Ninety days is the bar I use for building something without further argument. The maths is four numbers: hours a week times 46 working weeks, times what the hour costs you, times the share of the task that disappears. Divide the build price by one month of that saving. Three months or under, build it.
This is the method behind the ROI calculator on this site, written out so you can run it on the back of an invoice and see where it can be pushed around.
The four numbers, and what the calculator does with them
The calculator asks for four things and multiplies them in a fixed order. There is no cleverness in it, which is the point.
- Hours a week spent on the task, across everyone who touches it.
- What that hour costs you. Wages plus on-costs, so ACC, KiwiSaver, the phone, the ute. Not your charge-out rate.
- The share that goes away. The calculator offers 90, 70, 50 and 30 per cent, and defaults to 70.
- The build price, plus an optional monthly running cost for licences or hosting.
From those it works out annual hours as the weekly hours times 46. It multiplies annual hours by the hourly cost to get what the task costs you today. It multiplies that by the share removed, then subtracts twelve months of running cost, and that is the annual saving. Payback in months is the build price divided by one twelfth of the annual saving. Three-year net is three years of saving less the build price.
One consequence falls straight out of the arithmetic. To pay back inside 90 days, a build has to save a third of its own price every month.
Why 46 weeks and not 52
Nobody works 52 weeks. Take out four weeks of annual leave, the public holidays and the run of days around Christmas when the job does not happen, and about 46 working weeks is what is left. Using 52 inflates the annual hours by roughly 13 per cent, and everything downstream of that number inherits the error. The calculator states the convention on the results panel so the number can be checked rather than trusted.
Why 90 days is the bar
This is my judgement, not a benchmark anyone publishes. A build that pays for itself inside a quarter survives a bad quarter. If work goes quiet, a client pays late, or the year turns out worse than the one you planned for, a build that has already returned its price is not a decision you regret. One that needs two more years of stable trading to break even is a bet on the trading, on top of a bet on the software.
It is a demanding bar, deliberately. Most worthwhile automations do not clear it, and that is fine. The calculator's own verdict says anything inside twelve months is where most of these are worth doing. Treat 90 days as the line where the decision stops needing a conversation: under three months, build it; three to twelve months, build it if the process is stable; past two years, leave it alone.
A worked example, all of it
Say the job is quotes and invoices being retyped between the job tool and Xero. Two people touch it, twelve hours a week between them, which sits inside the 10 to 15 hours a week of leftover admin I describe in what custom automation actually costs a Canterbury trades business. The hour costs $45, the calculator's default. Seventy per cent of it goes away, because someone still handles the exceptions. The build is quoted at $4,000, the floor of the single-automation band. No running cost, because it sits on software already paid for.
- Annual hours: 12 × 46 = 552 hours
- What it costs today: 552 × $45 = $24,840 a year
- Annual saving: $24,840 × 0.7 = $17,388 a year
- One month of saving: $17,388 ÷ 12 = $1,449
- Payback: $4,000 ÷ $1,449 = 2.8 months
- Three-year net: ($17,388 × 3) − $4,000 = $48,164
Two point eight months. That one passes, and the bar it needed to clear was $1,333 a month. Put the same figures into the calculator with those numbers loaded and it returns the same answer, because it is the same four multiplications.
How to run it by hand
Weekly hours times 46. Times the hourly cost. Times the share that disappears, as a decimal. Divide by twelve for a month of saving. Divide the build price by that. Five operations, and you can do it in the ute before you ring anyone.
The hours number is the one to get right, and the way to get it right is to stop guessing. Have whoever does the job write down the time for two weeks, including the chasing and the double-checking. Two weeks of a scrappy tally beats anyone's confident estimate.
How the test gets gamed
Both of the common failures flatter the build, which is what you would expect from a number produced by someone who has already decided.
Optimistic hours, optimistic share. Take the example above and put honest numbers in: eight hours a week rather than twelve, and half of it going away rather than seventy per cent. Annual hours drop to 368, the annual cost to $16,560, the saving to $8,280, and a month of saving to $690. Payback moves from 2.8 months to 5.8. Same build, same price, no longer a 90-day payback. That gap is entirely in two guesses.
The owner's time to review, left out completely. Nothing lands finished. For the first quarter someone has to read the output, catch the odd wrong line and tell the builder about it. Two hours a week for thirteen weeks is 26 hours, and at $45 that is $1,170 of real cost the calculator never sees. Add it to the $4,000 build and the effective price is $5,170, which against $1,449 a month is 3.6 months. The 90-day pass quietly becomes a fail.
A third one is worth naming: counting a saving twice because two automations both claim the same admin hour. If you are running the test on three candidate builds, make sure the hours in each are hours the others are not also removing.
When 90 days is the wrong bar
The test measures hours removed, so it is blind to anything that is not an hour. A build that stops quotes going out four days late is worth money that never shows up in this arithmetic. So is not carrying the job in your head at 10pm. If the reason for the build is risk, compliance, or a customer experience problem, run the maths for information and then decide on other grounds.
It is also the wrong bar under about two hours a week. At that size, no build gets used properly, and a thing that is not used saves nothing however good the arithmetic looked. Fix the process or buy a subscription. Where the choice is between building and paying for one more tool, the shape of that argument is in what automating invoicing costs in NZ, and the platform version of the same call is in when Power Platform is the wrong answer.
Run it on your numbers
Book a free 30-minute call and we will put your hours through it together. If the answer comes back past three years, I will say so on the call rather than send you a proposal.
Questions about the payback maths
What is the 90-day payback test?
It is a screening test for whether to build an automation now or later. Work out the annual saving: hours a week times 46 working weeks, times what the hour costs you in wages and on-costs, times the share of the task that disappears, less any running costs. Divide the build price by one month of that saving. Three months or less and the build is worth doing without further argument. This is Kove's own bar, not an industry benchmark.
Why does the ROI calculator use 46 weeks instead of 52?
Because nobody works 52 weeks. Fifty-two less four weeks of annual leave, the public holidays and the days around Christmas when the job does not happen leaves about 46 working weeks. Using 52 inflates the annual hours by roughly 13 per cent, and every figure downstream of it, so the calculator uses 46 and says so on the page.
How much does a build have to save to pay back in 90 days?
A third of its own price every month. A $4,000 build has to save about $1,333 a month, which is close to $16,000 a year. That is a demanding bar on purpose, and most worthwhile automations sit between three and twelve months rather than inside the quarter.
How do people get the payback maths wrong?
Two ways, and both flatter the build. The hours figure is guessed high because it is being used to justify a decision already made, and the share that disappears is set at 90 per cent when nothing goes to zero. The third mistake is leaving out the owner's own time to review the output in the first quarter, which is real cost that never appears in the calculator. Two hours a week for thirteen weeks at $45 is $1,170, and adding it to a $4,000 build pushes a 2.8-month payback out to 3.6.
Two builds that tend to clear the 90-day bar easily because the saving is a phone call removed every time: automatic rebooking for recurring services, and an automatic review request once the job's done.