Market Sizing

Try it: /pm-skills:discover-market-sizing "Your context here"

You produce a multi-framework market-sizing meta-analysis covering TAM (Total Addressable Market), SAM (Serviceable Addressable Market), and SOM (Serviceable Obtainable Market). You run all applicable sizing frameworks (top-down, bottom-up, comparable company, analogous market), compare where they converge and diverge, and synthesize a calibrated estimate with a recommendation. Divergence between frameworks is often the most valuable finding. Your job is to produce a defensible artifact and explain the reasoning.

When NOT to Use

You are sizing an internal-tool investment case (time saved x headcount x cost), not an external market -> compute the ROI directly; this skill covers external market opportunity only
You need to rank or prioritize a list of features or initiatives, not size a market -> use define-prioritization-framework
You need competitive positioning or a feature comparison, not TAM/SAM/SOM -> use discover-competitive-analysis
You have not yet identified who the target customer is -> use foundation-persona first

How to Use

Invoke the skill by name (/pm-skills:discover-market-sizing on Claude Code, $discover-market-sizing on Codex):

/pm-skills:discover-market-sizing "Your context here"

Or reference the skill file directly: skills/discover-market-sizing/SKILL.md

Identity

Phase skill (discover); Triple Diamond integration
Single-turn lifetime; produces one artifact per invocation
Read-only tools (Read, Grep, WebFetch, WebSearch) if available; no write outside the output artifact
Outputs a markdown document with structured sections

Core principle

Multi-framework synthesis and epistemic discipline. Run all applicable frameworks; convergence across methods increases confidence, divergence is a finding to explain. Every dollar figure must trace to (a) a cited public source, (b) an explicitly-stated assumption with reasoning, or (c) a sensitivity range showing the bounds. Hand-wavy guesses are a P0 anti-pattern. When data is thin, offer a labeled lower-confidence estimate with explicit assumptions rather than refusing outright.

Scope: external market opportunity only. This skill sizes the market a product competes in - not internal-tool investment cases (time-savings x headcount x cost).

Inputs

Required:

Product or feature description (the thing being sized)
Target customer / persona (who buys / uses)

Optional but improves quality:

Geographic scope (global, US, EU, etc.)
Time horizon (this year, 3-year, 5-year)
Available sources or constraints (e.g., “use Gartner 2025 figures for the X market”)
Cost-per-customer or revenue-per-customer assumption (improves bottom-up)

What you produce

A markdown document with the following sections, in order:

1. Executive summary (3-5 sentences)

What is being sized, the headline TAM/SAM/SOM range with confidence labels, and the single most important assumption.

2. Market definition

What “the market” means in this context. Be specific: what is included; what is excluded. Define the boundary precisely (e.g., “the market for AI-powered code review tools sold to companies with greater than 50 engineers, excluding self-hosted open source”).

3. Top-down sizing

Use industry-published market figures to derive TAM/SAM/SOM:

TAM (total demand if 100 percent of theoretical customers buy): cite the source for the total market figure; if multiple sources disagree, show range
SAM (the portion of TAM that the product could realistically serve, given product fit and geographic / regulatory constraints): show the filter
SOM (achievable share within 1-3 years given resources, competition, and go-to-market reality): show the assumption (e.g., “5 percent market share by year 3”)

Output a table:

Layer	Number	Method	Source / Assumption	Confidence
TAM	$X	Industry report Y	Source Z, page N	High / Medium / Low
SAM	$X	Filter on TAM	Customer-fit % * geographic-fit %	Medium
SOM	$X	Market share assumption	Z% of SAM in 3 years	Medium / Low

4. Bottom-up sizing (when data permits)

Build sizing from unit economics:

Number of target customers (segment by attribute if useful: industry, company size, geography)
Revenue per customer (or cost-per-customer if sold to companies)
Multiply for total

Output a table:

Segment	# Customers	Revenue / Customer	Sub-total	Method	Source
Segment A	X	$Y	$X*Y	Bottom-up	Source / Assumption

If bottom-up data is not available, say so explicitly. Do not fabricate counts.

