What Questions Should a Board Ask About Their AI Strategy?

Your CEO comes to the board meeting with a new AI strategy. It’s exciting. It sounds like every other AI strategy: integrate LLMs, reduce costs, move faster, unlock new products. But nobody’s actually asking the questions that determine whether this works or becomes a nine-figure waste of time and compute budget.

As a board member, here are the questions you need answered.

“How Does AI Actually Make Us Money?”

This is the first question, and it’s almost never answered with actual numbers. Most companies have a vague notion that AI will help them reduce costs or build new features. But you need specificity.

Are you using AI to:

• Automate internal processes (finance, HR, support)? What’s the headcount savings? How certain are you? Is it 2 people or 20 people?
• Improve your product? What’s the customer willingness to pay? Is this a reason they pick you over a competitor, or just table stakes?
• Create entirely new revenue streams? What’s the addressable market? What’s your competitive moat?

If your CEO can’t answer this in terms of revenue impact within 12 months, press harder. “It’ll eventually be valuable” is not a strategy. “We’ll save $2–4M in support costs in 18 months if we automate our tier-1 support” is a strategy.

The honest answer is sometimes: “We don’t know yet, and we need to experiment.” That’s fine. But then ask: How much are we spending to find out? How long are we willing to spend it? What are the decision criteria for doubling down or cutting bait?

“What’s Our Competitive Advantage Here, Really?”

AI is becoming a commodity. Claude, GPT-4, open source models. They’re all getting better and cheaper. If your strategy is “we’ll use AI to do what everyone else is doing,” your advantage is that you moved faster this quarter. Next quarter someone else catches up.

So ask: What can we do with AI that’s harder to replicate? Is it our data? Is it the specific domain expertise we’ve built? Is it the integration with our product that competitors don’t have?

If the answer is “we’re first to market,” that’s real but temporary. If the answer is “we have unique data that trains a better model,” that’s real and defensible. If the answer is “we’ll build this, and it’ll be hard to replicate,” ask why. Is the technology hard, or is the competitive position hard?

The companies that win with AI aren’t the ones who adopt LLMs first. They’re the ones who have a defensible reason why their use of AI is better than everyone else’s.

“What’s the Cost Structure of This Strategy?”

Here’s where boards usually get surprised. AI is not free. Running GPT-4 API calls at scale is expensive. Training models is expensive. The infrastructure to support inference at your scale is expensive.

Ask your CTO: What’s our cost per transaction if we’re using an LLM for every customer request? What does that look like if we 10x our user base? If we’re training a fine-tuned model, what’s the infrastructure cost to run it? How does that compare to using the public API?

You should see a cost model. “We’ll integrate Claude into our support tool” is not a cost model. “Each support interaction will cost us $0.05 in API calls, we process 10K interactions per month, so $500/month in API costs, which is 10% of our current support spend” is a cost model.

If you can’t get this answer, that’s a red flag. It means they haven’t actually thought about whether this is financially viable.

“How Do We Maintain Quality?”

AI can do a lot of things, but “always correct” is not one of them. Every application of AI needs a quality gate. Sometimes that’s human review. Sometimes that’s automated testing. Sometimes that’s a small percentage of errors being acceptable because the cost of an error is low.

Ask: What’s our error tolerance? If we’re using AI to automatically categorize support tickets, and it gets it wrong 3% of the time, is that acceptable? What does the recovery process look like?

If you’re using AI to generate financial advice or medical recommendations, error tolerance is near zero. If you’re using AI to suggest content, error tolerance is higher.

The best AI strategies have a clear quality bar and a monitoring system that tells you when you’re below it. If your CEO says “we’ll ship it and see,” ask: How quickly will we know if it’s degrading? What’s the rollback plan?

“Do We Own the Relationship With the Customer, or Does the AI Model?”

This is subtle, and it matters for long-term positioning. If you’re using an LLM to interact with your customers, who are customers talking to? They’re talking to Claude or GPT-4. If you switch to a different model (or if OpenAI doubles their prices), what happens to your product?

This is why some companies are investing in fine-tuned models or open source model fine-tuning. You own the intelligence. You control the cost. You’re not dependent on another company’s API decisions.

Ask: Are we building on commodity models (Claude API, GPT-4 API)? If so, what’s our lock-in? What happens if our costs increase 10x? If we’re fine-tuning models, how much data do we need? How often do we retrain? Who owns the trained model?

The answer determines your long-term economics and your optionality.

“What’s Our Data and Privacy Story?”

Every AI strategy relies on data. Either training data or prompt data or both. You need to understand:

• What data are we using? Is it ours, or are we relying on public data?
• Are we sending customer data to external APIs? Have we encrypted it, isolated it, and made sure it’s compliant with our data governance?
• If we’re training models on customer data, have we anonymized it? Have we gotten explicit consent?
• What happens to the model weights if we train on customer data? Who can access them?

Some companies have strict rules: we never send customer data to external APIs. They fine-tune open source models locally. That costs more, but it gives them privacy control.

Other companies send anonymized data to third-party model providers. That’s cheaper, but it requires rigorous anonymization and audit trails.

You need to know which path you’re on, and you need to understand the implications.

“How Fast Can We Move, and How Stable Is This?”

AI models improve fast, but they also change fast. A model that’s state-of-the-art in March might be surpassed by June. API pricing changes. Model behavior changes with updates.

Ask: How dependent are we on current model capabilities? If Claude 4 comes out in six months and changes the game, can we upgrade our implementation in three weeks? Or are we locked into Claude 3.5 because we built something specific to its behavior?

The best AI strategies have built-in flexibility. You’re using models as a service (API), not as your core product. If that becomes impossible, you can migrate.

“What’s Our Hiring and Hiring Cost Story?”

If your AI strategy requires specialized AI engineers, that’s expensive and competitive. Startups are paying $250K–$400K+ for senior AI engineers. If your strategy requires five of them, that’s a $2M+ cost center.

Ask: Do we need specialized ML engineers, or can our existing engineers integrate AI into their workflows? If we need specialists, can we hire them? Can we afford them if they’re competitive?

The companies that win with AI often don’t have massive ML teams. They have product engineers who know how to integrate AI into user experiences effectively. That’s a different hiring problem and a cheaper hiring problem.

“What’s the Competitive Threat If We Don’t Do This?”

This is the flip side. Sometimes the board pressure is “everyone’s doing AI, we need to do it too.” That’s not a strategy. But it’s also true that if your competitors use AI to move significantly faster or at lower cost, you’re at a disadvantage.

Ask: What happens if we do nothing for 12 months? Do competitors eat our lunch? Or do we have time to learn, experiment, and build intentionally?

If the threat is real and immediate, your timeline is aggressive and you need to move fast (and accept some technical debt). If the threat is longer-term, you can afford to move deliberately.

The Real Question Behind All of These

The board’s job is to make sure your AI strategy is a business strategy backed by technology, not a technology strategy hoping to become a business.

Get the answer to “how does this make us money” first. Everything else flows from there.

If you can’t get a satisfying answer to that question in your next board meeting, that’s the conversation you need to have with your CEO. Not “more AI,” but “AI that works.”