Pin all nine blocks, then test whether they hold up.
Describe a product, its customers and how it operates today. Your assistant returns a complete Business Model Canvas, checks that the blocks support each other, runs an LTV > 3× CAC viability test and lists the assumptions that could break the model.
Key Partners
- Two produce suppliers
- Kitchen landlord, courier
- Mostly single-source: a risk
Key Activities
- Menu planning, kit production
- Monday cold-chain delivery
Key Resources
- Leased kitchen (14 months left)
- Van, ordering app
Value Propositions
- Ends the “part-time job” of daily catering
- $11.50 vs $16 hot catering (~28% gap, derived)
- Gap: no flex headcount yet
Relationships
- Self-service app plus office manager
- ~6% monthly churn
Channels
- Manager sign-up; 71% of orders in-app
Customer Segments
- 20–45 person studios, no kitchen staff
- Poor fit: under 15 people or hybrid <3 days
Cost Structure
- Implied ~$7.94 cost per meal (69% of $11.50, derived)
- Wage rise and packaging rule ahead
Revenue Streams
- $11.50 per meal, 40-meal weekly minimum
- ~$850 gross profit per office per month (derived)
What it is
A single instruction file that turns a capable AI assistant into a business model strategist. It works through all nine canvas blocks with guiding questions for each, from partners and activities on the left, through the value proposition in the centre, to channels, relationships and segments on the right, with costs and revenue underneath.
It does not stop at filling boxes. The Skill asks for an alignment check across the blocks, a viability test against LTV > 3× CAC and a list of key assumptions and risks. It also explains where the canvas falls short and when Lean Canvas or Startup Canvas fits better.
Who it is for
- Product managers documenting how an existing product makes money.
- Founders and operators checking whether unit economics survive scrutiny.
- Strategy teams preparing investor or board material that shows how the pieces connect.
- Students learning the Osterwalder canvas with worked reasoning.
Works well for
- Documenting an established business model end to end
- Analysing an existing model for weak links
- Investor or corporate-strategy material
- Pressure-testing unit economics before expansion
Not designed for
- Product vision, trade-offs or defensibility (the Skill itself says the canvas lacks these)
- Financial forecasts, valuations or investment advice
- Live market data: it works from what you supply
- Replacing customer research or accounting
What you provide, what you receive
You provide
- Product or service description
- Target customers and market
- Current operations or working assumptions
- Competitive context or industry dynamics
- Prices, volumes, churn or margin if you have them
You receive
- All nine blocks, each answered against its guiding questions
- An alignment check across the blocks
- An LTV > 3× CAC viability test
- Key assumptions and risks
- Notes on canvas limits and alternatives
Where it runs
Any AI assistant that can read a Skill file, including Claude Code and Codex. Download the free plaintext package from Finch, install it with your assistant’s Skill method, or ask the assistant to read SKILL.md.
No extra API key, script or service. You still need access to an assistant model, which may carry its own subscription or usage cost. The Skill has no built-in web research; it works from your notes.
Limits and licence
The canvas is a structured judgment, only as good as the facts supplied. Check every derived figure before relying on it.
MIT licence, Copyright (c) 2026 Pawel Huryn. Source: phuryn/pm-skills at commit 2a12295a7717, path pm-product-strategy/skills/business-model. The package adds a README, NOTICE, provenance note and a cover; the instructions are unchanged.
How it was checked
One controlled run on a fictional company (Claude Code only) produced all 13 parts the instructions require, including the viability test, which it expressed as a CAC ceiling because CAC was unknown. Re-checking the arithmetic found one slip: an annualized revenue figure. The published package was downloaded back from Finch and matched the audited files. Native installation in each assistant and real business outcomes were not tested. The Benchmark tab has the full record.