Short answer
An AI ROI calculator estimates whether an AI project pays for itself by comparing the staff time a task takes today with the time it takes with AI, including review and the cases AI cannot handle, then subtracting running costs. Enter your own volumes and NZ$ costs below to get hours saved per month, net monthly benefit, payback period and first-year return.
- Review time and the share of tasks AI handles move the result more than model prices.
- Exceptions still take today’s time, plus extra minutes for the failed AI attempt.
- If one plausible change flips the answer, run a pilot before building.
- Nothing you enter leaves your browser.
Use this calculator for a single workflow: one kind of task that happens often, such as answering a type of customer email, coding supplier invoices or summarising inspection reports. It works in NZ dollars and runs entirely in your browser. Start with the default numbers to see how it behaves, then replace them with your own.
AI ROI calculator
Estimates only. Exceptions are the share of tasks AI cannot handle, which still take today's time plus the extra minutes above. Nothing you enter leaves your browser.
What each input means
| Input | What to enter | Where to get it | Default |
|---|---|---|---|
| Tasks per month | How many times the task happens in a normal month | Your inbox, ticketing system, Xero or job system counts | 800 |
| Loaded cost per hour (NZ$) | Full cost of an hour of the people doing it | Payroll and overheads, divided by hours worked | NZ$65 |
| Minutes per task today | Average time from start to finish, now | Time 20 real tasks; do not estimate | 12 |
| Minutes per task with AI, including review | Time for a person to check and finish an AI-handled task | A pilot, or timed tests on real examples | 4 |
| Share of tasks AI can handle (%) | Tasks where the AI output is usable after review | Test on a mixed sample, not easy cases | 70% |
| Extra minutes per exception | Time lost on the failed AI attempt before doing it the old way | A pilot; start with 3 to 5 | 3 |
| One-off build cost (NZ$) | Setup, integration, testing, training, staff time | Quotes plus your own staff hours | NZ$30,000 |
| Monthly running cost (NZ$) | Model usage, licences, hosting, monitoring, support | Quotes, plus the AI cost calculator | NZ$600 |
How the calculation works
In plain words: the calculator works out total minutes spent on the task today, total minutes with AI, and converts the difference into hours and dollars.
- Minutes today = tasks × minutes per task today.
- Minutes with AI = (tasks × share handled × minutes with AI) + (tasks × share not handled × (minutes today + extra minutes per exception)).
- Hours saved per month = (minutes today − minutes with AI) ÷ 60.
- Net monthly benefit = hours saved × loaded cost per hour − monthly running cost.
- Payback period = one-off build cost ÷ net monthly benefit, in months. If the net benefit is zero or negative, payback is “Never”.
- First-year return = 12 × net monthly benefit − one-off build cost.
Worked example with the default numbers
Illustrative example: a team handles 800 tasks a month at 12 minutes each, which is 9,600 minutes. With AI, 560 tasks (70%) take 4 minutes including review, which is 2,240 minutes. The other 240 tasks take 12 + 3 = 15 minutes, which is 3,600 minutes. Total with AI: 5,840 minutes.
- Hours saved: (9,600 − 5,840) ÷ 60 = 62.7 hours a month
- Gross benefit: 62.7 × NZ$65 = NZ$4,073
- Net monthly benefit: NZ$4,073 − NZ$600 = NZ$3,473
- Payback: NZ$30,000 ÷ NZ$3,473 = 8.6 months
- First-year return: 12 × NZ$3,473 − NZ$30,000 = NZ$11,680
Which inputs move the answer most
Change one input at a time from the defaults and watch the payback.
| Change from default | Hours saved | Net monthly | Payback | First-year return |
|---|---|---|---|---|
| None (defaults) | 62.7 | NZ$3,473 | 8.6 months | NZ$11,680 |
| AI handles 50%, not 70% | 33.3 | NZ$1,567 | 19.1 months | −NZ$11,200 |
| AI handles 90% | 92.0 | NZ$5,380 | 5.6 months | NZ$34,560 |
| Review takes 7 minutes, not 4 | 34.7 | NZ$1,653 | 18.1 months | −NZ$10,160 |
| Exceptions cost 10 minutes, not 3 | 34.7 | NZ$1,653 | 18.1 months | −NZ$10,160 |
| 400 tasks a month, not 800 | 31.3 | NZ$1,437 | 20.9 months | −NZ$12,760 |
| Running cost NZ$1,500, not NZ$600 | 62.7 | NZ$2,573 | 11.7 months | NZ$880 |
Three lessons from the table. Review time is easy to underestimate, and three extra minutes of checking halves the benefit. The share AI can handle decides whether the project works at all, so measure it on a realistic mix of cases. And low volume is hard to rescue: a workflow at a few hundred tasks a month rarely justifies a custom build, though it may justify a subscription tool.
How to get inputs you can defend
- Time 20 real tasks. Spread them across a normal week and include awkward ones. Our lesson on calculating time saved by AI shows a simple timing sheet.
- Test the AI on 30 to 50 real examples. Count how many outputs are usable after review. That count is your share handled.
- Time the review, not the AI. The model takes seconds; the person checking it takes minutes.
- Run a short pilot before a full build. MBIE’s quickstart for businesses suggests starting small and trialling systems before committing (MBIE). Track results with the AI pilot scorecard.
- Re-run the calculator with pilot numbers. Then write it up with the AI business case template.
Common mistakes and when the estimate is unreliable
- The task varies a lot. If some tasks take 2 minutes and others 40, an average hides the cases AI handles badly. Split the workflow into task types and run each separately.
- Saved hours do not become money. The calculator values every saved hour at full cost. That is only true if the time is used for other paid work, avoided overtime or a hire you no longer need.
- The first months are slower. People learn the new process and prompts get fixed. Expect lower savings in months one and two.
- Quality changes are missing. Fewer errors or faster replies may matter more than time. Measure them separately rather than forcing them into hours.
- Running costs scale with volume. Model usage grows with tasks, so a busier year costs more to run. Our guide to what custom AI projects cost covers the other running costs.
- Inputs came from a demo. Vendor demos use clean examples. Only numbers from your own work count.
Next step
If the payback holds up across the sensitivity cases, write it into the AI business case template. If it only works in the best case, scope a narrow pilot first; our AI workflow automation work starts the same way, with real tasks timed before anything is built.
Sources
- Quickstart: Businesses using AI tools, Ministry of Business, Innovation and Employment, accessed 19 October 2026
Written by the Levy Street team in Wellington. Last updated 19 October 2026.