日本語DeutschFrançaisEspañolالعربية
Book a call
Home/Free tools/AI Agent ROI Calculator
Free tool · AI Agent ROI Calculator

What would one automated workflow save you?

Enter the hours your team spends on one workflow, what the agent costs to build and to run, and how long adoption takes. See the payback month and the three-year net, with every step of the math on the page.

40 hrs/wk
Total team hours spent on this one process, weekly.
$75 /hr
Salary + benefits + overhead, per hour.
$40,000
The one-off quote you are testing. Gigabit Agents run from $8,000 to $150,000, flat.
$3,000 /mo
Model usage, monitoring and upkeep. Managed AI Operations starts at $3,000/mo; lower it if your own team runs the agent.
3 months
Savings climb in a straight line until then. Most tools skip this and overstate year one.
Estimated annual savings, before costs
$78,000 – $109,200
40 hrs/wk × $75/hr × 52 wks = $156,000 spent per year
× 50–70% automatable = $78,000 – $109,200 back per year
midpoint $93,600 − running cost $36,000 = $57,600 net per year at full adoption
Payback on your build
10 months
Net after year 1
$9,800
Net after 3 years
$125,000
EngagementInvestmentPaybackNet, year 1
AI Quick Win Sprint$8,0004 months$41,800
AI Transformation Sprint$25,0007 months$24,800
Your build quote$40,00010 months$9,800

Get this as a one-page ROI memo.

The inputs, the savings range, payback and the three-year net, formatted to take to the board.

Turn the number into a pilot
How the math works

Five steps, nothing hidden.

The same arithmetic the calculator runs, so you can check it or rebuild it in a spreadsheet.

1 · What the workflow costs

Hours × hourly cost × 52

Total team hours spent on the one process each week, times the fully-loaded hourly cost (salary, benefits and overhead), times 52 weeks.

2 · What automation gives back

50–70% of that cost

The calculator assumes between half and seven-tenths of the work can be automated and shows both ends as a range. Everything after this step uses the midpoint, 60%.

3 · What the agent costs to run

Minus the monthly running cost

Model usage, monitoring and upkeep are paid every month from launch, whether or not the team has adopted the agent yet. You set the figure.

4 · How fast the saving arrives

A straight-line ramp

In month one of a three-month ramp the team gets a third of the saving, in month two two-thirds, and the full amount from month three on.

5 · Payback

The month the total turns positive

The build price is paid up front. Each month adds that month’s ramped saving and subtracts the running cost. Payback is the first month the running total reaches zero; net after one and three years is the same total at month 12 and month 36.

It leaves out revenue upside and error reduction, so read the result as a conservative first pass. For the reasoning behind each input, see how to calculate ROI on an AI project before you build it. For real figures to test against, see what ten client projects cost.

The inputs

What to enter, and where the sliders start.

Five inputs. The defaults are a starting point, not a forecast: replace each with your own figure.

InputWhat it meansStarts atWhy
Hours per weekTotal team hours spent on the one workflow each week, across everyone who touches it.40 hrs/wkOne full-time person's week: a workflow big enough to be worth automating, small enough to scope in one build.
Fully-loaded hourly costSalary plus benefits, payroll tax and overhead, divided by hours actually worked. Higher than the base hourly wage.$75/hrA round middle figure. Use your own: it moves the result more than any other input.
Build priceThe one-off quote you are testing, from us or anyone else.$40,000Inside the Gigabit Agents range ($8,000–$150,000, flat). The table under the result also shows our two fixed-price sprints.
Running cost per monthModel usage, hosting, monitoring and upkeep, paid every month from launch.$3,000/moThe entry tier of Managed AI Operations, if we run it. If your own team runs the agent, enter model usage plus the engineering time it takes.
Months to full adoptionHow long until the team has fully switched to the new way of working.3 monthsTeams rarely switch in week one. A three-month straight-line ramp is a cautious middle; most calculators assume day one.

What it counts

Cost saved on one workflow: the hours given back, priced at what they cost you, minus what the agent costs to run, ramped in as the team adopts it, against the one-off build price.

What it leaves out

Revenue upside, faster turnaround and fewer errors. Freed hours as capacity rather than cash: unless someone’s role changes, the saving shows up as more work done, not lower payroll. The cost of change management and of your own team’s time during the build. The long version is in how to calculate AI ROI: the formula.

Worked examples

Three workflows, the arithmetic written out.

The inputs are illustrative assumptions, stated in full. The results are what the calculator returns for them; load any example into it and move the sliders from there.

Example 1

One finance workflow, on the $8,000 Quick Win Sprint

Invoice matching in a finance team: one clerk spends most of each afternoon reconciling supplier invoices against purchase orders.