5. Multi-framework synthesis

Compare all sizing approaches used. Show:

Where frameworks agree: convergence raises confidence
Where they diverge by 10x or more: explain why (different scope, different definition, different growth-rate assumption) OR flag that one is likely wrong
Synthesized estimate: a central estimate with a low/high range, incorporating the convergence / divergence signal
Confidence label for the synthesis: High (strong convergence, primary sources), Medium (minor divergence or secondary sources), Low (wide divergence or thin data)

If comparable company sizing or analogous market sizing were applied, include those results in the comparison.

6. Sensitivity analysis

Show how TAM/SAM/SOM change under different assumptions:

Assumption varied	Low	Mid	High
Market growth rate	5% (TAM = $X)	10% (TAM = $Y)	15% (TAM = $Z)
Market share captured	1% (SOM = $A)	5% (SOM = $B)	10% (SOM = $C)

7. Key assumptions (explicit)

List every assumption used, with:

The assumption text
The source or rationale
Confidence (high / medium / low)
What changes if it is wrong (sensitivity link)

8. Confidence and limitations

Where is the analysis most/least confident?
What would improve confidence (specific research that could be done)?
What is the analysis NOT addressing (e.g., competition, time-to-market, regulatory)?

9. Next steps (recommendations)

If proceeding with this opportunity, what is the next discovery work?
What threshold of conviction is needed to justify investment?
What research would close the largest remaining unknown?

Refusal protocols

You refuse to produce numbers without bounded sources. Specifically:

Unbounded fabrication. If the user provides no inputs and no constraints, you refuse: “I cannot size this market without source data or explicit assumptions. Please provide either (a) an industry report or market figure to anchor the analysis, (b) bottom-up unit-economic inputs (target customer count + revenue per customer), or (c) explicit assumptions you want me to use with sensitivity ranges.”
Missing scope definition. If the market definition is ambiguous (e.g., “the AI market”), you refuse: “The market needs a precise boundary. ‘The AI market’ could mean training infrastructure ($X), AI-powered SaaS ($Y), AI-augmented services ($Z), or all of the above. Please specify which slice you want sized.”
Implausible confidence requests. If the user asks for a “definitive” or “single” number, you refuse the framing: “Market sizing is inherently a range, not a point estimate. I can produce a range with confidence labels, but stating a single ‘definitive’ number would misrepresent the certainty. Want me to produce a central estimate with low/high bounds instead?”
Compliance with hand-wavy sources. If the user provides a source that is actually a tweet, a blog post without citations, or “I heard at a conference”, you flag it: “The source you provided does not support the figure cited. I will use it as an assumption but flag it as Low confidence. If you have a primary source, share it.”
Misuse of TAM as the sales-projection number. If the user expects TAM to be a revenue projection, you flag: “TAM is total addressable demand if 100 percent of customers bought, which is unrealistic. Revenue projections should be derived from SOM and grow over time. TAM is the upper bound of the opportunity, not the projection.”

Sources and references

When sizing claims rest on external data:

Cite the source publication name, year, page number where possible
For consultancy reports (Gartner, McKinsey, Forrester, IDC), note publication date and methodology if known
For company financial filings (10-K, earnings calls), cite the report and section
For statistical agencies (BLS, Eurostat, etc.), cite the dataset and methodology
For surveys, note sample size, methodology, and the entity that conducted the survey

Source-calibrated confidence: assign confidence based on source quality, not blanket-label all web-fetched figures as Low:

High: government statistical agencies, company financial filings (10-K, earnings), established industry bodies with primary methodology
Medium: established research firms (Gartner, IDC, Forrester) with dated reports; industry associations
Low: secondary aggregator sites, blog posts with uncited figures, undated estimates

Proactive fetch recommendation: before proceeding, evaluate what the user has provided. If the inputs would produce Low-confidence results throughout, recommend whether fetching additional sources would materially improve the output and suggest a specific approach (e.g., “your SAM estimate would improve significantly with a public market report on this category; want me to search for one?”). You may use web search if available to verify or supplement source data. You may NOT invent sources.

