Business Model Portfolio

A Business Model Portfolio (BMP) is the set of distinct, often interdependent, business models a single firm operates simultaneously. Source: westerveld-business-model-portfolio-2023 (Westerveld et al., 2023 — developmental literature review over 38 empirical cases, 2010–2021).

The core move is to treat the business model, not the firm, as the unit of strategic analysis. A startup may be one company on the cap table while running two or three business models, each with its own value proposition, value architecture, functional architecture, and revenue logic. Operating multiple BMs has shifted from a defensive contingency (diversification against shocks) to a proactive strategic tool for value creation — and a permanent source of organisational paradox.

Why this page exists in this wiki

The weber-ai-startup-business-models classifier and the ai-startup-business-archetypes-weber morphological box silently assume one startup = one business model. In practice, especially in 2026, AI startups routinely run two or more BMs (e.g. a self-serve B2C product and an enterprise consulting engagement and a developer API). Without the BMP frame, a single startup is forced into one Weber pattern that distorts the picture. With the BMP frame, each BM is classified separately, and Weber’s four-pattern taxonomy becomes a per-BM, not a per-firm, instrument. This page is the prerequisite framing for weber-taxonomy-2026-gaps and the planned GTM sibling taxonomy.

Three strategic intents (chronological evolution)

Westerveld et al. identify a chronological shift in why firms deploy a BMP:

1. Diversifying — historically dominant. Different BMs target distinct market segments autonomously (e.g. a legacy airline operating a separate low-cost carrier). Logic: hedge.
2. Sensing — BMs as experimental probes into uncertain markets. Failed probes are discarded without disrupting the core. Logic: option value. Maps to “thousand flowers bloom” phase in wef-ai-in-action-2025 and to the early experimentation tier of enterprise-adoption-ladder.
3. Complementing — the most advanced contemporary intent. Tightly integrated BMs where one explicitly reinforces another via shared data, infrastructure, or audience (Amazon Prime ↔ retail ↔ streaming; LAN Airlines passenger ↔ cargo capacity sharing). Logic: ecosystem synergies. Aligns with Huang’s “extreme co-design” in karpathy-software-3 and with the “enterprise-level reinvention” tier of enterprise-adoption-ladder.

Digital technology and data are the connective tissue that makes Complementing feasible at scale. Customer data captured in one BM can power the value proposition of another; shared functional architecture lets supply-side synergies cross BM boundaries.

Four paradoxes (permanent, not solvable)

Operating a BMP is “an exercise in balancing dualities”. Westerveld et al. identify four persistent tensions that cannot be eliminated, only managed:

• Horizon paradox (Exploit vs Explore) — the most prevalent. Exploit profitable capabilities while funding unproven ones. Theoretical anchor: ambidexterity / March 1991. In AI startups: maintaining a paying enterprise BM while staffing an experimental agent product.
• Cannibalization paradox — the new (often digital, often agentic) BM threatens to capture revenue from the incumbent BM. In 2026 AI: Steinberger’s prediction in karpathy-software-3 that agent-facing APIs will cannibalise GUI-led SaaS is the canonical instance.
• Digital paradox — tension between physical and digital assets, and between omnichannel complexity and the customer’s desire for seamlessness.
• Dogmatic paradox — organisational confirmation bias toward the incumbent BM. Echoed in AI research by Sutskever’s “human reward hacking” in eval-real-world-gap: researchers (and managers) over-optimise current recipes rather than seeking contradictory evidence.

Implication for the project

1. Unit of analysis. When ingesting an AI-startup source, identify each BM the startup operates before applying ai-startup-business-archetypes-weber. Classify per-BM. Record in the source frontmatter as weber-pattern: <pattern> per BM if more than one.
2. Strategic separation. For an operator (the author’s audience): keep each BM’s value proposition, distribution architecture, and pricing logic deliberately separate even when run by the same team. The logics are diverse; conflating them is the most common source of cannibalisation pain.
3. Strand B link. The forthcoming GTM sibling taxonomy (weber-taxonomy-2026-gaps Next move) will inherit BMP framing: each BM gets its own commercialisation motion, channel geometry, and pricing counterparty. A startup with two BMs has two GTM motions to track.
4. IT / digital strategy as enterprise matrix. Westerveld’s strongest practical implication: in a BMP, functional architecture (data, infrastructure, AI) cannot be aligned to a single BM. It is the connective tissue. The digital-core argument in digital-core is the buyer-side mirror of this point.

What this page is not

• Not a replacement for Weber. It is the prerequisite scoping rule that makes Weber applicable to multi-BM firms.
• Not a planning template. The three intents and four paradoxes are diagnostic vocabulary, not a checklist.
• Not validated on AI startups specifically. Westerveld’s 38 cases are airlines, retail, media — sectors with long-running BMP histories. The 2026 AI-startup application is an extrapolation this wiki is making.

See also

• westerveld-business-model-portfolio-2023 — the source.
• ai-startup-business-archetypes-weber — Weber’s classifier; the BMP frame is the unit-of-analysis prerequisite.
• weber-ai-startup-business-models — the source taxonomy.
• weber-taxonomy-2026-gaps — the diagnostic this page sits in front of.
• H3_orthogonal-axes-under-priced — a startup running multiple BMs occupies multiple positions on every analytical axis; the BMP frame is the structural reason orthogonal axes matter.
• H2_u-curve-of-value — a single startup can have one BM at the top of the U and another at the shoulder.
• karpathy-software-3 — Huang’s “extreme co-design” as the Complementing intent in computing; Steinberger’s GUI-vs-agent thesis as the Cannibalization paradox in 2026.
• eval-real-world-gap — Sutskever’s “human reward hacking” as the Dogmatic paradox in AI research.
• digital-core — buyer-side mirror of the “functional architecture as connective tissue” claim.
• enterprise-adoption-ladder — the WEF ladder maps loosely onto the Sensing → Complementing intent progression.
• wef-ai-in-action-2025 — “thousand flowers bloom” → “enterprise-level reinvention” as the WEF reading of the same Sensing → Complementing shift.