THE PROBLEM WITH THE MARKET
AI governance advisors do not compete on methodology. They compete on relationship. Most institutions engage the firm they already use for audit, strategy, or technology — and accept whatever framing that firm brings.
Framing is not neutral.
A consulting firm aligned with a hyperscaler cannot credibly tell the institution to build its own infrastructure. A firm whose internal operations run on a partner model vendor cannot credibly recommend a different vendor. A software company that sells AI faces an inherent conflict when it also offers governance over that AI.
The conflict is not ethical. It is structural.
THE FIVE ARCHETYPES
01 · THE FRAMEWORK CONSULTANCY
What they sell. Board roadmaps, ethics principles, responsible-AI taxonomies. Deliverables are policies, committees, and guiding questions.
Structural conflict. A framework is not an architecture. It defines what the institution should do — not what the infrastructure enforces.
Failure mode. Governance philosophy without technical reality.
02 · THE HYPERSCALER-ALIGNED STRATEGIST
What they sell. Enterprise strategy paired with delivery capability built on a single cloud platform. Joint teams, co-funded assessments, shared commercial models.
Structural conflict. The firm's economics are tied to one infrastructure outcome. The strategic recommendation and the delivery revenue cannot be separated.
Failure mode. A compromised answer to Build, Rent, or Compose.
03 · THE MODEL-VENDOR-FUSED ADVISOR
What they sell. AI transformation programs built on a partner foundation model. Internal operations and client deliveries run on the same stack.
Structural conflict. The firm's own productivity depends on the model it recommends. The governance layer assumes the vendor it cannot replace.
Failure mode. The recommendation is the product.
04 · THE AUDIT-ANCHORED ADVISORY
What they sell. Inventory, risk-tiering, testing, monitoring, accountability. Traditional risk-function methodology extended to AI.
Structural conflict. Knowing which AI systems exist, and testing them, is not the same as controlling the infrastructure underneath them.
Failure mode. Audit dressed as control.
05 · THE GOVERNANCE SOFTWARE VENDOR
What they sell. A product that monitors, logs, and reports on AI use — typically operating within the vendor's own AI and data platform.
Structural conflict. The entity offering governance is the same entity providing the models, data platform, or infrastructure.
Failure mode. A vendor cannot govern itself.
WHAT THIS MEANS FOR THE INSTITUTION
The archetypes are not wrong. They are useful — for what they are.
A framework consultancy can help a board structure its AI committee. A hyperscaler-aligned strategist can help scale workloads on that hyperscaler. A model-vendor-fused advisor can accelerate deployment on that vendor's stack. An audit-anchored advisory can inventory and test a portfolio of existing use cases. A governance software vendor can monitor the AI running inside its own platform.
Each archetype has a structural economic interest in the outcome of the Build / Rent / Compose question — which is why an unconflicted answer requires a firm without that exposure.
Institutional AI is engaged when the institution needs the unconflicted answer. After that, the archetypes may still have a role — executing the direction the institution has committed to, under a control architecture the institution now owns.
HOW TO TEST THE CATEGORY
If there is uncertainty about which archetype a current advisor fits, the test is simple. Ask the advisor four questions:
- Does your recommendation depend on a specific cloud platform, model vendor, or software product?
- What percentage of your firm's annual revenue comes from a single infrastructure or model partner?
- If we concluded that Build was the right answer for our most sensitive workloads, how would your engagement economics change?
- Can you produce a control architecture that survives your departure?
The answers separate advisors from architects.