How to Calculate AI ROI
The formula, the inputs, and the three discounts most people skip
A practical method for working out whether an AI project pays back, including the arithmetic, a worked example you can check, and the corrections that separate a projection from a hope.
How do you calculate AI ROI?
Work out the annual value the AI creates (hours saved times fully-loaded cost per hour, plus any revenue added), subtract the build and annual run costs, and divide the net by the cost. Then apply three corrections almost everyone skips: an adoption haircut, a first-year ramp, and an honest answer on whether freed hours become money. Use payback period rather than a single ROI percentage to make the actual decision.
Key Facts:
- Annual value = hours saved × loaded hourly cost (+ revenue uplift)
- Total cost = one-time build + annual run cost (commonly 15% to 25% of build)
- Payback (months) = build cost ÷ monthly net value
- Apply an adoption haircut: nobody captures 100% of the theoretical saving
- Year one is never a full year of value, so ramp-discount it
Why Conservative Math Matters Here
Sourced industry figures, not our own claims. Primary sources only.
Of generative-AI pilots showed no measurable P&L return. Preliminary research, and contested, but it is the clearest signal that optimistic projections are the norm.
Source: MIT NANDA (2025, preliminary)Of agentic-AI projects Gartner expects to be canceled by the end of 2027, on runaway cost and unclear value.
Source: Gartner (2025)Expected AI spend per employee in 2026, up 50% year over year. The budgets being justified by these calculations are real money.
Source: Federal Reserve Bank of AtlantaThe Inputs You Actually Need
Seven numbers. If you cannot fill these in, the problem is not the calculation, it is that the project is not scoped yet.
| Input | What it means | How to get it honestly |
|---|---|---|
| Hours on the workflow | Person-hours per month spent on the work you want AI to take on | Add up the time across everyone involved. Ask them; do not estimate from the org chart |
| Loaded hourly cost | Salary plus benefits and overhead for the people doing that work | Usually 1.25x to 1.4x base pay. A $75K salary is roughly $50 to $55 per hour loaded |
| Automation rate | The share of that time AI can realistically remove | Well-scoped first projects often remove 30% to 50% of a repetitive workflow, not all of it |
| Build cost | One-time design, build and integration | From a scoped quote. If nobody has scoped it, this is a placeholder, not an input |
| Annual run cost | Model usage, infrastructure, monitoring and maintenance per year | Commonly 15% to 25% of build cost. Ask explicitly, it is often left out of proposals |
| Revenue uplift | Additional annual revenue attributable to the project | Optional. Set it to zero unless you can defend the causal link |
| Readiness | How prepared your data and team are | Drives how much of the theoretical saving you capture and how fast you capture it |
The Formula
Two calculations. The second one is the one that should drive the decision.
Step one: annual value
Annual value = (monthly hours × 12 × loaded hourly cost × automation rate) + revenue uplift.
This is the theoretical ceiling: what the work costs you today, multiplied by the share AI takes over. It is the number most vendor calculators stop at, and on its own it is close to useless, because it assumes perfect adoption from day one.
Step two: apply the three discounts
The adoption haircut accounts for the fact that teams never capture the full automation rate. Edge cases, exceptions, and people who keep doing it the old way all take a share. Depending on how ready your data and team are, expect to keep somewhere between 55% and 90% of the theoretical value.
The first-year ramp accounts for the fact that a system delivered in month three cannot deliver twelve months of value in year one. Between integration, training and the period where people are still learning to trust it, first-year realization commonly lands between 50% and 80%.
The redeployment test is the one nobody writes down, and it is the most important. Freed hours are only savings if someone decides where they go: to higher-value work, to slower hiring, or to more output. If the time is quietly reabsorbed into the working day, the return exists on the spreadsheet and nowhere else.
Step three: ROI and payback
First-year ROI % = ((annual value × ramp) − (build + run)) ÷ (build + run) × 100.
