To measure the ROI of AI automation, baseline what the manual process costs you today (hours times loaded labor rate, plus lost revenue), total the one-time build cost and monthly run cost, then divide the annual savings by that total cost. A Microsoft-sponsored IDC study of 4,000+ leaders found businesses earn an average of $3.70 for every $1 invested in generative AI, with payback inside about 13 months.
Why most ROI numbers feel fuzzy
The headline stats are real, but they're averages. The same IDC study notes top-tier deployers see roughly $10 back per dollar, while Boston Consulting Group found many late movers struggle to measure any positive ROI at all. The gap isn't the technology, it's the math. If you never baselined the manual process, you have nothing to compare the automation against. So before you buy anything, you measure what the status quo already costs you.
The 5-step ROI framework
- 1. Baseline the manual cost. Count the hours your team spends on the task each month, multiply by a fully loaded labor rate, and add any revenue you lose to slow or missed work (for example, leads that never get a callback).
- 2. Total the cost of the automation. Add the one-time build cost to the recurring monthly run cost (software, API usage, maintenance) to get a true annual cost of ownership.
- 3. Pick 2 to 4 metrics you can actually track. Hours saved, revenue recovered, error-rate reduction, and response time. Skip anything you can't pull a real number on.
- 4. Calculate ROI and payback. ROI = (annual gain minus annual cost) divided by annual cost. Payback (in months) = total build cost divided by monthly savings.
- 5. Re-measure after 90 days. Compare live numbers against your baseline, not against the vendor's promise. Adjust scope where the math underperforms.
Metrics worth tracking
| Metric | How to measure |
|---|---|
| Hours saved per week | Time on the task before automation minus time after. The St. Louis Fed found 33.5% of daily generative AI users saved 4+ hours a week. |
| Revenue recovered | Extra deals closed or leads saved (e.g. after-hours calls answered) times your average deal value. |
| Cost reduction | Loaded labor cost of the manual process minus the automation's annual run cost. |
| Payback period | Total build cost divided by monthly savings. IDC/Microsoft put the average GenAI payback near 13 months. |
| Error / rework rate | Mistakes or do-overs before automation minus after, times the cost to fix each one. |
A quick worked example
Say a missed-call problem costs a contractor 10 leads a month at a $1,500 average job. Recover even a third of those and that's roughly $5,000 a month in revenue the automation makes possible. Against a $4,000 build and $300/month to run, the build pays for itself before the second month, and the first-year ROI clears that $3.70-per-dollar benchmark several times over. Your real numbers will differ, which is exactly why you measure your own baseline first.