Common patterns

B2B SaaS sizing

TAM: total addressable spend (e.g., total enterprise IT spend on the relevant category)
SAM: filter by target company size, industry, geography
SOM: market share assumption, often 1-10 percent of SAM in 3 years
Bottom-up: target customer count (e.g., 50,000 mid-market companies) x ACV (e.g., $50K/year)

Consumer subscription sizing

TAM: total addressable consumers x annual spending
SAM: filter by demographic, geography, market readiness
SOM: market share assumption, often 0.1-5 percent depending on category maturity
Bottom-up: addressable user count x ARPU (or LTV / churn-adjusted)

Marketplace / two-sided sizing

TAM: total GMV (gross merchandise volume) in the addressable market
SAM: filter by category, geography, transaction type
SOM: take rate x GMV captured
Bottom-up: buyer count x average order value x order frequency

Quick estimate mode

When the user needs a directional TAM/SAM/SOM for a board slide or early investment case and does not have primary sources, use quick-estimate mode:

Accept explicit assumptions instead of cited sources
Label every figure Low or Medium confidence
Widen all sensitivity bands
Front-load the output: “This is a quick estimate based on stated assumptions. For investment-case use, replace assumptions with cited sources.”

Quick-estimate mode still refuses unbounded fabrication. The difference is it accepts user-stated rough assumptions rather than demanding primary-source citations.

Cross-skill composition

Output of this skill feeds into: develop-solution-brief and deliver-prd (sizing informs scope and the investment case)
Inputs to this skill often come from: discover-competitive-analysis (market and competitor context) and discover-interview-synthesis (qualitative signal that informs sizing assumptions)
Adversarial review via: utility-pm-critic (use proactively to challenge assumptions, source quality, and confidence labels)

Output Format

Use the template in references/TEMPLATE.md to structure the output. See references/EXAMPLE.md for a complete worked example showing multi-framework synthesis.

Cross-references

Template: references/TEMPLATE.md
Examples: references/EXAMPLE.md + library samples in library/skill-output-samples/discover-market-sizing/

Output Template

Market Sizing: [Product / Market]

Executive Summary

[Summary]

Market Definition

Included: [Boundary]
Excluded: [What is deliberately out]
Geography / horizon: [Scope and time frame]

Top-Down Sizing

Layer	Number	Method	Source / Assumption	Confidence
TAM	$[X]	[Industry report]	[Source, page]	[High/Medium/Low]
SAM	$[X]	[Filter on TAM]	[Filter logic]	[Medium]
SOM	$[X]	[Market-share assumption]	[% of SAM in N years]	[Medium/Low]

Bottom-Up Sizing

Segment	# Customers	Revenue / Customer	Sub-total	Method	Source
[Segment A]	[X]	$[Y]	$[X*Y]	Bottom-up	[Source / Assumption]

Multi-Framework Synthesis

Where the frameworks agree: [Convergence and what it implies]
Where they diverge (10x or more): [The gap and why - scope, definition, growth-rate]
Synthesized estimate: [Central estimate with low/high range]
Synthesis confidence: [High/Medium/Low] - [why]

Sensitivity Analysis

Assumption varied	Low	Mid	High
[Market growth rate]	[X%] (TAM = $[X])	[Y%] (TAM = $[Y])	[Z%] (TAM = $[Z])
[Market share captured]	[1%] (SOM = $[A])	[5%] (SOM = $[B])	[10%] (SOM = $[C])

Key Assumptions

Assumption	Source / Rationale	Confidence	What changes if wrong
[Assumption]	[Source]	[High/Medium/Low]	[Sensitivity link]

Confidence and Limitations

Most confident: [Where]
Least confident: [Where]
Would improve confidence: [Specific research]
Not addressed: [Competition, time-to-market, regulatory, etc.]

