/plea

Opportunity Solution Tree Reference

Purpose: Connect a single business outcome to the experiments that move it, via a visible tree of opportunities and candidate solutions. From Teresa Torres' Continuous Discovery Habits. Forces teams to enumerate opportunities before solutions and to design experiments under solutions — preventing solution-first thinking and output-framed roadmaps. In Plea's hands, the tree is populated with synthetic user opportunities (unmet needs, pain points, desires) channeled from personas, then routed to real teams for prioritization and validation.

Scope Boundary

• plea opportunity: builds an OST anchored on a stated outcome. Outputs the four-layer hierarchy (Outcome → Opportunity → Solution → Experiment) populated with synthetic content.
• plea request (default): list of feature requests, not structured into a tree. Use OST when the question is "where should we invest discovery effort?" rather than "what do users want?"
• plea need: lateral blind-spot scan. OST is hierarchical and outcome-anchored.
• plea jtbd: a JTBD analysis is upstream of OST — the job statement often becomes the outcome or first-level opportunity. JTBD is forces-and-stages; OST is hierarchy-and-experiments.
• plea 5whys: cleans up a single request into a root need. OST organizes many needs against one outcome.
• Field (elsewhere): real-user OST evidence. Field runs ongoing interviews and populates real opportunities from transcripts. Plea generates a synthetic seed tree to compare against.
• Spark (elsewhere): takes a chosen solution branch and produces a feature spec. Plea stops at the experiment level — Spark elaborates the solution into shippable scope.
• Experiment (elsewhere): designs and runs the actual test. Plea drafts experiment hypotheses; Experiment owns sample size, instrumentation, and analysis.
• Voice (elsewhere): real review/support sentiment that should populate opportunity nodes. Plea generates synthetic opportunities to triangulate with Voice.

Workflow

ANCHOR     →  state the outcome — a measurable change in user behavior
           →  reject output framing ("ship feature X") and OKR confusion ("hit target Y")

STRIP      →  enumerate opportunities under the outcome. Each is an unmet need,
           →  pain, or desire — *not* a solution. Phrased in user voice.
           →  cluster duplicates; split fat opportunities into specific sub-opportunities.

BRANCH     →  for each opportunity, sketch 2-4 candidate solutions.
           →  resist anchoring on the first solution that comes to mind.
           →  include "do nothing" or "non-product" options when honest.

EXPERIMENT →  for each solution, design the smallest test that would reduce
           →  the largest risk. Must have: hypothesis, leading indicator,
           →  decision rule (what would make us drop this branch?).

PRUNE      →  rank opportunities by impact × evidence × strategic fit
           →  prune low-evidence + low-impact branches; retain a few high-uncertainty
           →  high-impact branches as exploration bets.
           →  SYNTHETIC-TREE CAVEAT: every node is self-generated, so the evidence axis is
           →  near-zero by construction — do NOT cut a high-impact branch for "low evidence"
           →  (circular). The evidence axis must split "no real evidence yet → validate" from
           →  "implausible AI artifact ([synthetic-only]) → cut". Prune only the latter.
           →  Name the LOAD-BEARING opportunity: highest impact × highest synthetic-uncertainty —
           →  if it's wrong, the most of the tree collapses. Field validates this node FIRST.

CADENCE    →  schedule the tree as a living artifact — weekly review, weekly experiment
           →  treat opportunities as discovered (not invented); add as research arrives.

DELIVER    →  hand tree to Field for population validation,
           →  to Spark for solution elaboration,
           →  to Experiment for test execution.

The Four-Layer Hierarchy

Strict rule: every layer above must constrain the layer below. A solution that doesn't trace cleanly to an opportunity, which traces to the outcome, gets cut. This is the discipline OST enforces.

Outcome Anchoring

A real outcome is:

• Behavioral: something the user does differently
• Measurable: countable from product instrumentation
• Causally connected: improving it should drive business goals

If the stated outcome is wrong, fix it before populating the tree.

Opportunity Stripping

Phrase every opportunity as the user would describe their pain or unmet desire — not as the team would describe a problem.

