Stop Buying AI: Start Buying Decisions

I’ve sat through a lot of technology investment cycles over the last twenty years, and every one of them has followed roughly the same shape. ERP came in promising to transform the business, then CRM to transform customer interactions, then Business Intelligence, Big Data, the Data Lake, Cloud, Digital Transformation, the list is endless. Each wave of “transformation” got an even bigger budget than the previous one, as well as a steering committee and a slide deck about the future of the organization. Each one left behind a bit more architecture, a bit more governance, a bit more complexity for whatever came next to work around.

Guess what? AI is just the latest arrival in that queue, and I’m watching most boards wave it through using the same process they used for everything before it.

Nobody questions that AI matters. Where I think boards are falling short is much simpler than that, and it’s a question I’ve started asking directly in the room: is this investment going to change how we make decisions, or is it just more technology sitting on top of the technology we already have?

That’s not a rhetorical distinction. No business has ever won by owning software, because everyone has access to roughly the same models and the same vendors pitching the same roadmap. What separates one competitor from another is the quality of the decisions being made: what to price, where to put capital, which risk to take seriously before it becomes a problem, which customer to understand better than the business next door does. AI is only worth what it does to those decisions. If it doesn’t touch them, it’s just a cost.

Most of the AI strategies that land on my desk don’t start with decisions. They start with the technology, a long list of possible use cases, a platform someone’s already bought, an agent someone in IT is excited about, and the business case gets built backward from there. I’ve seen this play out before, more than once, with a client where the ERP system had quietly become the way people described the business itself. Nobody could tell me how the company made money without pointing at a screen. That’s not a technology problem. It’s what happens when an organization stops being able to explain itself in its own words, and it existed long before AI showed up. AI just makes it more expensive and faster to get wrong.

An agent can’t fix a business that doesn’t understand its own decisions. It just does more of whatever the business was already doing, faster.

Why the Canvas Starts where it starts

Every function has a reasonable but partial view of where an AI conversation should begin. The technology team wants to talk platforms, the data team wants to talk governance and quality, the Innovation team wants a use case list, and the Finance team wants the investment case. All of that is fair, and none of it is what a board needs, because a board isn’t approving a piece of software; it’s deciding where to put capital it can’t get back once it’s spent.

That’s the reason the Data & AI Strategy Canvas doesn’t open with technology. It opens with what the business is trying to achieve, works through the decisions and the value that follows from that, and only lets technology into the room once all of that is settled. In practice, most organizations I work with have that order backward, and putting it right is usually the first real piece of work.

My Data & AI Strategy Canvas – Samir Sharma

The ten questions I want answered before AI gets discussed

1. Strategic objectives: What is the business attempting to achieve? Get this wrong, or skip it, and everything downstream turns into tactics dressed up as strategy.

2. Key decisions and use cases: Which decisions move commercial performance? I’m not interested in a list of a hundred possible AI applications. I want the handful that genuinely matter.

3. AI and analytics opportunities: Can AI improve those decisions? Sometimes yes. Sometimes ordinary analytics does the job just as well. Sometimes the honest answer is that no technology is needed at all, and a board needs to be able to hear that without it feeling like a failure.

4. Value measures: How is value being tracked? I see too many programs reporting on how many models got built, or agents got deployed. A board isn’t a funding activity. It’s funding an outcome.

5. Data requirements and products: What data does this genuinely require, and what are we building with it: a model, an API, something a person or another system can use?

6. Consumers and users: Who’s on the receiving end of this? Increasingly, it’s other agents as much as it’s people, and that changes how you must design the thing.

7. Technology and systems. This is where technology finally enters the conversation, not where it starts. The platform exists to support the capability, not the other way round.

8. Portfolio management: How does this sit against everything else competing for the same money? Most organizations running twenty disconnected AI pilots skipped this question entirely, which is usually why they are disconnected.

9. Operating model and governance: This is the one boards overlook the most. Who owns the agent once it’s live? Who’s watching it? Who steps in when it does something nobody expected? Redesign the technology and leave the management structure exactly as it was, and you end up with capability nobody is accountable for.

10. Risk and ethics: Bias, regulation, IP, security, reputational exposure. This has to sit inside the investment decision from the start, not get bolted on afterward because compliance asked.

What I Would Add Today

If I were extending the Canvas right now, I wouldn’t add another box to it. I would add a habit of asking a different kind of question inside all ten. Why? Because three people have opened my worldview to Causal AI: John Thompson, Bill Schmarzo, and Mark Stouse. Each has taught me that extending and asking deeper questions is what gives most organizations the edge that they need.

For example, most analytics tells you what has already happened. A predictive model has a go at what might happen next. Neither one gets close to why it happened, or what would happen if you changed something. That is the causal question, and it’s a genuinely different way of thinking about the problem, not just a fancier dashboard, and that is what makes the difference between just doing it because everyone else is doing it and really drilling into the hard questions. I really thank my colleagues John, Bill and Mark for this.

Once a board starts asking what is causing an outcome, and which of those levers it can realistically pull, the whole conversation shifts. You stop reacting to last quarter’s numbers and start managing the conditions that produce the next.

Every board should therefore be asking:

• What genuinely causes this outcome?

• Which variables can management influence?

• What assumptions sit underneath our understanding of it?

• What evidence would prove those assumptions wrong?

• What unintended consequences might an intervention create?

• Which leading indicators tell us the change is already underway?

• At what point do we intervene rather than keep watching?

Most boards never get past the first one. That’s usually where the real problem lies.

What I Would Leave the Board With

None of this is about producing another strategy document, and it’s not about producing a longer list of AI projects. It is about making sure the decision gets better before the money leaves the building.

Boards have always had to allocate capital without perfect information; AI hasn’t changed that job, it’s just made the mistakes faster and more expensive. As we head into the next few years, no doubt all organizations will have agents running around, but that kind of capability is going to be a commodity soon enough, available to anyone with a card to swipe. The edge is going to sit with the businesses that understand how they make money, and have the discipline to back that understanding with real capital decisions rather than enthusiasm.

That’s what I built the Canvas for. Not to get a business to adopt more AI, to stop a board from finding out the expensive way that it should have asked a different question first.

About the Author

Samir Sharma is a senior data, analytics, and enterprise applications leader with over 20 years of experience and specializes in helping boards and executive leadership teams operationalize data and AI strategy across business processes, decision-making, and service delivery. He is the author of The Strategy Canvas: A Field Guide for Data & AI — Closing the Strategy-Execution Gap.