The 5-Question AI ROI Test for Small Businesses

You're not short on AI tool ideas. You're short on a way to size up the ones you're eyeing. So the call gets made on a mix of "this looked cool in a podcast," "a vendor pitched me hard," and "everyone else seems to be using it." Buy that way for a year and your bank statement fills up with subscriptions nobody opens.

There's a better way, and it takes about five minutes per tool. Five questions. Ask them before you sign up. Ask them again at thirty days. They're simple questions. Most owners just don't ask them, which is why most AI spending doesn't pay back what it should.

You don't need a tech background for this. You need the same instincts you already use on any business expense. What does it do? What does it cost? Who's on the hook for it? And how will you know if it's working?

Question 1: What task does this replace or augment?

A real use case begins with a specific task you can name. If you can't name the one this tool will change, you're chasing a shiny thing.

A task is a repeatable sequence of steps that spits out a defined result. "Marketing" isn't a task. "Writing the weekly customer newsletter" is. "Customer support" isn't a task. "Drafting first replies to refund requests" is.

If you can't fill in this sentence cleanly, stop.

"We're going to use [tool] to [replace or augment] our current process for [specific task]."

So before you subscribe, spend thirty minutes watching your team work, find the recurring task that's a real headache, and point the tool at that.

Question 2: How many hours per week does that task consume?

You'll probably guess wrong in one direction or the other. Someone who's done a task for years swears it takes "a couple of hours." Watch it happen and it's twenty minutes. Or it's eight hours spread across three people. Guess at that number and you're making a money decision on a hunch.

Think of it like hiring a part-timer for the task. You'd want to know how many hours it really takes before you decided what to pay for the help, right? An AI subscription is the same math.

• The task eats 30 minutes a week. Even at 80% time savings, you're getting back 24 minutes a week, two hours a month. A $50/month tool has to beat that to earn its spot, and most won't once you count the time spent learning it and switching gears.
• The task eats four hours a week. At 50% time savings, a fair baseline for most AI-assisted work, you're saving two hours a week, eight hours a month. Now a $50/month tool pays for itself easily, and even a $200/month tool clears the bar.
• The task eats fifteen hours a week. This is your best AI candidate by a mile. Even small improvements pile up fast. Start here.

So time the actual task before you decide what to buy. The tools worth your money go after your highest-hours work, no matter how slick the demo looks.

Question 3: Whose job changes?

Every AI tool you bring in changes somebody's job. Name that person and pull them in early, and you've got a tool people use instead of one that collects dust.

The change usually lands one of three ways.

• The job gets better. They spend less time on the tedious part and more on the part they care about. This is the easy one, and it's what good AI is for. The busywork comes off their plate, and they take to it fast.
• The job changes a lot. Now they spend part of the day checking and editing AI output instead of doing the work themselves. That can work, but it takes a real conversation. Nobody signed up to be a proofreader. Some folks adapt fine. Some will resent it, and you can't blame them.
• The job goes away. The AI does enough of the work that the role itself is in question. Be honest with that person, early. Pretend otherwise and you break trust.

If you can't answer "whose job changes," you're not ready to roll this out.

Question 4: What does success look like in 30 days?

If you can't say what success looks like at thirty days, you can't tell whether the tool is working.

This is where most of these decisions fall apart. You subscribe, the team uses it for a while, and eventually someone asks "is this actually working?" and nobody has an answer everyone agrees on.

A good 30-day success line is concrete and measurable.

• "By day 30, the operations lead is using this tool at least three times a week without being reminded." Adoption metric.
• "By day 30, the time-to-draft for our weekly newsletter has dropped from four hours to under one hour." Time savings metric.
• "By day 30, we've published at least eight pieces of content using this tool." Output volume metric.
• "By day 30, the team has rated the output quality as acceptable on at least 70% of attempts." Quality metric.

Pick one. Two if you want both adoption and outcome.

Write the success line down ahead of time, where everyone can see it. Come back to it on day 30. Be honest about the answer.

Question 5: What does failure look like, and what will you do about it?

Before you commit, say what failure looks like and what you'll do when you see it.

• "If the operations lead isn't using this three times a week by day 30, we cancel."
• "If output quality is rated unacceptable on more than 50% of attempts, we cancel."
• "If we haven't shipped any output through this tool in two weeks, we cancel."

So the day you sign up, drop a 30-day reminder on your calendar with the failure criteria attached. When it pings, check them.

It stings to cancel something you talked everyone into. But learn to do it cleanly and your AI stack stays lean and actually useful. Cling to the duds and it bloats.

The five-minute version

For any AI tool you're considering, write down answers to these five questions on one page:

1. Task: What specific, recurring task does this change?
2. Volume: How many hours per week does that task consume today?
3. Person: Whose job is affected, and have they been consulted?
4. Success at 30 days: What's the specific, measurable criterion?
5. Failure at 30 days: What's the specific criterion that triggers cancellation?

Can't fill in all five without hand-waving? You're not ready to subscribe. Get more specific, or go find a tool that maps to a task you can actually answer.

This isn't a high bar. It's about the lowest bar that gets you a good AI buy for a small business.

Run the test before you sign up. Run it again at thirty days. Cancel what didn't earn its keep. Do that, and you're already ahead of the businesses down the road still buying tools on a hunch and wondering where the returns went.