We Were Wrong: Why Being Honest About Use Cases is So Important

Part of the AI Readiness Series

In May 2026, I argued that AI has to live inside the systems people already use, and that a separate portal puts adoption at risk. Where the tool sits does, of course, matter, but we made it the fifth foundation of our Five Foundations of AI Readiness, and I now believe that was a mistake. The question that decides whether an AI effort succeeds comes earlier: is this a use case your organization can actually handle today?

What “Honest” Means Here

Most organizations pick their AI use cases from a list of what looks exciting, what a vendor demonstrated or what a senior leader read about. These lists of “greatest hits” rarely get tested and reviewed against the organization that would have to operate and support the result. An honest evaluation strategy would test each use case candidate against three realities as they exist today, not as they’re planned to be.

1. Data. Does the information this use case depends on exist, and can the people involved get to it? Is it accurate enough that you’d act on the results it produces, and does everyone agree on what the fields mean? A use case that depends on customer records from two systems that disagree about who the customer is has a data problem, and the AI will inherit it.
2. Process. Is the work done the same way by every person every time? AI applied to a process nobody has written down, or one that varies by who’s on shift or one that has known pain points and bottlenecks tend to magnify the problems, not solve them. Sometimes the honest finding is that the process needs fixing first, which probably costs less than any AI project.
3. Leadership direction. Is there a named owner, a measure of success and agreement on what happens with the time or capacity freed up? Without those, there’s no way to tell afterward whether the effort worked, and the pilot drifts.

A use case can be attractive and still fail one of these tests. When it does, the honest answer is “not yet,” along with what would have to change. That’s a useful result; it’s cheaper to learn it in a working session rather than after the build.

Why This is Hard to Do on Your Own

It’s difficult to judge your own organization’s readiness with a straight face. The people proposing the use case usually want it to work. The people who’d have to say the data isn’t good enough are often the ones who maintain it. And nobody wants to be the one who tells leadership their brilliant idea has to wait.

That’s where a second set of eyes helps, and there are a few places to get one:

• A peer inside your organization who has nothing riding on the project. Someone from a different function will ask what the acronyms mean and where the numbers come from, which is just what you need.
• A peer at another organization in your industry. Maybe they’ve tried something similar, and they can tell you where their plan went off the rails.
• A trusted advisor who regularly assesses readiness and outcomes across industries, and is not aligned to any specific technology solution, like the professionals on our Digital Workplace Solutions and Management Consulting team.

Any of the three can do the job. What matters is that they ask you to show the evidence behind each answer: the data, the process as it’s actually run and who owns the outcome.

What Comes Out of an Honest AI Use Case Evaluation

Once a use case candidate passes, there’s still more to decide. You might configure tools you already license, buy a product built for the task, have people use a general AI assistant with clear rules and a checking standard, or build an integration. Where the tool lives becomes just one input to that choice, which is a smaller role than I gave it in May. Embedding the tool still makes sense for work that lives inside a transactional system. But a separate AI workspace is often the right place for drafting and research.

For each candidate that moves ahead, the recommendation should state the costs to start and keep running, including review time and upkeep. It should also say what you expect to get back in operational terms, what it depends on and where the estimate is soft. Management makes the decision, and the advisor’s job is to make sure the choice is an informed one.

Here’s a hypothetical: A finance team wants AI to match supplier invoices to purchase orders. The honest check turns up two problems. Purchase order numbers are entered several different ways, and nobody owns the exception process. The team decides to standardize the numbering and designate an owner first, then revisit. Nothing was built initially, and a failed pilot didn’t get added to the list of reasons AI doesn’t work here. That’s still a win.

I was wrong to treat deployment location as the deciding question. The deciding question is whether you can honestly say what your organization is ready for today. “Honest Use Cases” becomes our Fifth Foundation of AI Readiness because it’s the logical culmination of the assessment you do in the first four Foundations: Clear Direction, Safe Usage, Training and Data Readiness.

80% of CIOs surveyed by Gartner said they don’t know which use cases to prioritize. Honest use cases is where to start.