Guide · ROI
How to Calculate the ROI of AI Automation (Formula and Worked Example)
"AI will save us 20 hours a week" is not a business case. To decide whether an AI project is worth it, you need what you'd want for any investment: the value it creates, the full cost, and how long it takes to pay back.
Key takeaways
- Count only the freed time your team actually puts to productive use.
- Count gross profit from new sales, not revenue.
- Subtract every cost: the build, the running costs, and the months before the system goes live.
- Report three numbers: first-year ROI, payback period, and three-year ROI.
Below is a practical method for estimating the return on an AI automation project. It's the same model behind our ROI calculator, so you can follow the math here and then plug in your own numbers.
Why most AI ROI estimates are too optimistic
- They treat every saved hour as cash. Unless freed time goes into paid work or avoids a new hire, it's extra capacity, not savings.
- They count revenue instead of profit. A $2,500 deal at a 40% gross margin adds $1,000 of profit, not $2,500.
- They ignore costs. Build fees, software subscriptions, and AI usage all come out of the return.
- They assume value on day one. Most systems take weeks to build and launch, and benefits only start at go-live.
The formula
Annual value = (hours saved per year × hourly cost × productive-use share) + (added sales per year × gross margin)
First-year ROI = (value in year one − cost in year one) ÷ cost in year one
Payback period = months to go live + build cost ÷ (monthly value − monthly running cost)
Value and running costs start when the system goes live; the build cost is paid up front. For a multi-year view, repeat the same calculation over 36 months.
Step by step
1. Measure the hours
Count the people who do the work and the hours each spends on it per week. Multiply by working weeks per year (48 is a reasonable default) and by the share of the task you expect to automate. Automating part of a task is far more realistic than automating all of it; our calculator starts at 40%.
2. Value the time correctly
Multiply those hours by the fully loaded hourly cost (wages plus benefits), then by the share of freed time that will go into productive work. If freed hours are likely to disappear into the day, count less of them.
3. Estimate the revenue lift conservatively
For lead response or follow-up projects, estimate a relative lift in your close rate. A 15% relative lift on a 15% close rate takes it to 17.25%, not 30%. Multiply the extra deals by your average deal value to get added sales.
4. Convert revenue to gross profit
Multiply added sales by your gross margin. This is the value the new revenue actually adds to the business.
5. Add up every cost
Include the one-time setup and build, monthly running costs (software, AI usage, hosting, and support), and your team's time if the project needs a significant amount of it.
6. Account for time to go live
If the system takes two months to launch, the first year only has ten months of benefits, but it still carries the full build cost. That single adjustment often changes the first-year picture more than any other.
A worked example
Take a service business where five people each spend 10 hours a week on repetitive admin at an average hourly cost of $45. It receives 80 new leads a month, closes 15% of them, and its average deal is $2,500 at a 40% gross margin. The project costs $12,000 to build and $800 a month to run, and it goes live after two months. We'll assume 40% of the admin is automated, 75% of the freed time is used productively, and faster follow-up gives a 15% relative lift in the close rate.
| Line | Calculation | Result |
|---|---|---|
| Hours saved per year | 5 × 10 × 48 × 40% | 960 hours |
| Value of time | 960 × $45 × 75% | $32,400 |
| Added sales | 80 × 12 × 15% × 15% × $2,500 | $54,000 |
| Added gross profit | $54,000 × 40% | $21,600 |
| Annual value | $32,400 + $21,600 | $54,000 ($4,500 a month) |
| Year-one value | $4,500 × 10 live months | $45,000 |
| Year-one cost | $12,000 + ($800 × 10) | $20,000 |
| First-year ROI | ($45,000 − $20,000) ÷ $20,000 | 125% |
| Payback period | 2 + $12,000 ÷ ($4,500 − $800) | 5.2 months |
| Three-year ROI | ($153,000 − $39,200) ÷ $39,200 | 290% |
Stress-test your assumptions
A sound business case still works when the optimistic assumptions don't hold. Here's the same project under tougher conditions:
| Scenario | First-year ROI | Payback | Three-year ROI |
|---|---|---|---|
| Base case (above) | 125% | 5.2 months | 290% |
| No revenue lift (time savings only) | 35% | 8.3 months | 134% |
| No revenue lift, half of freed time productive | −10% | 14 months | 56% |
The last scenario loses money in year one but still returns more than it costs over three years. That is exactly the kind of insight a single optimistic number hides, and it's why payback and three-year ROI belong next to the first-year figure.
What to measure after launch
- Record baselines before launch: hours per task, response time, close rate, and revenue per customer.
- Track the same metrics every month after go-live.
- Compare actual running costs with your estimate, including AI usage.
- Revisit the assumptions each quarter and update the business case with measured results.
Frequently asked questions
What is a good ROI for an AI project?
There's no universal benchmark. Compare it with your other uses of the same money, and look at the payback period and three-year return, not only the first-year ROI.
Should we count our team's time as a cost?
Yes, if the project needs significant time for interviews, testing, and training. Add it to the build cost at your team's hourly rate.
How do we estimate the revenue lift before launch?
Start conservatively, test on a subset of leads if you can, and replace the assumption with measured results after launch.
Is time saved the same as money saved?
Only if the freed time goes into revenue-generating work or avoids a hire. That's why the formula includes a productive-use share.