Ideas & Frameworks

AI readiness is a clarity problem

Not a data problem, and not a technology problem. Organisations that are clear on what their data means, who decides, and how much risk they'll carry are ready sooner than they think.

AI-ready data: you're more ready than you think (Altis)

Reversibility sets the quality bar

Start from the decision the data or AI supports. The harder that decision is to undo, the more rigour it deserves. Easily reversed decisions can move fast.

Humans in control, not just in the loop

A person clicking "approve" isn't control. Control means people set the boundaries, own the escalation paths, and can step in before harm, not just after.

The Meaning Gap

When meaning isn't defined, AI fills the gap with confident guesses. Ask about "active customers" and the model will pick an interpretation and answer fluently, right or wrong. Closing the gap means agreeing on and recording what things mean.

Trust by design: why AI fails when meaning is unclear (Altis)

Semantic Layer as AI Foundation

The semantic layer is broader than metric definitions and joins. For AI, we're stepping into meaning and ontology: what things are, how they relate, and what they mean to the business.

• Two meanings of "semantic layer"
• AI needs context, not just data (Altis)

Derived Meaning Store

Use an LLM to derive meaning from unstructured data once, then store it as governed, versioned data you can reuse everywhere. Stable and explainable, instead of asking the model again each time.

Read the introduction. A follow-up is coming.

AI Governance Matrices

Three simple matrices for deciding how much control an AI use case needs:

• Autonomy × Grounding
• Autonomy × Reversibility
• Verifiability × Reversibility

Place a use case on each and the right level of oversight becomes a conversation about evidence, not opinion.

Shadow AI Is a Governance Signal

Shadow AI isn't a compliance failure. It shows you where people need help and where your current approach isn't meeting them. Treat it as information, not just a breach.

MVG → MVO → AI Readiness

A maturity ladder for getting AI-ready without waiting for perfect foundations:

• Minimum Viable Governance (MVG): the smallest set of ownership, definitions and decision rights that lets people use data and AI safely.
• Minimum Viable Ontology (MVO): the core business concepts and how they relate, written down so people and AI share the same meaning.
• AI readiness: with both in place, AI works from governed data and agreed meaning.

Each rung is small enough to start now and builds on the one before.