ai-native-business-model-canvas

Designs the shift from an AI-enhanced business (AI bolted onto existing operations) to an AI-native one, using an extended Business Model Canvas with four AI-specific lenses: value proposition, data m

SKILL.md

GitHub

AI-Native Business Model Canvas

Purpose

Designs the transition from a traditional "AI-enhanced" business (a business improved by AI) to an "AI-native" business model — a business model built on AI from the ground up, whose entire value proposition depends on the ML ecosystem.

Anchored in research

  • A research report supplied by the user, "AI Business Designer in the Age of AI" (2026) — the extended AI Business Model Canvas
  • Business Model Canvas (Osterwalder & Pigneur) — the base structure this framework extends with four AI-specific lenses
  • A research report supplied by the user, "Methods, Frameworks, and Competencies for Identifying AI Opportunities and Capacity in Business" (2026) — the "AI Value Canvas" (the same four-lens structure under different names: Unique AI Value Proposition, Data Moat & Flywheel, Human-AI Interaction Model, Cost of Compute vs. Marginal Revenue). This report confirmed three lenses that already existed (points 1–3 below) and explicitly surfaced a fourth — the Human-AI Interaction Model (point 4 below) — which this skill hadn't yet treated as its own point.
  • The proprietary learning-engine principle in Method step 6 (added later) is grounded in current AI-moat/data-engine discourse (Y Combinator's "7 Real Moats for AI Startups," McKinsey's "From AI Table Stakes to AI Advantage," among others), consistently making the same point: proprietary, accumulating feedback loops are the durable moat, not the underlying algorithms, which commoditize faster than data streams do. Recursion Pharmaceuticals is used as the reference case — see ../../references/ai-native-reshuffle-heuristics-research.md for the full grounding and why this was added as a deepening of an existing step rather than a new skill.

Method

  1. Value Proposition. Determine how AI personalizes, scales, or creates new value in real time — not just speeds up an existing process. Example: a hyper-personalized learning platform that adapts content and tone to the user's emotional state. Test: if AI were removed, would the value proposition collapse, or would it just be slower? (The latter = AI-enhanced, not AI-native.)
  2. Key Resources. Identify proprietary data — unique data that competitors can't easily copy or acquire. Map algorithms and the orchestration layer as a core company asset, not a support process.
  3. Cost Structure. Model the economics of training and running models (inference costs): cloud compute pricing, API costs at scale, and how these costs behave as user/transaction volume grows (linearly or sublinearly).
  4. Ecosystem & Partnerships. Decide what models/capabilities are built in-house (Build), what's used through ready-made interfaces (Utilize), and who to partner with (Partner). For a deeper look at this decision: ../build-vs-buy-vs-partner-ai/SKILL.md.
  5. Human-AI Interaction Model. Explicitly decide which interaction mode humans and AI use to work together in this business model — this is a separate choice from the level of human oversight (see below):
    • Copilot — AI assists human-driven work in real time, the human remains the primary actor.
    • Autonomous Agent — AI independently performs a bounded set of tasks, the human sets the goal and checks the result.
    • Generative Interface — the interface itself is generative and adaptive rather than fixed (see ../../../specialisation-packs/ai-native-startup-design/skills/ai-native-conversational-os-design/SKILL.md for a broader treatment of designing a conversational interface architecture, if this is the form chosen). This choice is a different question from the LEVEL of human oversight (in/on/outside-the-loop, see ../../../specialisation-packs/ai-native-startup-design/skills/closed-loop-process-and-human-oversight-design/SKILL.md) — the interaction mode describes HOW humans and AI work together, the oversight level describes HOW MUCH the human checks. Use both together: e.g. "Autonomous Agent + human-on-the-loop" is a valid combination.
  6. Data flywheel check — and the proprietary learning-engine principle behind it. Assess whether usage feeds data back into the model so the product improves on its own through use — this is one of the key differences between an AI-enhanced and an AI-native model. Go one level deeper than the check itself: the durable moat in an AI-native business is essentially never a single AI module on its own — individual modules commoditize quickly once unbundled, because a competitor can assemble an equivalent module from the same commodity building blocks. The moat is the CONTINUOUS, PROPRIETARY INTEGRATION of modules into one feedback loop spanning the full chain — from experimentation, through delivery, to the go-to-market signal that comes back from real usage — getting measurably stronger with every transaction. A well-documented reference case: Recursion Pharmaceuticals records millions of cellular images to train models on how diseases change cell morphology — a proprietary "wet-lab" feedback loop independently assessed as more durable than the algorithms it runs, precisely because the algorithms alone are increasingly commoditized while the integrated, accumulating dataset isn't. When evaluating a canvas's Key Resources and this flywheel check together, ask explicitly: if a competitor could buy or copy every individual AI module this business uses, would the business still have an advantage? If the honest answer is no, the "moat" is an illusion built on components, not on an integrated loop. Once a flywheel is DESIGNED this way, validate whether the claim actually holds using ../../../data-strategy-and-literacy/skills/data-monetization-model-selection/SKILL.md's four-point flywheel checklist (unique collection channel, measurable model improvement, observable UX improvement, the loop closing as growth) — that skill audits whether a flywheel claim is real; this step is about designing the architecture so it can be.
  7. Produce a structured output (a canvas table or equivalent, see ../../references/ once added) and validate it with stakeholders or your own experience-based checklist.

What this skill does NOT do

  • Doesn't make the final business-model decision for you — it produces a structured draft to support a human decision.
  • Doesn't confirm exact inference or cloud-cost figures from memory — it uses the inputs you provide, or marks an assumption clearly ([assumption — verify]).
  • Doesn't make the build-vs-buy-vs-partner decision finally — it structures it as part of the canvas but points to a deeper analysis in another skill.
  • Doesn't assess technical feasibility or PoC scoping — that's ai-use-case-feasibility-and-poc-scoping's job.

Refinement notes

Areas to keep deepening with real practice:

  • your own rules of thumb and heuristics for this technique
  • concrete templates (into ../../references/, e.g. a canvas template)
  • reference cases / your own examples of AI-native business models
  • what this skill deliberately does not do (guardrails, common mistakes) — add to the list above

Once this section is filled in and validated in practice, update the maturity field in skills_index.json to draft, validated, or canonical (see ../../../meta/maturity_levels.md). Don't add new fields to the frontmatter — name and description are the only ones allowed (see ../../../meta/frontmatter_schema.md).

Continue from here

  • Preceding skill in this pack: ../ai-opportunity-portfolio/SKILL.md — identifies and prioritizes AI use cases before business-model design.
  • Next in this pack: ../ai-use-case-feasibility-and-poc-scoping/SKILL.md — determines the technical boundary conditions of an AI use case and scopes the PoC phase.
  • A deeper-dive skill for the ecosystem decision: ../build-vs-buy-vs-partner-ai/SKILL.md
  • Related skill in another pack (once the model is validated and a full business case is needed): ../../../business-case-and-analysis/skills/business-case-builder/SKILL.md
  • A ready-made skill chain for this situation: see ../../../playbooks/
  • This pack's shared guardrails: ../../CLAUDE.md

References

  • ../../references/ai-native-reshuffle-heuristics-research.md — grounding notes for the proprietary learning-engine principle in step 6
  • ../../references/ — the pack's shared background material
  • ../../CLAUDE.md — the pack's shared guardrails