What is AI governance?

Why AI broke traditional IT governance

Classic IT governance assumes a project has a knowable spec, cost, and benefit. AI breaks all three. Value is uncertain, behavimy is probabilistic, and the cheapest way to learn whether something works is often to try it. Stage-gate processes built for ERP rollouts approve AI bets on slideware and discover the truth far too late.

The result is the pattern MIT’s NANDA initiative measured: 95% of enterprise GenAI pilots fail to move the P&L, and BCG found only 22% of companies have moved past proof-of-concept while just 4% create substantial value. The governance gap, not the technology, is what burns the money.

AI governance is really an ROI problem

Most AI governance conversations start with risk, ethics, and compliance. Those matter, but they are not where the money leaks. The money leaks when the organisation cannot say which AI bets are actually returning value and keeps funding the ones that never will.

So the first job of AI governance is ROI accountability: make every bet comparable on value, evidence, adoption, governance risk, and time-to-impact, then set explicit fund, fix, or kill thresholds so decisions do not drift on sunk cost. Risk controls sit on top of that, they do not replace it.

• Make every AI bet comparable on value, evidence, adoption, governance risk, and time-to-impact.
• Require behavioural evidence, not just a business case, before a bet earns significant capital.
• Set explicit fund, fix, or kill thresholds so decisions do not drift on sunk cost.
• Keep an auditable trail the board can defend to investors and regulators.

Governance by evidence, not by audit

This is where Leslie Barry is different from the auditors and consultants who dominate the AI governance conversation. They fly in, run interviews and workshops, and hand you a framework, a maturity score, and a PowerPoint. You are left with a document, not a decision, and no new evidence about whether any specific bet will pay off.

I run governance as an operating discipline. Using pretotyping, the method created at Google and taught at Stanford, I produce cheap behavioural evidence on your highest-stakes bets in days, then rank the whole portfolio on that evidence. Rapidly, my innovation-management software, keeps it as a living system of record, so governance stays current instead of aging in a slide deck.

How Leslie Barry operationalises it

My AI Bets Audit is governance you can run in two weeks. I map your live and proposed bets into one portfolio, score them on a common framework, pretotype the highest-stakes ones for real evidence, and hand the board a defensible fund, fix, or kill decision for each.

It draws on the same method I’ve run across 4,000+ enterprise experiments, work that has saved teams like Tabcorp, AGL, and RACQ from funding bets that were never going to pay off.