans out of 80 eligible plans, or 65%. The team proposes a 72% target for a future review period. Neither figure comes from a real customer account. A denominator note excludes training depots and records the eligibility rule. Preserve that definition when comparing periods, or a change in the eligible population could look like improved confirmation.
Leadership believes earlier confirmation could reduce last-minute dispatch coordination. That business connection is a hypothesis. The tree does not establish lower labor costs, fewer missed shifts or increased revenue merely because its product metric is well defined. Record the proposed target separately from any later measurement.
Carry specific research evidence into each opportunity
Nessa’s input pack contains four synthetic story summaries. Each describes a particular past event rather than a feature preference. In actual work, retain the authorized source, date, participant context and location within the recording or notes; the short IDs below illustrate traceability without inventing real research participants.
| Evidence ID | Authored story context | Observed difficulty inside the fictional story | What the record does not establish |
|---|---|---|---|
| Q1 | Supervisor at a small depot checks Thursday’s plan | She opens each shift separately to find two missing acknowledgments, then writes a personal follow-up list | That all supervisors need automatic reminders |
| Q2 | Supervisor revises a Wednesday assignment | A driver responds to an older exported copy; the supervisor compares timestamps to identify the current plan | That acknowledgment delay caused every late plan |
| Q3 | Supervisor inherits a colleague’s Thursday handover | The notes do not distinguish acknowledged shifts from people still awaiting contact | That drivers refused shifts or ignored a message |
| Q4 | Supervisor checks a printed sheet after a change | The sheet lacks a visible revision reference, so he calls the dispatcher before acting | That a mobile application is the required remedy |
Q1 and Q3 support a candidate opportunity about recognizing unresolved acknowledgments. Q2 and Q4 support a different opportunity about recognizing the current revision. These are reasoned interpretations of the authored stories, not participant quotations or evidence of prevalence across the customer population.
The four stories do not produce an opportunity-size estimate. Two mentions in this deliberately constructed pack are not “50% of customers,” and the 80-plan baseline is not an interview denominator. Mixing those populations would create a persuasive number without a defensible meaning.
Keep contradictory or incomplete details beside the interpretation. Q3 concerns a handover between supervisors; its difficulty may involve ownership as well as acknowledgment visibility. The tree should retain that uncertainty instead of flattening the event into whichever problem the team already wants to solve.
A completed tree with one selected opportunity
The team creates the following version 1 record. Indentation represents parent–child links. “Candidate” means worth exploring, not validated; the tree is complete as a documented decision structure even though its proposed tests remain unrun.
OUTCOME O1: Increase eligible plans confirmed by Thursday 17:00
Metric: 52/80 = 65% illustrative baseline; proposed target 72%
Audience: depot supervisors preparing next week's driver shifts
OPPORTUNITY P1: I cannot identify which required acknowledgments remain
Evidence: Q1 individual checks; Q3 ambiguous handover notes
Selected for the next comparison, with ownership uncertainty retained
S1: A daily supervisor status digest
A1: Required statuses can be summarized without stale assignments
Next evidence: inspect synthetic revision/status cases with Jo
S2: An unresolved-acknowledgment queue in the existing planning view
A2: Supervisors can identify the next action without opening every shift
Next evidence: task-based prototype comparison with Arun
S3: A structured supervisor follow-up worksheet
A3: Handover owners can maintain the worksheet without duplicate tracking
Next evidence: bounded workflow walkthrough with Nessa
OPPORTUNITY P2: I cannot tell which plan revision is current
Evidence: Q2 older export; Q4 printed sheet without revision reference
Retained, not selected for this comparison
Solution exploration deferred; no build commitment
OUTSIDE CURRENT TREE: automatic reminder feature request
Candidate solution only; underlying need and notification risks unresolved
P1 is selected because the team can compare visibility and handover approaches within its current remit. P2 remains supported and important. Deferring it does not mean its users matter less or that the revision problem is solved. The selection note should describe the actual reason rather than assigning unsupported precision to a scoring matrix.
The team also records a dependency: a digest based on obsolete assignments could misrepresent what remains unresolved. That concern links P1 to the revision context in P2 without merging two different needs into a single vague “communication problem.” One solution might eventually address both, but its arrows would still need explanations.
All three P1 candidates offer different ways to support the same task. None requires a new driver application as a starting assumption. The candidate descriptions expose different timing, maintenance and access questions: a scheduled summary can become stale, a queue needs reliable status definitions, and a manual handover aid needs an update owner.
