What Boards Will Ask About AI and How to Answer It

Board conversations about AI have shifted from opportunity to accountability, and the questions are operational. This piece covers what is asked and what answering requires, drawing on FISTA Solutions' AI enablement governance work.

What are the questions?

Six that recur, with what each requires.

Question	What answering requires
What AI is running here?	A maintained inventory
What does it decide?	Documented scope per system
What does it cost to operate?	Full cost attribution
How do you know it works?	Evaluation evidence over time
What is the worst case?	Risk assessment per system
Who is accountable?	A name, with authority

Why did the questions change?

Because AI moved from experiment to dependency.

While systems were pilots, the board conversation was about strategy and competitive position. Once customers are affected by automated decisions, it becomes a risk and control conversation, which is the board's core function.

Regulatory attention accelerated this. Directors reading about enforcement in other organisations ask whether their own house is in order, and they ask specifically. See the coming audit of AI systems.

Why is the inventory the hardest question?

Because most organisations genuinely do not know.

AI appears in team-built tools, in vendor products as features, in spreadsheets, and in processes nobody registered centrally. The list the technology function maintains is usually a subset.

Building it requires asking every function what they use, including purchased software with AI capability. It is tedious and it is the foundation for every other answer. See AI service catalog template.

What does the cost question uncover?

That the full cost is not known.

Model spend is usually available. Review hours, evaluation maintenance, corpus upkeep, and monitoring are usually not, which means the reported cost is a fraction of the real one.

Directors asking about return will get a number built on incomplete cost, and that becomes a problem later. Building full cost attribution before the question arrives is the sensible order. See the operating cost of intelligence.

What quality evidence satisfies a board?

Trends over time, incidents with resolutions, and evidence that regressions were caught.

A dated evaluation report showing quality stable or improving, a log of incidents with what changed afterwards, and a description of what is monitored. That demonstrates a system under management.

What does not satisfy is confidence. Directors have learned that confident assurance about systems nobody measures is the pattern that precedes incidents. See AI eval report template.

Why does accountability need a name?

Because governance without an accountable individual is not governance.

When a committee owns a system, nobody owns it. Boards have learned to ask who specifically is responsible, and the absence of an answer is itself the finding.

That person needs real authority — to change the system, to pause it, to refuse a deployment. Accountability without authority is an unfair position and an ineffective control.

How should this be reported?

Regularly, briefly, and with the same numbers each time.

A short standing report — inventory count and changes, cost, quality trend, incidents, and open risks — read quarterly builds the understanding that makes a crisis conversation possible.

The alternative is an ad hoc deep dive triggered by an incident, which is a worse conversation from a worse starting position. See AI board reporting checklist.

What is the counter-argument?

The counter is that this level of scrutiny slows adoption and treats AI more harshly than other technology. There is something in that. The response is that the scrutiny follows consequence: systems making decisions about customers attract governance, and that is proportionate rather than punitive.

What does this change for engineering teams?

It means the artefacts governance needs — inventory, cost attribution, evaluation results, incident records — should be produced automatically rather than assembled for meetings.

Assembled-for-meetings reporting is expensive, late, and usually wrong. Generated reporting is cheap and correct.

What does this change for buyers?

It means asking vendors for the evidence you will need to report upward: what their system decides, what it costs, how quality is monitored, and what audit trail you receive.

A vendor who cannot supply that has left you unable to answer your own board.

What should leaders do about it now?

Build the inventory now, including vendor products with AI features. It is the question you will be asked first and the one you are least able to answer quickly.

Then name an accountable person per significant system and give them authority to pause it.

What about agents specifically?

They raise the stakes on every question, because they act rather than advise. Directors will ask what an agent can do without human approval and what the limits are.

An organisation that can produce a written scope, limits, and approval thresholds per agent answers well. One that cannot has a governance gap that is visible immediately. See the shift from chatbots to agents.

How will you know if this is happening?

Watch for AI appearing on the audit committee agenda, for questions about specific automated decisions, and for requests for an inventory. Each means the accountability phase has started.

How FISTA Solutions reads this

FISTA Solutions builds and operates production AI systems through AI agents, AI enablement, and forward deployed engineering: a maintained inventory covering vendor AI features, named accountability per system, and governance artefacts generated automatically rather than assembled for meetings, decisions documented with their reasoning, and handover that leaves your team able to maintain what was delivered. The record is 150+ projects for 50+ companies across 12+ countries.

To discuss what this means for your roadmap, message FISTA on WhatsApp, or read AI governance framework.