Part 3 of 3 — The AI ROI Series
In Part 1, we covered the green flags; conditions where AI is likely to deliver real ROI. In Part 2, we covered the red flags; the conditions that quietly sink AI projects after the contract is signed. Part 3 is about putting it all together and know if you should commit.
Now comes the part where we stop talking concepts and do some math.
Because here’s the thing: gut feel and vendor case studies don’t really create a business case. The only number that matters is what this specific process, at this specific company, with these specific costs, looks like on paper.
My ROI calculator has 5 parts. 4 parts input, 1 part analysis.
Current State
The foundation of any honest AI evaluation is knowing what the process costs today. Not a manager’s estimate. Not a rough guess based on job titles and salary ranges. A real time study.
That means tracking the actual people who do the job for at least a few days. Ideally, one full cycle of normal variation, at least a couple weeks to capture representative data. You’re measuring:
- How often the task happens per month
- How long it takes per instance, on average, including prep, rework, and wrap-up
- How much time goes to exceptions and edge cases specifically
That last number matters more than people expect. Exceptions are where AI struggles most, and where human oversight costs accumulate. If 30% of your task time is exceptions, your realistic AI savings rate is capped well below 100% regardless of what the tool can do.
Once you have the time study data, calculate what that time actually costs. Not base salary, fully loaded. Salary plus benefits, payroll taxes, and an overhead allocation. A reasonable rule of thumb is 1.3 to 1.5 times base salary.
That’s your baseline. Everything else in the analysis gets measured against it.
One-time Setup Costs
To get your project started there will likely be onboarding costs. While vendors may have some templates to start the automation, any process that deviates from their standard scenario requires customization. And while some tasks may be performed by internal staff, that time is not really free.
For the ROI calculation, you will want to gather the following information:
- Integration and onboarding fees
- Prompt engineering and configuration time
- Staff training
- Parallel testing period — running old and new processes simultaneously to validate results
You may want to add some contingency to the setup costs. It almost always takes longer than projected.
Monthly Ongoing Costs
AI, people have ongoing costs. The purpose of this evaluation is to compare them. The monthly costs will include:
- Subscription or licensing fees
- API and usage-based charges (these grow with volume, budget accordingly)
- Internal management time to oversee and maintain the tool.
The ROI calculator also asks for the percentage of human time AI actually replaces. The answer should never be 100%. If you expect an AI solution to be set and forget it, you may be disappointed. Just like the best employee, at the very least, you need to spot check the results. Is there exception handing? That will be routed to a human. New scenario or change in the business process? AI may need addition training and new prompting.
Use numbers that reflect your actual workflow, not the best-case scenario. A conservative savings rate that holds up is worth far more than an optimistic one that collapses on contact with reality.
Growth & Cost Escalation
A month-one snapshot tells you whether the economics are in the right ballpark. A multi-year projection tells you whether they hold. If you are evaluating a longer time frame, there are additional factors to consider.
Business growth rate — More volume means AI savings grow. If you’re planning to scale, the ROI picture often improves significantly over time. This is the legitimate “we won’t have to hire additional people” argument.
AI tool cost escalation — SaaS vendors raise prices. Usage fees grow. Budget for 10–20% annual cost increases on your AI tooling and see how it affects the long-term picture. If the economics only work under the assumption that tool costs stay flat, that’s a risk worth identifying.
Results
Claude and I built an ROI calculator that works through all of this systematically — time study inputs, fully-loaded labor costs, setup costs, AI tool costs, growth rate, and cost escalation — and projects cumulative ROI over 12, 24, or 36 months. The calculator gives a YES, NO or MAYBE recommendation with the data to support it.
The ROI increases over time and the breakeven is in the desired timeframe? YES! Are the numbers kind of flat? MAYBE. You may choose to proceed. Maybe you have a busy season. AI can provide seasonal labor, The important thing is making decision based on data, not guesses.
The Bottom Line
AI is not a strategy. It’s a tool. And like any tool, it works brilliantly in the right application and expensively in the wrong one.
The businesses that get the best long-term results from AI aren’t the ones that move fastest. They’re the ones that asked the right questions first; about volume, consistency, error tolerance, true costs, and realistic savings. Then they made decisions based on what the numbers said.
That’s not skepticism. That’s how you make AI work.
Want to see the calculator in action? Watch the video; Coming Soon!
This is Part 3 of a 3-part series on AI ROI for small and mid-sized businesses. Part 1: Green Flags — When AI Is Actually Worth the Conversation Part 2: Red Flags — When AI Will Cost You More Than It Saves
Have a process you’re evaluating? Questions about the framework? Get in touch.