Two branch corrections that change the decision
The first draft had more confident language than its inputs justified. Rather than erase the history, Arun retains two removed labels and explains what replaced them. This correction record is a substantive part of the tree, because it changes which work the team considers ready for exploration.
| Original branch | Evidence check | Completed correction | Consequence for the next decision |
|---|---|---|---|
| “Drivers need automatic SMS reminders” under opportunities | Q1 supports the supervisor’s difficulty finding missing acknowledgments; it says nothing about driver channel preference, consent or reminder timing | Replace the opportunity with P1. Move SMS reminders outside the selected candidate set, labeled an unexamined solution request | Do not turn the branch into an SMS implementation ticket; first establish the relevant user task and constraints |
| “Drivers do not care about confirming shifts” under P1 | No story establishes motivation; Q3 only describes ambiguous handover notes | Delete the motive claim. Add an uncertainty note: contact ownership and status meaning need clarification | Compare visibility and ownership approaches without blaming drivers or interpreting silence as refusal |
These are two different failures. The first confuses a solution with a need. The second assigns a motive to people who have not supplied that explanation. Rewording both as “improve engagement” would hide the problems instead of correcting them.
Nessa keeps the original request in a separate request log so its origin remains available. Removing it from the opportunity layer does not require pretending nobody asked for it. If later research supports a reminder-related need, the team can revisit the idea with the new evidence and date.
The deletion also affects presentation. The stakeholder view no longer shows an automatic-message branch as an approved route to the 72% target. It shows a selected supervisor need, three alternatives and named uncertainties. That presentation lets stakeholders see what the team has selected and what remains undecided.

Attach assumptions without claiming tests have passed
Product Talk’s assumption-testing guide, dated October 18, 2023, recommends making the beliefs beneath a solution specific and evaluating the ones that carry material risk. It distinguishes checking an individual assumption from testing an entire idea, and discusses desirability, feasibility, usability, viability and ethical considerations. The categories help expose missing questions; this article does not treat category labels as an approval process.
For this tree, Jo owns A1 and prepares three authored revision/status cases for review. The open question is whether a digest could show the right assignment revision and acknowledgment meaning. No existing integration is assumed to provide that information, and no engineering feasibility finding is recorded yet.
Arun owns A2 and drafts a prototype task: locate unresolved required acknowledgments for one synthetic plan and explain the next action. The comparison must distinguish recognizing a status from completing a follow-up. A participant finding a row would not demonstrate that a real plan becomes confirmed earlier.
Nessa owns A3 and drafts a supervisor handover walkthrough. It must examine who updates the worksheet, when updates stop being reliable and whether the worksheet duplicates an existing record. A manual option is not automatically cheap if people must reconcile conflicting copies every day.
The version 1 record marks A1, A2 and A3 “proposed; not run,” with Jo, Arun and Nessa named as their respective owners. It records dependencies before scheduling: authorized participants, permitted data, agreed task materials and the relevant accessibility or access needs. The synthetic example contains no driver absence reasons, medical details or private contact information. Actual research should use the organization’s approved handling and access rules.
This is enough to explain why the next step is a comparison rather than a build request. It is not a completed test protocol or a claim that any candidate is effective. Detailed test design belongs in its own record, with a return link to the assumption and the exact tree version it informed.
Review the tree as evidence changes
At the stopping point, version 1 contains one defined outcome, two supported candidate opportunities, three possible solutions for P1, two documented corrections and three unrun assumption activities. No solution has been selected for delivery. P2 remains on the tree, and the proposed 72% target has not been achieved or measured.
When another story arrives, retain its context before deciding whether it supports an existing branch, contradicts an interpretation or suggests a separate need. If an assumption activity produces evidence, record what happened and which candidate it affects. Do not turn a positive task observation into a forecast of the overall outcome.
Further workplace guides are available from LandOffer.ai. Take the current tree into the next team review with its evidence IDs, correction history and open assumptions. Ask reviewers to confirm P1, the three candidate descriptions and the owners of A1–A3. Record any disagreement against the relevant evidence ID, and keep delivery and outcome claims pending until their own evidence exists.
Sources and Further Reading
Teresa Torres, Product Talk: Opportunity Solution Trees — original practitioner explanation of the method; December 6, 2023; checked October 28, 2026.
Teresa Torres, Product Talk: Assumption Testing — assumptions beneath solution candidates and bounded evidence gathering; October 18, 2023; no tests performed here.
Teresa Torres, Product Talk: Shifting from Outputs to Outcomes — output versus intended impact; July 17, 2024; checked October 28, 2026.


