Stop Looking for AI Use Cases. Find the Business Problems Worth Solving.

We need to talk about this

Walk into almost any company's AI strategy meeting and eventually someone asks:

“What are our AI use cases?”

A few weeks later, there is a spreadsheet.

Forty-seven ideas. Customer service agents. Automated reporting. Sales copilots. Contract review. Forecasting. Internal knowledge search.

Everything gets ranked by feasibility, impact and effort.

It looks like progress.

But there's a problem.

The conversation started with AI instead of the business.

McKinsey's latest research illustrates the disconnect. Eighty percent of respondents report individual productivity benefits from AI, yet only 37% attribute any EBIT impact to it. Only about 6% qualify as AI high performers. BCG separately found that more than half of CEOs see a missing link between AI and P&L, while only 14% clearly define P&L impact across all AI initiatives.

The issue isn't a shortage of things AI can do.

AI can do thousands of things.

The question is which ones are worth doing.

Start with the value leak

Before discussing models, agents or software, ask:

Where is the business currently losing money, time, capacity, customers or decision quality?

Look for situations like:

A sales organization generating plenty of opportunities but losing deals because quotes take five days.

A manufacturer carrying excess inventory because forecasting is slow and disconnected across regions.

An insurer whose adjusters spend hours gathering information before they can exercise thirty minutes of actual judgment.

A finance team whose monthly close requires hundreds of manual reconciliations across systems.

Those are business problems.

Now AI has something useful to work on.

Bain has made the same point bluntly: your business doesn't need a separate AI roadmap; your business roadmap needs AI. PwC similarly recommends concentrating investment in a few high-value areas rather than spreading AI across dozens of disconnected initiatives.

Use a simple opportunity filter

For every candidate problem, answer six questions.

1. What is happening today?

Establish a baseline.

How long does it take? How much does it cost? How often does it happen? What is the error rate? What revenue is being lost?

If you don't know the starting point, you won't know whether AI improved anything.

2. What is the value at stake?

Put a dollar sign—or another meaningful business metric—behind the problem.

If reducing quote time from five days to one could reasonably produce $4 million in additional annual gross profit, you have something worth investigating.

If automating a task saves twelve people three minutes per month, you probably don't.

3. Where is the actual bottleneck?

Don't assume the obvious task is the problem.

Maybe quote generation isn't slow because employees write slowly. Maybe pricing approvals require four departments.

Automating the document doesn't fix the bottleneck.

4. Could AI materially change it?

Now—and only now—bring in the technology.

Can AI eliminate work, improve a decision, reduce a handoff, surface information faster or create a capability that previously wasn't possible?

5. Can we measure the outcome?

Define success before implementation.

“Employees use the AI tool” is not success.

“Average quote turnaround falls from five days to 24 hours while maintaining gross-margin thresholds” is.

6. Who owns the result?

Every initiative needs someone accountable for the business outcome—not merely the technical delivery.

If nobody owns the number, don't be surprised when nobody owns the result.

Then rank the problems, not the AI

You don't need 47 priorities.

Start with one to three problems where the combination of economic value, frequency, measurability, feasibility and organizational ownership is unusually strong.

Then build.

This may sound less exciting than announcing an enterprise-wide AI transformation.

It is also much more likely to work.

AI isn't the strategy.

The business is.

The companies that create the most value from AI won't necessarily find the cleverest use cases.

They'll become exceptionally good at identifying expensive problems—and solving them.