Ask a small business owner whether AI is worth it and you’ll usually hear “it saves us time.” Ask how much time, and whether that time turned into money, and the answer gets vague. That’s the gap this guide closes. AI ROI isn’t hard to measure for a small business, but it does take a baseline, honest cost numbers and a few weeks of tracking.
Below is what the research says, a simple formula, what to measure for common tasks and a worked example you can adapt.
What the research says about AI ROI
The evidence points both ways, which is exactly why you should measure in your own business:
- Time savings are real but modest. In a St. Louis Fed survey, workers who used generative AI saved an average of 5.4% of their work hours in November 2024, or about 2.2 hours in a 40-hour week. A 2025 update found 37.4% of working-age adults used it at work by August 2025.
- Gains are uneven. In a study of 5,179 customer support agents published by NBER, AI raised issues resolved per hour by 14% on average and by 34% for newer, less experienced agents. The most experienced agents saw minimal gains.
- People misjudge their own speed. In a METR study of 16 experienced software developers, tasks took 19% longer with AI. Afterward, the developers still estimated that AI had made them 20% faster.
- Pilots without a plan don’t pay. A widely cited MIT report, summarized by Virtualization Review, found that 95% of organizations saw no business return from their generative AI spending. It pointed to tools that didn’t fit into daily workflows.
Small businesses are already using AI. A May 2026 survey by the U.S. Chamber of Commerce Foundation and Ipsos found that half of small business workers use AI at work, mostly for drafting, summarizing and brainstorming. However, only about 1 in 10 had been offered formal training.
A simple AI ROI formula
AI ROI = (value gained − total cost) ÷ total cost
Value gained can include:
- Hours saved, multiplied by the loaded hourly cost of the person who saved them
- Extra revenue, like more appointments booked or faster collections
- Avoided costs, like overtime, temp help or missed calls
Total cost should include:
- Subscriptions for every seat, used or not
- Setup and integration time
- Training time for each person
- Ongoing admin, like managing accounts and updating instructions
Review time matters too. When someone spends 10 minutes checking an AI draft, subtract that from the time saved.
Measure your baseline before you start
To measure AI results, pick one to three tasks that happen often and take real time. Then track them for two weeks before anyone uses AI:
- Time per task. A rough log is enough.
- Volume. How many emails, quotes or invoices in a typical week.
- Quality. Errors, redo rate or customer complaints.
- Turnaround. How long a customer or coworker waits.
Use numbers your systems already keep where you can, because they’re more reliable than memory. The METR result is a good reminder that how fast work feels isn’t the same as how fast it is.
What to measure for common tasks
| Task | What to measure | Where to find it |
|---|---|---|
| Customer emails and messages | Time to first response, messages handled per hour | Shared inbox or help desk |
| Invoicing and collections | Days to get paid, overdue balance | Accounts receivable aging report |
| Phone answering | Missed calls, after-hours bookings | Phone system and calendar |
| Proposals, quotes and reports | Hours per document, rounds of revisions | Time tracking |
| Bookkeeping data entry | Transactions processed per hour, corrections | Accounting system |
| Marketing content | Pieces per month, time per piece | Calendar and platform analytics |
These numbers turn AI time savings into something you can compare month to month. Our guides to automating invoicing in QuickBooks and AI customer service go deeper on the metrics for those tasks.
A worked AI ROI example
Here’s an example with assumptions you can swap for your own. An eight-person office gives four staff members Microsoft 365 Copilot Business at $21 per user per month on an annual plan, from Microsoft’s pricing page. We’ll assume a loaded labor cost of $40 an hour.
Costs for the first year:
- Subscriptions: 4 seats × $21 × 12 months = $1,008
- Setup and training: 18 hours × $40 = $720
- Total: $1,728
Measured savings: After 90 days, each person saves a net 1.5 hours a week after reviewing AI output. That’s below the St. Louis Fed average of 2.2 hours. Over 48 working weeks, that’s 288 hours, worth $11,520.
AI ROI: ($11,520 − $1,728) ÷ $1,728 = 567%.
That number assumes every saved hour is worth full value. Saved hours only become money if they go to billable work, more sales or less overtime. A more conservative approach counts only half the hours: ($5,760 − $1,728) ÷ $1,728 = 233%. Report both, and be clear about which one you’re using.
Common mistakes when measuring AI ROI
- Counting time that isn’t really saved. Review and rework eat into the gain.
- Measuring too early. People get faster with practice, so measure at 30, 60 and 90 days.
- Paying for unused seats. Check who actually uses their license each month.
- Ignoring quality. Faster work with more errors isn’t a win.
- Spreading too thin. One workflow done well beats five half-finished pilots.
When to scale, fix or stop
Set a decision rule for AI ROI before you start. At 90 days, scale if the net value is positive and quality held steady. Fix the setup if results are close, which usually means better instructions or training. Stop if the numbers aren’t there, and move the budget to a different task.
Our AI rollout plan and 30-day plan to start using AI cover picking the first tasks. Our guide to what AI implementation costs breaks down the cost side, and our comparison of Copilot, ChatGPT, Claude and Gemini covers subscription prices.
Getting help in Asheville and across North Carolina
Good People Technologies helps with AI implementation for small businesses in Asheville, Western North Carolina and across the state. That includes setting up the baseline, choosing the first workflows and building a simple AI ROI scorecard you can review every month. To talk it through, start with a free 15-minute operations and technology audit.
Frequently asked questions
What’s a good AI ROI for a small business?
There’s no standard number. Compare it with your other uses of the same money and time. Simple tools with monthly subscriptions should show positive net value within a few months, or they’re not the right fit.
How long does it take for AI to pay off?
Measure at 30, 60 and 90 days. Subscriptions start right away, but AI time savings grow as people learn the tools.
Should we count time saved as money?
Only when the time turns into revenue or avoided costs. Otherwise, report hours saved separately from dollars.
What if employees say AI helps but the numbers don’t show it?
Trust the measurements, then ask why. The METR study showed that people can feel faster while actually working slower.
Sources, checked November 2026: the St. Louis Fed on time savings and adoption in 2025, NBER, InfoWorld on the METR study, Virtualization Review on the MIT report, the U.S. Chamber of Commerce Foundation and Microsoft 365 Copilot pricing.
Hi, I’m Benji, founder and CTO of Good People Technologies in Asheville, NC. We help small and mid-sized businesses put AI to work safely and connect it to the systems they already run, from QuickBooks to NetSuite.
For over a decade, we’ve helped businesses solve complex operations and technology problems. When not working, I’m hiking or enjoying rivers in western North Carolina with my wife and two dogs.
