discover-market-sizing
À propos
Cette compétence estime l'opportunité de marché (TAM, SAM, SOM) en appliquant et en recoupant plusieurs cadres de dimensionnement, tels que les analyses descendantes et ascendantes. Elle produit une fourchette calibrée avec des étiquettes de confiance, mettant en lumière les points de convergence ou de divergence des données comme signaux stratégiques. Utilisez-la pour les dossiers d'investissement, les décisions de lancement ou d'abandon, et les présentations aux parties prenantes, lorsque vous avez besoin d'une analyse de marché structurée et notée selon la fiabilité des sources.
Installation rapide
Claude Code
Recommandénpx skills add product-on-purpose/pm-skills -a claude-code/plugin add https://github.com/product-on-purpose/pm-skillsgit clone https://github.com/product-on-purpose/pm-skills.git ~/.claude/skills/discover-market-sizingCopiez et collez cette commande dans Claude Code pour installer cette compétence
Documentation
Market Sizing
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.
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-briefanddeliver-prd(sizing informs scope and the investment case) - Inputs to this skill often come from:
discover-competitive-analysis(market and competitor context) anddiscover-interview-synthesis(qualitative signal that informs sizing assumptions) - Adversarial review via:
/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.
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
Cross-references
- Companion command:
commands/market-sizing.md - Template:
references/TEMPLATE.md - Examples:
references/EXAMPLE.md+ library samples inlibrary/skill-output-samples/discover-market-sizing/
Dépôt GitHub
Compétences associées
llamaguard
AutreLlamaGuard est le modèle de Meta, doté de 7 à 8 milliards de paramètres, conçu pour modérer les entrées et sorties des LLM selon six catégories de sécurité comme la violence et les discours haineux. Il offre une précision de 94 à 95 % et peut être déployé avec vLLM, Hugging Face ou Amazon SageMaker. Utilisez cette compétence pour intégrer facilement le filtrage de contenu et des garde-fous de sécurité dans vos applications d'IA.
cost-optimization
AutreCette compétence de Claude aide les développeurs à optimiser les coûts du cloud grâce au redimensionnement des ressources, aux stratégies d'étiquetage et à l'analyse des dépenses. Elle fournit un cadre pour réduire les dépenses cloud et mettre en œuvre une gouvernance des coûts sur AWS, Azure et GCP. Utilisez-la lorsque vous devez analyser les coûts d'infrastructure, redimensionner les ressources ou respecter des contraintes budgétaires.
quantizing-models-bitsandbytes
AutreCette compétence quantifie les LLMs en précision 8 bits ou 4 bits à l'aide de bitsandbytes, permettant une réduction de 50 à 75 % de la mémoire utilisée avec une perte de précision minime. Elle est idéale pour exécuter des modèles plus volumineux sur une mémoire GPU limitée ou pour accélérer l'inférence, prenant en charge des formats comme INT8, NF4 et FP4. La compétence s'intègre à HuggingFace Transformers et permet l'entraînement QLoRA ainsi que l'utilisation d'optimiseurs en 8 bits.
dispatching-parallel-agents
AutreCette compétence Claude déploie plusieurs agents pour enquêter et résoudre simultanément 3 problèmes indépendants ou plus. Elle est conçue pour des scénarios impliquant des défaillances non liées qui peuvent être résolues sans état partagé ni dépendances. La capacité fondamentale est la résolution de problèmes en parallèle, en assignant un agent par domaine problématique indépendant afin de maximiser l'efficacité.