Cluster duplicates. Split overly broad opportunities (e.g. "the experience is slow") into specific friction points until each is testable.

Solution Branching

For each opportunity, generate at least 2 solution candidates before picking one. Single-solution opportunities indicate anchoring bias.

OPPORTUNITY_NODE:
  text: "[User-voice unmet need]"
  evidence: { real: [refs], synthetic: [refs] }
  calibration: "[hypothesis]"   # [synthetic-only] if it reads as an AI artifact; [supported]/[validated]
                                # only with a cited real/Voice/Field match (reference/calibration.md).
                                # Only [validated]/[supported] nodes are growth-acceptance Insight-Ledger-citable.
  parent_outcome: "[Outcome ID]"
  solutions:
    - id: S1
      summary: "[Intervention]"
      experiments:
        - hypothesis: "[If we ship S1, then [behavior change], because [reason]]"
          riskiest_assumption: "[The thing most likely to break]"
          test: "[Smallest action that tests it]"
          leading_indicator: "[What we'd measure in <1 week]"
          decision_rule: "[Threshold below which we drop this branch]"
    - id: S2
      summary: "[Alternative intervention]"
      experiments: [ ... ]

Experiment Design

Each experiment must answer: what is the smallest test that would reduce the largest risk?

Bad experiments are large, late, and lack a kill rule. Good experiments are small, early, and have explicit failure criteria.

Weekly Cadence (Continuous Discovery Habits)

Torres' habits are weekly, not quarterly:

• Weekly interviews: at least 1 customer interview per week (Field).
• Weekly tree update: add new opportunities, deprecate stale ones, mark experiments closed.
• Weekly experiment: at least one experiment running.
• Weekly opportunity selection: pick the next opportunity to invest in based on evidence and impact.

The tree is a living artifact. A static OST is a roadmap by another name and defeats the purpose.

Anti-Patterns

• Solution-first thinking: starting with a solution and reverse-engineering an opportunity to justify it. Build top-down.
• Output framing: anchoring the tree on "ship X" instead of a behavioral outcome.
• OKR confusion: using a quarterly business KPI as the outcome (too lagging, too far from user behavior). Use a user-behavior outcome that drives the OKR.
• Single solution per opportunity: indicates anchoring; force at least 2 candidates.
• No kill rule on experiments: without decision rules, every branch lives forever — defeats pruning.
• Tree as one-time artifact: built once, never revisited. OST only works as a continuous habit.
• Mixing real and synthetic evidence without tags: synthetic opportunities that look real cause over-confident prioritization. Always tag synthetic: true and a per-node calibration on Plea-generated nodes.
• Skipping the prune step: a 50-node tree without pruning is a wishlist, not a discovery artifact.
• Circular prune on a synthetic tree: cutting a high-impact branch because it has "low evidence" — but YOU generated every node, so all evidence is synthetic. Cut only [synthetic-only] implausible nodes; high-impact [hypothesis] nodes go to Field to validate, not the bin.
• No load-bearing node named: shipping the tree without flagging which opportunity, if wrong, collapses the most branches — Field then validates in the wrong order.

Handoff

• To Field: synthetic tree as scaffolding. Field runs weekly interviews and replaces synthetic opportunity nodes with real-user-evidenced ones, starting with the load-bearing opportunity (highest impact × uncertainty). Tag swap status and promote calibration ([hypothesis] → [supported]/[validated]) as real evidence arrives.
• To Spark: chosen solution branch → feature spec. Plea hands the solution + winning experiment results; Spark elaborates scope.
• To Experiment: experiment hypothesis + riskiest assumption + decision rule → test execution. Experiment owns instrumentation and analysis.
• To Rank: opportunities + impact × evidence × strategic-fit scores → prioritization output.
• To Scribe[unified]: outcome-level statement + retained opportunity branches → spec-package "why" section.
• To Voice: cross-check synthetic opportunity nodes against real review/support text — convergence raises confidence; divergence flags hypothesis weakness or coverage gap.
• To Cast: PERSONA_FEEDBACK on which personas surfaced which opportunity branches — coverage signal for the persona registry.