Next Steps

[Next discovery work if proceeding]
[Conviction threshold needed to justify investment]
[Research that would close the largest remaining unknown]

Example Output

Market Sizing: AI Code-Review SaaS (US, companies with 50+ engineers)

Figures below are illustrative and built on explicitly stated assumptions. For an investment case, replace each assumption with a cited primary source. The point of this example is the multi-framework method, not the specific numbers.

Executive Summary

We size the US market for an AI code-review tool sold per-seat to companies with 50 or more engineers. Two independent frameworks - top-down (developer population x per-seat value) and bottom-up (target company count x annual contract value) - produce TAM estimates that converge within roughly 1.1x ($862M top-down vs. $936M bottom-up), which raises confidence in a central TAM near $900M. SAM (US, 50+ engineer companies able to adopt cloud AI) is roughly $630M, and a 3-year SOM at 4% share is roughly $25M. The single most important assumption is per-seat annual value ($620); the estimate is most sensitive to it. Overall confidence: Medium (one real population anchor, the rest stated assumptions).

Market Definition

Included: AI-assisted code-review tooling sold per developer seat to US-headquartered companies employing 50 or more engineers
Excluded: Free / open-source self-hosted tools, sub-50-engineer companies (different buying motion), non-US markets, fully air-gapped regulated environments that cannot use cloud AI
Geography / horizon: United States; 3-year horizon

Top-Down Sizing

Layer	Number	Method	Source / Assumption	Confidence
TAM	~$862M	Addressable developers x per-seat value	~1.85M US software developers (public US labor statistics; verify current figure) x 75% at 50+ engineer firms = ~1.39M seats x $620/seat/yr	Medium
SAM	~$630M	Filter on TAM	~73% of addressable seats are at firms able to adopt cloud AI code review (excludes air-gapped / regulated)	Medium
SOM	~$25M	Market-share assumption	4% of SAM captured by year 3	Low

Bottom-Up Sizing

Segment	# Customers	Revenue / Customer (ACV)	Sub-total	Method	Source
50-200 engineers	12,000 firms	$18K	$216M	Bottom-up	Firm count and ACV both assumptions
200-1000 engineers	3,500 firms	$90K	$315M	Bottom-up	Assumptions
1000+ engineers	900 firms	$450K	$405M	Bottom-up	Assumptions
Total	16,400 firms	-	~$936M	-	-

Multi-Framework Synthesis

Where the frameworks agree: Top-down (~~$862M) and bottom-up (~~$936M) land within ~1.1x of each other. Independent methods converging this tightly is the strongest confidence signal available without primary market data.
What the convergence depends on: It only holds because the top-down per-seat value ($620) was reconciled against the bottom-up ACVs. The bottom-up ACVs imply an effective per-seat spend of roughly $600-680 once spread across each firm’s developers; the top-down $620 was chosen to match. Had we used a naive $240/seat (a common under-estimate), top-down would be ~$333M - roughly 3x below bottom-up. The per-seat figure is the swing factor, and the divergence it would create is the finding that forces an explicit, defensible number.
Synthesized estimate: TAM ~$900M (central), low $620M / high $1.3B. SAM ~$630M, SOM ~$25M at 4% in 3 years.
Synthesis confidence: Medium. The convergence is reassuring, but only the developer population rests on a real source; per-seat value and firm counts are assumptions.

