Eager or Hesitant, Defining the Right AI Use Case Drives Proven Results

Why most AI efforts fail before they start

When leaders sit down to discuss AI with me, their organizations usually fall into one of two camps. Hesitant organizations are bogged down by how to evaluate and select the right tool out of the overwhelming volume of available options. Eager organizations have a board- or executive-level directive to “just use AI,” and are frantically implementing without a clear sense of what problem they are trying to solve. 

Both paths are equally risky. The hesitant path leads to missed opportunities, while the eager path leads to expensive, ungrounded experiments.  

Both camps are asking the wrong question. One is asking “Which tool do we use?” and becomes paralyzed trying to answer; the other is asking “What can we do with this tool?” and recklessly forges ahead. Neither camp is asking the question that determines whether AI will work for them. Instead of looking at tools as your starting point, first think about what problem you’d like AI to solve. 

The fastest way to make AI work for your organization is proving, with your own data, whether it can perform accurately within a specific use case. 

Your answerable question

The antidote to both frantic activity and inertia is to identify an AI use case. A use case is simply an answerable question tied to a measurable business outcome. 

That question defines what you’re trying to learn. It doesn’t tell you whether the answer is trustworthy. 

If you’re overly eager, this forced focus slows you down just enough to ensure you’re not wasting resources. If you’re hesitant, it gives you a specific criterion to evaluate against. 

• Where does our supply chain spend diverge from our competitors? 
• Who is really buying from us? 
• Which portfolio companies are leaking profit and where? 

The question you want to answer reveals exactly what data matters and keeps you from overengineering everything else. You don’t need enterprise-wide data perfection. You need data that’s reliable and accurate for that specific use case.  

The truth about data quality

There’s a dangerous misconception that you can apply AI on top of dirty data to get directional information. 

Let me be clear: If your data is wrong, your AI is wrong.  

AI is designed to produce answers. In fact, unless you specifically direct it otherwise, it can’t tell you it doesn’t know the answer. AI behaves like a straight-A student in their first ambiguous, high-stakes job: it will give you a definitive, highly polished response every time, even if it has to hallucinate the entire thing to do so.  

There’s no such thing as a directional trend from bad data; there is only a definitive answer that leads you in the wrong direction. 

This is especially relevant for eager organizations. When you move fast to implement a tool without vetting your data against a specific question first, you risk building decisions on answers that were never grounded in reality. 

Once a use case is defined, the next step is to test whether your data can answer it reliably. That’s where grounding comes in.  

Grounding is a process that validates the AI output by asking it questions you already know the answer to. If it gets the answers right, we know the approach is working. If it gets them wrong, we refine the model. This ensures the result is a validated, truth-grounded proof you can trust. 

Testing AI use cases as a strategic diagnostic

Testing a specific use case with AI serves as the ultimate stress test for your data and your strategy. The use case defines the test. The grounding process determines whether the result holds up. 

For the hesitant, it provides a concrete reason to move forward without having to solve every future scenario first, and it reveals if the data you have can answer the question you’re asking. 

For the eager, it provides a working demonstration that proves technical feasibility before over-investing in a tool that might not solve your most pressing problem. 

The most valuable outcome of testing a use case often isn’t the model, but the clarity. Sometimes grounding reveals that your data isn’t ready to reliably answer that question yet.  

In a world of frantic AI adoption, that’s a massive strategic win. It prevents you from making operational decisions based on bad data and tells you exactly where to focus your data modernization efforts next. 

The discipline of starting small to build a repeatable practice

AI doesn’t have to be a mystery or a race. The organizations that make progress build a repeatable way to test, learn, and decide. 

They start with a clear question and test it with real data. They ground the results to make sure they hold up, and they use what they learn to decide what to do next. 

That cycle builds confidence. It gives hesitant teams proof they can move forward without having to choose the perfect tool, and eager teams a way to validate what’s worth scaling before they over-invest.  

Start with one answerable question. Then prove it. 

Let’s find your first answerable question.