How do you actually calculate the ROI of AI automation before you build it, instead of finding out six months after the invoice?
Most sales pages answer that question with a story, not a formula. This one gives you the formula, plus what the honest version of the ROI looks like once you compare a single scoped automation against the slower, murkier numbers primary research reports for broad enterprise AI rollouts.
TL;DR:
- Deloitte's 2025 executive survey found typical AI projects take 2 to 4 years to deliver satisfactory ROI, well past the 7 to 12 month payback businesses expect from a standard tech investment.
- Gartner predicts over 40 percent of agentic AI projects will be canceled by the end of 2027 due to escalating cost, unclear business value, or inadequate risk controls.
- Those numbers describe broad, company-wide AI transformation efforts, not a single scoped workflow automation, which is a different calculation with a much shorter payback.
- The formula needs five inputs: task frequency, time per task, fully loaded hourly cost, build cost, and monthly running cost.
What the pitch promises versus what the research finds
Scroll through any roundup of ai automation roi calculators and the implied timeline is weeks, not months. Plug in a few numbers, see a six-figure annual saving, sign the contract.
Deloitte surveyed 1,854 senior executives across 14 countries in August and September 2025 at organizations already running operational AI, and found the typical AI use case takes 2 to 4 years to deliver satisfactory returns, against a 7 to 12 month payback expectation for ordinary technology investments. Only 6 percent of respondents achieved payback within a year. For agentic AI specifically, only 10 percent of organizations currently report realizing significant, measurable ROI.
Gartner's research points at the same gap from a different angle. The firm's June 2025 prediction that more than 40 percent of agentic AI projects will be scrapped by the end of 2027 names escalating costs and unclear business value as the leading causes, not model performance. The agent itself usually works. The business case underneath it never got measured honestly enough to survive a budget review.
Why the broad number and the scoped number are not the same bet
A company-wide AI rollout has to prove value across dozens of workflows simultaneously, and the benefit of any one of them gets diluted by the coordination cost of touching all the others. That is most of why Deloitte's and Gartner's numbers land where they do.
A single scoped automation skips that overhead entirely. It has one trigger, a handful of steps, and a result you can count in a spreadsheet by the end of the first month. Our AI automation team builds these at the scoped level deliberately, because that is the only version of "automate it" where the ROI math is actually knowable before you sign anything.
The practical move is to treat the two as separate purchases. Calculate the scoped automation's ROI on its own, prove it, then decide whether to expand.
The five inputs you need
Before any ROI number means anything, you need these five, specific to the one process you are evaluating, not an average across the business:
- Frequency - how many times this task happens per month.
- Time per occurrence - how many minutes a person spends on it today.
- Fully loaded hourly cost - wage plus payroll taxes and benefits for whoever does the task now, not just their base pay.
- Build cost - the one-time cost to design and ship the automation.
- Monthly running cost - the ongoing tool subscription plus any review or maintenance time it still needs.
Skip any one of these and the ROI number you get back is a guess dressed up as math. Frequency and time-per-occurrence in particular get rounded up by sales conversations; pull them from an actual log of call volume, ticket count, or form submissions instead of a gut estimate.
A worked example (illustrative, not a quote)
As an illustration of the formula in use, not a real client figure: a business handles 400 inbound support tickets a month, and a person currently spends 6 minutes per ticket on first-response triage, at a fully loaded cost of 28 dollars an hour.
That is 2,400 minutes, or 40 hours, of triage a month, worth 1,120 dollars at that hourly rate. If an automation cuts triage time to 1 minute per ticket by routing and drafting the first reply, the labor cost drops to roughly 187 dollars a month, a savings of about 933 dollars a month.
Against a hypothetical build cost of 3,500 dollars and a monthly running cost of 150 dollars, net monthly savings land around 783 dollars. Payback period is build cost divided by net monthly savings: 3,500 divided by 783, or about 4.5 months. That example math is the shape of the calculation, not a promise about any specific business's numbers, which depend entirely on actual ticket volume and current handling time.
The payback period formula
Two formulas cover the whole calculation:
- Monthly net savings = (frequency x time per occurrence x fully loaded hourly cost) minus monthly running cost.
- Payback period in months = build cost divided by monthly net savings.
Run both before you commit to a build, not after. As a rule of thumb, if the payback period comes out past 12 months for a single scoped workflow, that is a signal the process either does not happen often enough to justify automating yet, or the build is scoped too broadly and should be split into a smaller first phase.
Where the number breaks down after launch
A payback calculation done at launch assumes the automation keeps performing at that level indefinitely. It does not, automatically. Edge cases get discovered, source systems change their data format, and a workflow that was 95 percent reliable at launch can quietly drift without anyone noticing until a client complains.
That is also why "automate everything with AI" rarely produces the ROI a single scoped build does. Our breakdown of AI automation for small business covers the build order that avoids this failure mode: rule-based automation first, AI layered in only where a judgment call is genuinely required. If the process you are evaluating touches multiple systems with no AI judgment involved at all, a plain workflow automation build is often the faster, cheaper path to the same payback period, without carrying an AI model's ongoing review cost.
Budget for a monthly check against the original five inputs. As a rule of thumb, if frequency, time saved, or running cost drift by more than 20 percent from what you calculated at launch, the payback period you quoted internally is no longer accurate, and it is worth re-running the math before anyone asks.
Before you build anything
- Pull real frequency and time-per-occurrence numbers from a log, not a guess, before calculating anything.
- Calculate the scoped automation's payback on its own; do not let a company-wide AI rollout's slower numbers talk you out of a fast, narrow build.
- Re-check the math monthly for the first quarter after launch, since drift in volume or handling time changes the payback period you originally quoted.
Tell our AI automation team the one process you are evaluating, with its actual monthly volume, and we'll run the payback math with you before any build starts.