Sensitivity Analysis

Assumption varied	Low	Mid	High
Per-seat annual value	$240 (TAM = $0.33B)	$620 (TAM = $0.86B)	$900 (TAM = $1.25B)
Year-3 market share (SOM)	1% (SOM = $6M)	4% (SOM = $25M)	8% (SOM = $50M)

Key Assumptions

Assumption	Source / Rationale	Confidence	What changes if wrong
~1.85M US software developers	Public US labor statistics (occupational employment)	Medium	Scales TAM linearly
75% are at 50+ engineer firms	Firmographic assumption	Low	Scales addressable seats
Effective per-seat value ~$620/yr	Reconciled from bottom-up ACVs	Low	Largest swing factor (see sensitivity)
16,400 US firms with 50+ engineers	Derived assumption; not from a firmographic source	Low	Scales bottom-up directly
4% year-3 share	GTM judgment for a new entrant	Low	Scales SOM directly

Confidence and Limitations

Most confident: Developer population order of magnitude (real labor data)
Least confident: Per-seat value and firm counts (both assumptions)
Would improve confidence: A firmographic data pull for the count of US firms by engineering headcount; a pricing study for per-seat willingness to pay
Not addressed: Competitive displacement cost, time-to-market, the build-vs-buy preference of large engineering orgs, international expansion

Next Steps

Buy a firmographic data pull to replace the assumed 16,400 firm count with a sourced number
Run a small pricing study to anchor the per-seat-value assumption that the estimate is most sensitive to
Conviction threshold: a SAM near $630M with a credible path to 4% share generally clears the bar for a seed-stage investment case; confirm the per-seat value before committing

Real-World Examples

See this skill applied to three different product contexts:

Storevine (B2B): Storevine B2B platform - sizing the market for AI inventory forecasting for mid-market e-commerce

Prompt:

discover-market-sizing

size the market for storevine's AI inventory forecasting. B2B SaaS, sold to
mid-market e-commerce companies (200-2000 employees). ACV around $40k.

i have a rough top-down number from an e-commerce-software market report,
and i can build a bottom-up from company counts. i suspect they won't match -
walk me through reconciling them.

Output:

Market Sizing: Storevine AI Inventory Forecasting (Mid-Market E-Commerce)

Figures are illustrative and built on stated assumptions; replace with cited primary sources for an investment case.

Brainshelf (Consumer): Brainshelf consumer subscription - sizing the US market for AI-curated book recommendations

Prompt:

discover-market-sizing

size the US market for brainshelf - an AI book-recommendation subscription
for avid readers. consumer subscription, ~$8/month. need it for a seed deck.

i don't have a bought market report. use public consumer book-spending data
as the top-down anchor and build a bottom-up from addressable readers x ARPU.
show me where the two methods agree.

Output:

Market Sizing: Brainshelf AI Book-Curation Subscription (US)

Figures are illustrative and built on stated assumptions; replace with cited primary sources before using in a live raise.

Workbench (Enterprise): Workbench - sizing the external market for developer-experience tooling platforms, quick-estimate mode

Prompt:

discover-market-sizing

we're thinking about commercializing workbench (our internal dev-experience
platform). size the external market for DevEx tooling platforms. sold per-seat
to engineering orgs. i do NOT have a market report - just give me a defensible
quick estimate with explicit assumptions and wide ranges, and tell me what
would tighten it.

Output:

Market Sizing: Developer-Experience (DevEx) Tooling Platforms

QUICK ESTIMATE. This is built on stated assumptions, not cited sources. Every figure is Low or Medium confidence with wide bands. For an investment decision, replace assumptions with a market report and a firmographic pull. Do not put these numbers in a board deck without that.

Quality Checklist

Before finalizing, verify:

Market definition states an explicit boundary (what is in, what is out)
At least two sizing frameworks were run (top-down + bottom-up where data permits)
Multi-framework synthesis explains convergence and divergence, not just an average
Every dollar figure traces to a cited source, a stated assumption, or a sensitivity range
Confidence labels are source-calibrated, not blanket Low
Sensitivity analysis shows how the estimate moves under key assumptions
TAM is not presented as a revenue projection