ROI & Getting Started
How to Measure AI ROI: The Honest Math
Most ROI claims about AI are marketing, not math. Here's how to run the numbers yourself — with a formula simple enough for a napkin.
What it is
Every AI tool vendor will tell you their product "pays for itself." Some do. But vendors measure ROI in the rosiest possible light: best-case users, cherry-picked tasks, and cost figures that leave out setup time. If you're a small business owner, the only ROI that matters is yours: does this tool save more money than it costs you, in your actual shop, with your actual staff?
Measuring AI ROI yourself is not complicated. The core formula is: (value of time or output gained) minus (cost of the tool plus the time spent on it). The hard part isn't the arithmetic — it's honestly estimating the inputs. This guide walks you through doing exactly that, before and after you adopt a tool.
There is one more trap to avoid: counting "hours saved" as money in your pocket when nothing about your staffing changes. If a tool saves your receptionist 3 hours a week but you don't reduce hours or reassign her to revenue work, you haven't saved money — you've bought slack. That's still valuable, but call it what it is.
Who it's for
Anyone considering paying for an AI tool, or wondering whether a free one is worth the staff time it takes to use. This is especially important before signing up for per-user monthly plans, which compound fast across a team.
Skip the formal math for genuinely free tools that take no setup — trying ChatGPT's free tier to draft a flyer costs you ten minutes; just notice whether it worked. Pull out the calculator once money or real staff hours are involved.
What it costs
Doing this measurement costs you nothing but an hour of thinking and a simple spreadsheet. Most tools offer a free tier you can use for the trial period, and paid plans are usually billed per user per month — often roughly $10–$50/user/month for mainstream AI assistants, but check the vendor's current pricing, it changes.
The real cost to account for is staff time: setup, learning, and the daily minutes of using the tool. That time is the part vendors never include in their ROI claims.
Target ROI: the honest math EXAMPLE
Here's the full formula in plain form: net monthly gain = (hours saved × hourly value of that time) + (extra revenue the tool plausibly helped create) − tool subscription − (hours your team spends using and managing the tool × hourly value). All of these are estimates — the point is to estimate them conservatively, not optimistically.
Example: If you pay staff $25/hr and this saves 5 hrs/week, that's about $500/month in time. If the tool costs $30/month, the example net is ~$470/month in freed time — before you count setup hours.
What to actually measure:
- Time per task before and after the tool (time it with a stopwatch for a week, both ways).
- The fully loaded hourly value of whoever's time is saved — wage plus a rough allowance for taxes and benefits, or what you'd pay a contractor for that hour.
- Error and rework rates: are outputs actually good, or are you spending saved time fixing AI mistakes?
- Whether "saved" time converted into something real: more customers served, more output, or genuinely reduced hours.
How to set it up
- Pick one task to measure. Choose the specific job you're considering AI for — e.g., "writing weekly social posts" or "summarizing timesheets." Don't measure "AI" in general; measure one task.
- Time the old way. For one to two weeks, have whoever does the task note the start and end time each time they do it. Write the times in a shared note or spreadsheet. This is your baseline.
- Price your time. Decide the hourly value to use: the employee's loaded hourly cost, or what you'd pay a freelancer to do it. Write it down and use the same number everywhere.
- Run a 30-day trial. Use the tool on the same task for a month. Track the time it takes now — including prompting, reviewing, and fixing output. Include any subscription cost and the setup hours.
- Do the subtraction. Old cost per month (time × hourly value) minus new cost per month (time × hourly value + subscription + setup amortized). Positive means it's paying; negative means it isn't yet.
- Check the quality, not just the clock. Are customers, staff, or your accountant noticing any difference in the output? A tool that halves the time but doubles the error rate is not a win.
- Decide: keep, fix, or drop. If the math is clearly positive, keep it and re-check quarterly. If it's close, change how you use it (better prompts, less review). If it's negative after an honest month, drop it without guilt.
Watch-outs and honest limitations
The biggest trap is counting savings that never materialize: "saved" hours that just become slower workdays, or output increases nobody asked for. Only count what converts into reduced costs or real revenue. Another common one is measuring the honeymoon period — the first two weeks when everyone's enthusiastic — instead of steady-state use after the novelty wears off.
Also watch for cost creep: free trials that convert to paid plans, per-user fees multiplying as you add staff, and add-on charges for extra usage. Read the pricing page's fine print before the trial ends. And remember that your honest-math number is an estimate, not accounting precision — it's a decision tool, not a tax filing.
What to measure in your first 30 days
- Minutes per task, old way vs. new way, logged the same way each time.
- Total subscription and usage costs actually billed during the trial.
- Setup and training hours spent by you and your staff.
- Rework: how often the AI's output needed meaningful fixing before use.
- One quality signal: an error rate, a customer comment, or a staff opinion — something beyond the stopwatch.