Updated April 2026 — expanded from three foundational questions to ten, reorganized around where organizations actually are in their AI journey.
Most AI initiatives fail not because the technology is immature. They fail because the foundational questions never got asked. These ten are worth asking now… whether you’re starting something new or managing what’s already running.
The conversation around AI in the enterprise has shifted. A year ago the question was whether to invest. Today the question is how to manage what’s already proliferating across the organization… some of it sanctioned, some of it not, most of it moving faster than the governance thinking.
These ten questions are organized around where organizations actually are in that journey. The first group is about what you’re building. The second is about what you’re already running. The third is about whether it’s sustainable at scale.
Before You Build
1. What specific business outcome does this serve?
AI is not a strategy. It’s a tool. The question isn’t “how can we use AI?” but “what measurable outcome are we trying to improve, and is AI the most efficient path to get there?”
Good answers look like: reduce manual document review time by 60%, increase lead qualification accuracy from 40% to 75%, cut customer support response time from 4 hours to 15 minutes.
Bad answers look like: “we need an AI strategy,” “our competitors are doing it,” “the board wants to see innovation.”
If you can’t tie the initiative to a number that matters to the business, stop and redefine the scope before spending engineering cycles. This is the question that separates AI initiatives that deliver value from AI initiatives that generate slide decks.
2. Is this an agent or is this automation?
This distinction matters more than most leaders realize, and most vendors are deliberately blurring it.
Traditional automation executes a defined sequence of steps. It does exactly what it’s told, in exactly the order it’s told to do it. An AI agent makes decisions. It interprets context, applies judgment, takes actions on behalf of humans, and can operate autonomously across multiple systems without a human present for each step.
The reason this question matters: agents create accountability exposure that traditional automation doesn’t. When an automated workflow misfires, you debug a script. When an agent makes a consequential decision without adequate oversight, you explain it to a regulator, a client, or a board. Before you build, be honest about which one you’re actually building… and whether your governance model is designed for it.
3. Do you have the data to support it?
Every AI initiative is a data initiative in disguise. Before evaluating models or platforms, audit what you actually have.
Do you have enough data to train or fine-tune, or is this a zero-shot problem better suited to a foundation model? Is your data labeled, clean, and representative, or are you building on a foundation of inconsistent exports from a legacy system? Can your team actually reach the data they need, or is it locked behind compliance restrictions, organizational silos, or a system nobody wants to open up?
The honest assessment here saves months. Teams that skip this step end up building elaborate pipelines for data that turns out to be unusable.
4. Who owns this after launch?
AI systems are not set and forget. Models drift. User behavior changes. Edge cases surface in production that never appeared in testing. Vendor pricing changes overnight.
Before you start, define who watches model performance post-launch, how end users report problems, how often you’ll update prompts or retrain, and who has budget authority when the cost model changes. The most successful AI initiatives treat launch as the midpoint, not the finish line. The teams that plan for ongoing ownership from day one are the ones that sustain results.
As You Deploy
5. Do you know what’s already running?
This is the question most organizations are least prepared to answer cleanly.
Before building anything new, do you have visibility into what AI systems and agents are already deployed across the organization? In most companies the answer is partial at best. Agents are embedded in SaaS renewals nobody fully reviewed. They’re running in workflows stood up by individual teams without IT involvement. They’re operating with credentials that were provisioned for a proof of concept and never tightened for production.
You can’t govern, defend, or claim value from AI systems you can’t account for. And when a board, a regulator, or an acquirer asks for a complete picture, “we’re still building the inventory” is not an acceptable answer.
6. Who’s accountable when something goes wrong?
Not who built it. Not who deployed it. Who answers for it.
This is the question that elevates AI from a technology discussion to a leadership one. AI agents don’t get deposed. They don’t sit in front of regulators. They don’t sign their name to anything. When an agent makes a decision that causes harm, the human in the loop is holding the bag… which means putting people in that position without clear accountability structures and adequate oversight isn’t responsible governance. It’s exposure.
For every significant AI deployment, there should be a named owner who understands what the system does, can explain how it makes decisions, and has the authority to shut it down if something goes wrong. In most organizations, that person doesn’t exist.
7. What’s your cost model?
Most organizations have no idea what their AI operations actually cost at the action level because flat-rate subscriptions have been hiding the real numbers.
That’s changing. Vendors are repricing, subscription terms are tightening, and the economics of unlimited agentic usage on flat-rate plans don’t work at scale. When the pricing changes, as it already has for some platforms, organizations that haven’t built a cost model discover the gap in the worst possible way.
Nobody has a budget line for token burn yet. That needs to change. Per-agent cost, cost-per-outcome, spend by workflow… these aren’t exotic metrics. They’re the basics of managing a new category of operational cost that doesn’t fit cleanly into existing budget structures.
8. Are your agents compliant with what’s coming?
The regulatory window is closing faster than most organizations realize.
EU AI Act high-risk provisions take full effect in August 2026. NIST’s AI Agent Standards Initiative published its framework in early 2026 and it’s moving from voluntary guidance to procurement requirement to litigation standard faster than most technology standards do. If your organization operates in financial services, healthcare, legal services, or any other regulated industry, the agents currently deployed are going to face compliance scrutiny whether you’re ready or not.
The question isn’t whether compliance will be required. It’s whether you’ll be building the documentation, audit trails, and governance infrastructure ahead of the mandate or scrambling to produce it after one.
As You Scale
9. What does your workforce actually look like now?
Not the org chart. The real workforce.
How many AI agents are operating alongside your human teams? What are they doing? Who authorized them? What systems do they have access to? Are the autonomy levels they’re running at the result of deliberate decisions or deployment defaults?
Most leadership teams can answer those questions for their human workforce. Almost none can answer them for their digital one. That gap is a governance problem, an operational risk, and eventually a competitive disadvantage. It compounds as the agent population grows.
10. Can you pass the diligence test?
This is the question that matters most to anyone thinking about a transaction, a partnership, or a significant investment.
When a buyer, a PE firm, or a strategic partner asks for a complete picture of AI systems in use… what’s running, what data it touches, what the liability exposure is, what the vendor dependencies are, who owns it… can you produce that cleanly and quickly?
Most companies can’t. The inventory doesn’t exist in a form that’s auditable. The accountability structures haven’t been documented. The vendor dependencies haven’t been mapped. The cost model hasn’t been built. And the governance framework that would make all of this legible to an outside party is still aspirational.
AI due diligence is becoming a standard part of transaction processes. The organizations that have built the governance infrastructure proactively will be able to answer these questions in hours. The ones that haven’t will spend weeks trying to reconstruct a picture that should have been built in real time.
The Bottom Line
These ten questions are not glamorous. They won’t generate a slide deck that impresses a boardroom. But they map to the failure modes that are actually ending AI initiatives right now… vague outcomes, invisible infrastructure, absent accountability, and governance that exists only in documents nobody reads.
The technology is capable. The constraint is almost always organizational. Start with the questions.