Everyone wants to talk about AI governance. ISO 42001. Model risk. Agentic workflows. Fine — but most of the organizations rushing to stand up AI governance programs haven’t done the boring work underneath. And AI doesn’t forgive shaky foundations. It exposes them, faster and at scale.
You can’t govern what models touch if you don’t know where your data lives, who has access to it, or which systems actually run the business. Automation magnifies whatever’s already there — including the gaps.
Start with how the money is made.
Before any framework, any tool, any control catalog — map your revenue engine. What products generate cash? Which systems support them? What data flows through, and who touches it? Which people are critical to keeping it running? That’s your crown jewels list. Everything else billows out from there: access, monitoring, vendor risk, AI use cases, privacy obligations.
Most programs skip this step and start with a control framework. Then they spend two years protecting things that don’t matter and missing things that do. If you want to allow your privacy, security or AI governance programs enable the business, you HAVE to understand it.
Use the standard guidance. It works.
CIS Controls v8 (the Top 18, formerly Top 20) covers the security fundamentals that prevent the majority of real-world incidents. ISO 27701 gives you a defensible privacy backbone. Neither is exciting. Both work. If you’ve genuinely implemented the first six CIS controls — inventory, software inventory, data protection, secure config, account management, access control — you’ve eliminated most of the attack surface that AI deployments will inherit and amplify.
Simply: you can apply practical automation and innovation with AI because you understand the full picture, what you have, how it works and the controls it must inherit.
Privacy is the same story. RoPA (data processing records), lawful basis, retention, DSR workflows, DPIAs for high-risk processing. Unsexy. Foundational. Required before you can credibly say anything about AI governance.
Don’t get caught up in tools.
You don’t need a GRC platform to start. You don’t need CSPM to know which S3 buckets are public. A well-maintained spreadsheet in Google Drive, a structured Notion workspace, a deep knowledge of your cloud infrastructure, or a disciplined OneDrive folder beats an expensive platform nobody updates. Tools amplify discipline. They don’t create it. Buy the platform when the process is real and the manual version is slowing you down — not before.
Then layer AI governance on top.
Once you know your crown jewels, your data flows, your access model, and your vendors — then AI governance becomes tractable and has sustainable velocity. Model registry maps to your data inventory. Pre-deployment risk assessment uses the same risk register. Vendor AI risk slots into existing TPRM. ISO 42001 stops being a parallel program and becomes an extension of the work you already did.
Skip the fundamentals and AI governance becomes theater. Documented policies nobody can operationalize. Model registries with no underlying data lineage. Human oversight requirements that fail because access controls were never enforced in the first place.
Fundamentals win championships. Same in security. Same in privacy. Same — especially — in AI.