The 5-Question AI ROI Test for SMBs

Most small businesses I work with aren't short on AI tool ideas. They're short on a way to evaluate the ones they're considering. The default decision pattern is some combination of "this looked cool in a podcast," "a vendor pitched me hard," and "everyone else seems to be using it." That pattern leads, predictably, to subscription bloat and underused tools.

There's a better way, and it takes about five minutes per tool. Five questions. Run them before you sign up. Run them again at thirty days. The questions are simple, but most owners don't ask them — which is why most AI spending doesn't return what it should.

You don't need a technology background to do this. You need the same instincts you already use to evaluate any business expense: What does it do? What does it cost? Who's responsible? And how will I know if it's working?

Question 1: What workflow does this replace or augment?

This is the foundation question. If you can't name the specific workflow this AI tool will change, you don't have a use case — you have an attraction.

A workflow is a specific, repeatable sequence of steps that produces a defined output. "Marketing" isn't a workflow. "Writing the weekly customer newsletter" is. "Customer support" isn't a workflow. "Drafting first responses to refund requests" is.

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

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

If the workflow is fuzzy, the AI tool will be fuzzy too. The output will be impressive in demos and useless in your actual operation.

The fix: get specific before you subscribe. Watch your team work for thirty minutes. Identify a real, recurring workflow with a real frustration attached to it. Match the tool to the workflow — not the other way around.

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

You can't compute ROI without knowing the input. This question quantifies the problem.

Most owners wildly overestimate or wildly underestimate. The person who's done a task forever assumes it takes "a couple of hours." Watching it actually happen, it might be twenty minutes — or it might be eight hours spread across three people. Without measurement, you're making a financial decision on guesswork.

Think about it the same way you'd think about hiring a part-time employee to handle a task. You'd want to know how many hours the task actually takes before you decided what to pay for help. AI subscriptions are the same math.

• Workflow consumes 30 minutes a week. Even at 80% time savings, you're recovering 24 minutes a week — two hours a month. A $50/month tool needs to deliver more than that to make sense, and most don't once you factor in the time spent learning and switching contexts.
• Workflow consumes four hours a week. At 50% time savings — a realistic baseline for most AI-assisted work — you're saving two hours a week, eight hours a month. Now a $50/month tool is paying for itself clearly, and even a $200/month tool clears the bar.
• Workflow consumes fifteen hours a week. This is your highest-leverage AI candidate. Even modest improvements compound to significant savings. Pay attention to these workflows first.

The fix: time the actual workflow before you decide what tool to buy. The tools worth picking are the ones aimed at your highest-hours workflows — not the ones with the most polished sales demos.

Question 3: Whose job changes?

This question separates AI investments that succeed from ones that fail. Every AI workflow changes someone's job. Identifying the person — and getting their input early — is the difference between a tool that gets used and one that collects dust.

Three patterns to watch for:

• The job gets better. The person spends less time on the tedious part of their work and more time on the part they actually care about. This is the easy case. Adoption almost always follows.
• The job changes substantially. The person now spends part of their day reviewing and editing AI output instead of producing the work directly. This can be fine — but it requires a real conversation. People didn't sign up to be proofreaders. Some will adapt well. Some will quietly resent the shift.
• The job becomes redundant. The AI replaces enough of the work that the role itself is in question. Be honest with the person involved, early. Pretending otherwise destroys trust in ways that outlast the tool.

If you can't answer "whose job changes," you're not ready to deploy. Find that person first. Then have the conversation. They'll also know things about the workflow that you don't — which almost always improves the rollout.

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

If you can't define success at thirty days, you can't tell whether the tool is working.

This is where most SMB evaluations break down. The owner subscribes, the team uses it for a while, and at some point the question "is this actually working?" surfaces — but there's no agreed-upon answer. Different people have different criteria. Some say it's working because the output looks fine. Others say it isn't because no one's using it. Without a written success metric, the conversation goes in circles.

A useful 30-day success definition 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 workflow." 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. More than three and you lose the signal — you'll end up debating definitions instead of making a clear call.

The fix: write the success criterion down, in advance, where everyone can see it. Revisit it on day 30. Be honest about the answer.

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

This is the question that separates serious operators from casual experimenters. You have to define what failure looks like — and what action that triggers — before you commit.

Without this, AI tool failures linger for months. The team stops using it, but no one wants to call it. The owner forgot they signed up. The subscription auto-renews. Eventually someone notices the line item and asks what you're paying for — and the answer is: "I'm not sure anymore."

Define failure in advance:

• "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 workflow in two weeks, we cancel."

The fix: when you sign up, put a 30-day calendar reminder on your calendar with the specific failure criteria attached. When the alarm goes off, check the criteria. If they're met, the tool earned its place. If not, cancel — without sentiment.

This is harder than it sounds. There's genuine emotional resistance to canceling something you advocated for. Owners who can do it cleanly end up with leaner, more productive AI stacks than owners who can't.

The five-minute version

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

1. Workflow: What specific, recurring task does this change?
2. Volume: How many hours per week does that workflow 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?

If you can't fill in all five — without hand-waving — you're not ready to subscribe. Either get more specific, or find a different tool that maps to a workflow you can answer cleanly.

This isn't a high bar. It's the lowest bar that produces good AI investment decisions for a small business. Most owners clear it easily once they slow down enough to think it through. The discipline is in slowing down. The market doesn't reward speed of subscription — it rewards quality of decision.

Run the test before you sign up. Run it again at thirty days. Cancel what didn't earn its place. That's the entire framework — and it puts you ahead of most peer businesses, who are still buying tools the old way and wondering why the returns are invisible.