How to Calculate AI ROI Before You Buy

The formula

The value is the share of the work the AI takes over, priced at what that work really costs you, plus any extra revenue you can actually measure. The total cost is the setup plus the running costs plus the time your team spends checking the AI's work. The four steps below fill in each piece.

Step 1: Price the work, not the salary

If the AI takes over part of a job, the value is that part of the job's full cost: salary plus payroll taxes, benefits, equipment, and hiring. A $38,000 receptionist really costs about $53,230 a year. Then estimate the share of the work the AI can realistically take on. It is never 100%. Half is a solid assumption for most customer-facing roles, and 60% is achievable when the work is very repetitive.

Step 2: Count every cost

• Setup: the build or configuration, plus your team's time explaining prices, policies, and rules.
• Running costs: the subscription or monthly fee, usage, phone lines, and the AI model behind it.
• Oversight: hours a week spent checking its work, priced at the hourly cost of whoever does it.
• Maintenance: updates when your prices, hours, or tools change.

Step 3: Choose a time horizon

The first year carries the full setup cost, so year-one ROI always looks worse than year two. Use a three-year view to compare options, and use the payback period to see how long you are out of pocket. A setup cost that pays back in a few months is a very different decision from one that takes two years.

Step 4: Add revenue gains only if you can measure them

Answering every call or replying to leads in seconds can win jobs you used to miss. That is real, but it is also where estimates get inflated. Only count revenue you can actually track, such as appointments booked by the agent or calls answered that used to go to voicemail, and keep it separate from the cost savings so you can see which one is doing the work.

A worked example

Take a receptionist who costs $53,230 a year fully loaded, and an AI agent that takes on 60% of the work.

• Value of the work it takes on: 60% × $53,230 = $31,938 a year.
• Running costs: $500 a month plus $250 a month in usage, plus three hours a week of checking (about $3,992 a year) = $12,992 a year.
• Net savings: $31,938 − $12,992 = $18,946 a year.
• Setup: $5,000 one-time.
• 3-year ROI: (3 × $18,946 − $5,000) ÷ ($5,000 + 3 × $12,992) ≈ 118%.
• Payback: $5,000 ÷ ($18,946 ÷ 12) ≈ 3.2 months.

The calculator below runs this exact math, so you can swap in your own numbers.

How to measure AI ROI after launch

Before launch, record a baseline for a few weeks: calls answered and missed, how fast leads get a reply, bookings, and the hours your team spends on the task. After launch, track the same numbers along with what the AI costs each month. Compare month over month, and count only the work it actually took over. Ignore the first month or so, which is mostly tuning.

Common mistakes

• Using salary instead of the fully loaded cost of the work.
• Assuming the AI handles the whole job.
• Leaving out the time spent checking its work.
• Forgetting the setup cost when quoting ROI.
• Counting revenue you can't measure.
• Judging it on the first month, before it's tuned.