/spark

Spark Opportunity Sizing Reference

Purpose: Produce a defensible estimate of an opportunity's size and confidence before a proposal advances past DISCOVER. Sizing connects a target persona and job-to-be-done to a bounded market, a realistic reachable slice, and a willingness-to-pay signal — so that prioritization frameworks downstream score on real numerators, not wishful reach.

Scope Boundary

• Spark opportunity: sizing the opportunity upstream of scoring — TAM/SAM/SOM, reach estimates, willingness-to-pay signals, opportunity tree mapping.
• vs Rank: Rank orders items with ICE/RICE/WSJF once the numerators exist; opportunity supplies the numerators Rank consumes.
• vs Void: Void prunes scope once sizing exposes that the reachable slice does not justify build cost.
• vs Echo[demand]: Echo[demand] role-plays synthetic users to surface unmet needs; opportunity quantifies the market those needs live in.
• vs Experiment: Experiment validates the hypothesis after a proposal ships as test; opportunity estimates the upper bound on what validation can possibly earn.

If the question is "how big is this bet?" → opportunity. If it is "which of these sized bets do we pick?" → Rank.

TAM / SAM / SOM Sizing

Always produce all three. A TAM-only number is a sales deck, not a proposal.

Require two independent paths to each number: top-down (market report) and bottom-up (funnel × conversion × ARPU). If the two diverge by more than 2x, surface the gap explicitly — do not average them.

Reach × Impact × Confidence

When the opportunity is internal (an existing product feature, not a net-new market), size it in RICE-compatible units so handoff to Rank is clean.

Evidence tiers for Confidence:

• <= 50%: meeting discussion, analogy, gut
• 50–70%: qualitative interviews (N≥5), small-N surveys
• 70–85%: quantitative analytics, prior experiment, large-N survey
• > 85%: live A/B result or shipped-feature telemetry on same audience

Bottom-Up vs Top-Down

Use both, label which is which, reconcile.

Top-down:   TAM -> SAM -> SOM  (market report, then filter)
Bottom-up:  eligible_users × reach_rate × conversion × ARPU × retention

Bottom-up is where product bets live and die. A 1% of 10B TAM line looks strong until the bottom-up funnel says realistic SOM is 40k users at $20 ARPU = $800k — which may or may not clear the bar for build.

Willingness-To-Pay Signals

Size is hollow without demand evidence. Accept these signals, ranked by strength:

If the strongest available signal is "interview enthusiasm", flag the proposal as UNPRICED and require Experiment or a fake-door before Rank scoring.

Market-Timing Assessment

Sizing a correct market at the wrong time produces killed proposals. Assess:

• Why now? — what changed (regulation, platform, cost curve, behavior shift) that makes this viable now but not 2 years ago?
• Why not yet? — what enabling condition is still missing? If the answer is "nothing", the opportunity is likely already contested or already failed by others.
• Window half-life — if this opportunity exists for 6 months, build speed dominates; if 3+ years, platform quality dominates.
• Adjacent-move signals — are larger players signaling entry? (public roadmaps, job listings, acquisitions in the space)

Opportunity Tree Mapping

Use Teresa Torres's Opportunity Solution Tree to connect sized opportunities to outcomes.

Desired Outcome  (KPI-aligned, from OKRs)
    |
    +-- Opportunity A  (pain / desire / unmet need)  ← sized here
    |     +-- Solution A1  (candidate feature)
    |     |     +-- Experiment  (smallest test)
    |     +-- Solution A2
    |
    +-- Opportunity B
          +-- Solution B1

Rules:

• An opportunity is a customer problem statement, never a feature shape.
• Parent opportunity reach is the sum of child opportunity reach ceilings (de-duplicated by user).
• If a child opportunity is reachable but does not move the parent outcome, it belongs on a different tree — surface the tree mismatch rather than forcing it.
• Limit each level to 3–7 branches; more than 7 means the opportunity is not yet decomposed to actionable size.

Sizing Output Template

## Opportunity Sizing: [Opportunity Name]

Target outcome: [KPI from OKR]
Target persona: [segment]
Job-to-be-done: [progress sought, not activity]

Market:
  TAM (top-down):  $[X]  source: [report]
  SAM (filtered):  $[X]  filters applied: [geo / channel / price]
  SOM (12–24mo):   $[X]  bottom-up path: [funnel math]
  Reconciliation:  [explain >2x divergence, if any]

Reach (internal-feature sizing):
  Eligible segment: [N users], window: [quarterly]
  Reach rate assumption: [%], evidence: [tier]

Willingness-to-pay signal: [tier + detail]
Market timing: why-now / why-not-yet / window half-life

OST placement:
  Outcome -> Opportunity -> candidate Solutions [A1, A2]

Confidence: [tier, with evidence]
Blockers to higher confidence: [named evidence gaps]

Anti-Patterns

• Quoting TAM without SAM and SOM — a TAM-only deck hides reachability.
• Averaging top-down and bottom-up when they diverge — record the divergence, do not smooth it.
• Using total registered users as Reach — use segment-specific active users in a consistent window.
• "1% of the market" assumptions — always derive SOM from a funnel, never from a percentage pulled from thin air.
• Accepting survey "would you pay?" as willingness-to-pay — discount heavily or re-route to a fake-door test.
• Sizing a feature (activity) instead of an opportunity (progress) — features are solutions; opportunities are problems.
• Flat 80% confidence because "we interviewed some users" — map to the evidence tier explicitly.
• Opportunity trees that retrofit existing roadmap items — if every child solution is already being built, this is confirmation, not discovery.
• Ignoring market timing — a correctly sized opportunity at the wrong time still kills.

Handoff / Next Steps

• If SOM clears the bar and WTP signal is Strong or above → hand to Rank for RICE/WSJF scoring against peers.
• If SOM is ambiguous but WTP signal is weak → hand to Experiment for a fake-door or Van Westendorp before scoring.
• If SOM is small but strategic (wedge into larger market) → hand to Magi for Go/No-Go with explicit strategic rationale.
• If opportunity tree shows the parent outcome is not moved by any reachable solution → hand to Void to prune and re-frame.
• If willingness-to-pay requires synthetic user probing before survey design → hand to Echo[demand].

Record the sized opportunity in .agents/spark.md under phantom/underused features so future proposals inherit the sizing work.