The 3-Question Test for Any AI Investment

Every week, new AI tools launch. Every week, business owners ask some version of the same question: is this worth it? Here’s a simple framework that cuts through the noise.

Question 1: What Specific Task Does This Replace?

Not “what category of work” — the specific task. “AI for customer service” is too broad. “Answering the 15 most common questions we get via our contact form” is specific enough to evaluate. If you can’t name the task, you can’t measure whether the tool is doing it well, and you can’t calculate whether the cost is justified.

If a tool’s marketing can’t clearly answer this question, that’s your first red flag.

Question 2: How Much Time or Money Does That Task Currently Cost?

This is where most businesses skip a step. Before adopting any tool, benchmark the current cost. How many hours per week does someone spend on this task? What’s the hourly cost of that labor? What’s the error rate, and what does each error cost?

You don’t need a precise number. A rough estimate works. “This takes my assistant about 5 hours a week, and she earns $25/hour, so it costs roughly $500/month” is enough to evaluate a $49/month tool that could eliminate 80% of those hours.

Question 3: What Breaks If the AI Gets It Wrong?

AI tools make mistakes. The question isn’t whether errors will happen — it’s what happens when they do. A mistake in your internal scheduling tool is annoying. A mistake in an automated email to your best client is a relationship problem. A mistake in a compliance document could be a legal problem.

The higher the stakes of an error, the more human oversight you need in the loop. This doesn’t mean avoiding high-stakes workflows — it means designing the right checks before you automate them.

Running the Test

Pick any AI tool you’re considering right now and run all three questions. Specific task. Current cost. Consequence of error. In most cases, this exercise either makes the decision obvious or surfaces the gaps in your thinking before you’ve committed anything.

AI investments that pass this test tend to deliver. Those that can’t answer question one rarely do.

If you want help running this analysis on a workflow in your business, reach out here. It usually takes less than an hour.