Payback in months = build cost ÷ ((annual value − annual run cost) ÷ 12).
Use payback for the decision. A single ROI percentage compresses timing out of the picture, and timing is usually what determines whether a project gets funded and whether it survives contact with a budget review.
If a calculator only ever produces a large positive number, it is not a model, it is a brochure. An honest one will sometimes tell you the project does not pay back, which is the most valuable output it can give you.
A Worked Example You Can Check
A 50-person professional services firm automating document processing. Every line is arithmetic on the line above it.
| Step | Figure | Calculation |
|---|---|---|
| Time on the workflow | 200 hrs/month | Measured across the team |
| Loaded cost per hour | $55 | ~$75K salary plus overhead |
| Annual cost of that work | $132,000 | 200 × 12 × $55 |
| Automation target | 40% | Realistic for a repetitive workflow |
| Theoretical annual value | $52,800 | $132,000 × 0.40 |
| Adoption haircut | × 0.70 | Medium data and team readiness |
| Realistic annual value | $36,960 | $52,800 × 0.70 |
| Build cost | $35,000 | One scoped workflow |
| Annual run cost | $7,000 | 20% of build |
| Year-one ramp | × 0.60 | Medium readiness |
| Year-one value | $22,176 | $36,960 × 0.60 |
| Year-one net | −$19,824 | $22,176 − ($35,000 + $7,000) |
| First-year ROI | −47% | −$19,824 ÷ $42,000 |
| Monthly net (steady state) | $2,497 | ($36,960 − $7,000) ÷ 12 |
| Payback | ~14 months | $35,000 ÷ $2,497 |
| Three-year net | ~$40,100 | $36,960 × 2.6 − $21,000 − $35,000 |
Read the two ROI lines together. First-year ROI is negative 47% and the project is still a good investment, because it pays back in fourteen months and clears roughly $40,000 over three years. This is why payback beats a first-year percentage: judged on year one alone, a sound project looks like a failure.
The Four Ways These Calculations Go Wrong
In roughly the order we see them.
No run cost
The build is budgeted and the ongoing model, infrastructure and maintenance cost is not. This error compounds: it looks small in year one and dominates the three-year picture.
Full automation assumed
The model assumes the AI handles 100% of a workflow that has exceptions, edge cases and judgment calls. Real automation rates on first projects are typically a third to a half.
Hours counted as cash
Time saved is booked as a saving with no decision about where the hours go. Unless headcount, hiring plans or output actually change, the money never appears in the accounts.
No baseline captured
Nobody measured the workflow before the project started, so the result cannot be proven either way. Capture the baseline first: it takes an afternoon and it is impossible to recreate later.
Run Your Own Numbers
Our free AI ROI calculator applies the same conservative model used in this guide, including the adoption haircut and first-year ramp, and it will tell you plainly when the math does not work.
- Conservative and optimistic ranges, not a single flattering number
- Payback period, first-year net, and a three-year view
- An honest verdict when the project does not pay back
Calculating AI ROI, FAQ
Common questions about the math behind an AI business case.
Want someone to pressure-test your numbers?
We will look at your inputs, tell you where the assumptions are doing too much work, and give you a straight answer on whether the project is worth scoping.
Related Services
Explore our other technical consulting services
Tell us what you're building.
Bring us the problem you're solving. We'll tell you how we'd build it, what it takes, and how Simple Engineers can help your business scale its technology.
Get in Touch
hello@simpleengineers.com
We typically respond within 2-4 hours
Phone
+1 (888) 966-0773
Mon-Fri 9AM-6PM EST
Location
Global Remote Team
Serving clients worldwide
Response Time
Within 24 hours
Emergency support available
Why Partner with Simple Engineers?
- Senior engineers who build and ship, not just advise
- We stay on to run and grow what we build
- You own everything: your code, your cloud, your keys
- Scaling companies from startup to enterprise since 2016
Send Us a Message
Tell us about your project or goals and we'll get back to you within one business day.