Assumptions

  • 15 hours a week at a loaded $45/hr
  • AI Quick Win Sprint, $8,000 fixed
  • $250/mo of model usage, run by your own team (the Sprint includes 30 days of light support)
  • 2 months to full adoption

The math

15 hrs × $45 × 52 = $35,100 a year
× 50–70% = $17,550 – $24,570
midpoint $21,060 = $1,755 a month
− running $3,000 = $18,060 net a year at full adoption
Payback on $8,000
6 months
Net after year 1
$9,183
Net after 3 years
$45,303

A small workflow pays back on a small build because the running cost is low. Put the same workflow on a $3,000-a-month managed service and it never pays back: the monthly saving is under $1,800.

Example 2

A support agent, built for $8,000 and run by us

Customer support for a Shopify brand: three full-time agents answering order-status, returns and product questions.

Assumptions

  • 120 hours a week at a loaded $40/hr
  • Gigabit Agents build at the entry price, $8,000
  • Managed AI Operations, Light tier, $3,000/mo
  • 3 months to full adoption

The math

120 hrs × $40 × 52 = $249,600 a year
× 50–70% = $124,800 – $174,720
midpoint $149,760 = $12,480 a month
− running $36,000 = $113,760 net a year at full adoption
Payback on $8,000
3 months
Net after year 1
$93,280
Net after 3 years
$320,800

The calculator’s 60% midpoint is close to what our D2C support case measured: 64% of tickets resolved with no human. A full deployment like that one is a larger quote than the entry price, so put the real quote in the build field. If you are weighing this against a per-resolution product, see custom support agent vs Intercom Fin: the breakeven.

Example 3

An operations workflow, starting with the $25,000 Sprint

Order exceptions in a 200-person distributor: an operations team chasing missing data, holds and supplier changes across three systems.

Assumptions

  • 60 hours a week at a loaded $85/hr
  • AI Transformation Sprint, $25,000, ending in a deployed pilot
  • $3,000/mo to run it
  • 4 months to full adoption

The math

60 hrs × $85 × 52 = $265,200 a year
× 50–70% = $132,600 – $185,640
midpoint $159,120 = $13,260 a month
− running $36,000 = $123,120 net a year at full adoption
Payback on $25,000
5 months
Net after year 1
$78,230
Net after 3 years
$324,470

This treats the pilot as the system that goes live. If the Sprint ends with a scoped production build, add that quote to the build price; the Sprint fee credits toward a Gigabit Agents build.

Calculator FAQ

Questions about the numbers

How do you calculate the ROI of an AI agent?

Start with what the workflow costs today: weekly hours × fully-loaded hourly cost × 52. Estimate the share that can be automated; this calculator uses 50–70%. Subtract what the agent costs to run each month, allow for the months it takes the team to adopt it, and set the result against the one-off build price. Payback is the month the running total turns positive.

What does it cost to run an AI agent after it is built?

It depends on who runs it. If Gigabit operates it, Managed AI Operations is $3,000–$20,000/mo depending on how many systems are covered, starting at $3,000/mo for one. If your own team runs it, the cost is model usage plus the engineering time to monitor and maintain it. Enter whichever applies; the calculator subtracts it every month.

Why does the calculator say an agent never pays back?

Because at the numbers entered, the monthly running cost is higher than the monthly saving, so no build price can be recovered. That is a real result for small workflows. A larger workflow, a cheaper way to run the agent, or plain rules-based automation is the better answer there.

What does this calculator leave out?

Revenue upside, faster turnaround, fewer errors and the value of capacity freed for other work. It measures cost saved on one workflow, on a straight-line adoption ramp, so treat the result as a conservative first pass.

Where does the 50–70% automation range come from?

It is an assumption, shown on the page so you can judge it. Well-scoped workflows with clear rules and good data sit near the top; messy ones with many exceptions sit near the bottom or below. For a real figure to test against, our D2C support case measured 64% of tickets resolved with no human involvement.

Can I use the calculator to check another vendor's quote?

Yes. Put their build price in the build field and their monthly fee, or your own running cost, in the running cost field. The arithmetic does not depend on who builds the agent.

What is a good payback period for an AI project?

There is no universal number, but the worked examples on this page give a feel: a support agent on a large workflow paid back in 3 months, a finance workflow on an $8,000 sprint in 6, an operations workflow starting with a $25,000 sprint in 5. If your result is measured in years, the workflow is probably too small for the running cost, and rules-based automation may be the better tool.

Do I need to give an email to see the result?

No. The savings range, payback, net figures and the math appear on the page as you move the sliders. An email is only needed if you want the result sent to you as a one-page memo.