How to Find Your First AI Use Case (The $10M+ Business Guide)

Every CEO I work with asks the same first question:

“What should we actually build?”

It’s the right question. It’s also the question most “AI strategy” content avoids — because the honest answer requires you to understand the specific business, not the generic playbook.

I’ve run this exact exercise dozens of times for manufacturers, industrial services companies, and federal contractors. Here’s the methodology I use. You can run a rough version of this yourself in an afternoon.

Step 1: Stop Thinking About AI

Counterintuitive but critical. If you start by asking “where can we use AI?”, you’ll end up with a list of cool things that probably aren’t valuable.

Start instead with this question:

“Where in this business does a smart person spend hours doing work that’s mostly pattern-matching?”

That’s it. That’s the whole frame. AI is good at pattern-matching. So is a smart human — but a smart human is expensive, scarce, and gets tired. If you find a place in your operation where the bottleneck is “we don’t have enough smart people doing repetitive thinking,” you’ve found a candidate use case.

Step 2: Walk the Floor

Sit with five people for one hour each. Not your leadership team. Your operators. The people who do the actual work.

Ask them these questions:

• What’s the most repetitive part of your job?
• What do you wish someone would just do for you?
• Where do you spend time finding information instead of using it?
• What decisions do you make over and over that follow a pattern?
• What does the customer wait the longest for?

Write down what they say. Don’t filter. You’re looking for high-frequency, pattern-shaped friction.

In manufacturing, this exercise consistently surfaces the same three categories:

• Quoting and engineering review (someone reads specs and pattern-matches against the catalog)
• Document intelligence (someone reads PDFs, drawings, RFQs and routes them)
• Customer service triage (someone reads inbound, classifies, and assigns)

You’ll find your version of these in your business.

Step 3: Score by Dollar Impact, Not Feasibility

This is where most companies go wrong. They start filtering by “can we technically build this?” before they’ve asked “is this worth building?”

For each candidate, estimate:

• Frequency: How many times per week does this happen?
• Hours saved per instance: Realistic estimate. Not best case.
• Fully-loaded cost of those hours: Salary × 1.4 ÷ working hours.
• Quality lift: Will customers notice? Will errors drop?
• Risk avoided: Compliance, safety, regulatory — what does failure cost?

Multiply. You’ll find that most “exciting” AI use cases are worth $20K/year. Some boring ones are worth $400K/year.

Build the boring ones first.

Step 4: Filter by AI Suitability

Now — and only now — ask the feasibility question. A good AI use case has three traits:

1. Inputs are mostly text, images, or structured data. Not exotic sensor streams. Not multi-modal video. The first project should use building blocks that already exist.

2. The “right answer” can be checked. You can tell within minutes of running a sample whether the AI is producing useful output. If you can’t tell good from bad, you can’t iterate.

3. The decision isn’t life-critical on day one. Your first AI project should not be the thing that fires people, ships product, or signs contracts unsupervised. Augment humans first. Automate humans later — maybe.

A use case that hits all three of these can be shipped in 60–90 days with a small team and a moderate budget. A use case that fails one of these is a research project, not a deliverable.

Step 5: Pick the One That Pays Back in 90 Days

You’re looking for a single project that:

• Is worth at least $100K/year in saved time or unlocked revenue
• Can ship to production in 60–90 days
• Has one identifiable internal champion who’ll use it daily
• Won’t get killed by IT, compliance, or legal at month two

If you have three candidates that meet this bar, pick the one with the strongest internal champion. Champion strength matters more than technical elegance. A medium-good system that gets used daily by a believer beats a perfect system that the team avoids.

The Honest Math

When I run this exercise for a $20M manufacturer, the first project we identify is typically worth $200K–$500K/year and costs $30K–$80K to ship.

That’s a 4–6x first-year ROI. Realistic, not promotional.

The mistake CEOs make is hunting for the “transformational” use case before they’ve shipped the boring one. There is no transformational use case. There’s a sequence of boring ones that compound into a different company.

What to Do This Week

If you’re a CEO reading this, here’s the action:

1. Block 90 minutes on Friday.
2. Pick five operators across different functions.
3. Run the questions in Step 2 with each of them for 15 minutes.
4. Write down the patterns you hear.
5. Score the top three by dollar impact.
6. Sleep on it. Pick one Monday morning.

You can run the rest of the project from there. Or you can call someone who’s done this 50 times and skip the false starts.

---

The AI roadmap session I run with manufacturers and industrial services companies is exactly this exercise, sharpened by a few years of pattern recognition. It costs $2,500 and ships a written, board-ready plan. Book a session.