Most board AI conversations focus on productivity, cost reduction, and competitive positioning. These are the right questions for the CFO. They are the wrong questions for the board. The board's role is governance — and the governance question nobody is asking is who owns the reasoning layer your AI is learning to replicate.
01 —
If our primary AI vendor exited tomorrow, how long would it take us to restore full operational capability?
This is the AI equivalent of a business continuity question. Every board asks it about critical infrastructure. Almost no board asks it about AI vendors.
The answer reveals your actual dependency. If the honest answer is months or years, the board has approved an architectural dependency that has not been risk-assessed.
► Ask management: what is our AI vendor exit strategy? The answer tells you more about AI architecture risk than any security audit.
02 —
Are we training our vendors with our institutional intelligence — and do we have a policy about it?
Every time your team approves an exception, resolves a dispute, or overrides a system recommendation through a vendor AI platform, that judgment is absorbed by the vendor's model. Not yours.
Your finance team's pattern for identifying a strategic customer worth protecting. Your operations team's instinct for when a supplier exception is worth making. This is Business Intelligence Compromise — and most enterprises are transferring it to vendors for free, at scale, without a line item on any risk register.
► Ask management: who owns the intelligence layer being built from our data — and can we take it with us if we leave?
03 —
Do we own the three layers of our AI architecture — or have we handed two of them to vendors?
Enterprise AI architecture has three distinct layers. Most boards are only governing one of them: the Transaction Layer (ERP, CRM — vendor-owned, that is fine), the AI Layer (models, copilots — must be disposable by design), and the Collective Reasoning Layer (intent schemas, governance rules, tribal knowledge — you must own this).
No vendor has economic incentive to build the layer that makes you independent of them. That layer must be yours.
► Ask management: can we swap our AI model without rebuilding our architecture? If no, we do not own the AI layer. We are renting it.
04 —
What decisions are we permitting AI to make autonomously — and who defined those boundaries?
Autonomous has meant the same thing since the industrial revolution: human judgment redistributed, not eliminated. A collections AI that does not know a 15-year customer just placed a $3M order will optimize you out of that relationship in one automated message. That is not an AI failure. That is a governance failure.
The governance layer does not have to be complete before deployment. It has to be intentional. Answer three questions before any agent acts: what does a defect look like, who has permission to stop the line, and what does fix-at-source mean here.
► Ask management: are escalation thresholds in our AI deployments set by the vendor or by us?
05 —
What is our Integration Tax — and has the CFO ever calculated it?
The Integration Tax is the hidden cost of maintaining connections between enterprise systems — customizations, integrations, and workarounds that accumulate over years of vendor lock-in. Industry research puts this at 30 to 60 percent of enterprise IT budgets. Most CFOs have never calculated their number. When they do, they are shocked.
AI does not reduce the Integration Tax. In most deployments it compounds it — through Double-Dip pricing where vendors charge legacy seat licenses plus AI add-on fees plus API call charges simultaneously.
► Ask management: what percentage of our IT budget goes to maintaining existing integrations versus creating new value?
06 —
Are we building an Agentic Babel Tax that will cost us more than the agents save?
When multiple AI agents — from different vendors, built on different models, with different context windows and memory architectures — are deployed across enterprise workflows and cannot natively coordinate, the enterprise builds translation layers between them. This is the Agentic Babel Tax. The Integration Tax of the agentic era.
Every enterprise that deployed SaaS point solutions in the 2000s without an integration strategy spent the next decade paying the Integration Tax to connect them. Every enterprise deploying AI agents today without a coordination strategy is building the same problem one layer up.
► Ask management: do we have a standard for how our AI agents communicate with each other — or are we deploying agents the same way we deployed SaaS tools in the 2000s?