Software ROI, Calculated Properly: Payback Period, Three-Year Cost of Ownership, and the Hours That Do Not Count
Every software purchase gets justified the same way: the list price on one side, "hours saved" on the other, and a percentage that looks too good to question. The percentage is usually wrong in both directions — the cost side is missing half its lines, and the benefit side is counting hours nobody will actually get back. This is the arithmetic done with every line in it, for a subscription and for a build, so the number you take to the budget meeting survives contact with the finance person.
Quick answer
Software ROI is (benefit − total cost) ÷ total cost over a fixed period, and payback is total cost ÷ monthly benefit. A $500-a-month tool with a $2,000 setup fee and 20 hours of implementation costs about $9,200 in year one. If it saves two people five hours a week at a $60 loaded hourly cost, the benefit is $31,200 a year: ROI of 239% and payback in 3.5 months. Apply the usual 50% haircut to hours saved and it is still 70% and 7 months. A three-engineer-month build of the same thing costs $90,000 to own over three years against $20,000 to buy, and only wins once the vendor's price passes about $2,450 a month.
The cost side has five lines, not one
The subscription price is the line everyone counts. The others are the ones that turn a "$6,000 a year" tool into a $9,200 first year.
| Cost line — a $500/month tool, year one | Amount | Where it hides |
|---|---|---|
| Subscription, 12 months | $6,000 | The invoice |
| Setup or onboarding fee | $2,000 | The order form, page 3 |
| Implementation — 20 hours at $60 loaded | $1,200 | Nobody's timesheet |
| Training — 4 people × 2 hours at $60 | $480 | Same |
| Integration upkeep, data cleanup, renewals | ~$500 | Next year's surprise |
| Year-one total cost | $10,180 | |
| Years two and three, each | ~$6,500 | Subscription plus upkeep |
Call it $9,200 counting only the subscription, setup and implementation, or $10,200 with training and upkeep; the point is the gap between the invoice and the cost. "Loaded" hourly cost — salary plus benefits, payroll tax and overhead, usually salary × 1.3 — is the right rate for every hour in this table, including the implementer's. An $85,000 salary is about $53 an hour loaded; $100,000 is about $62 — and a fully loaded engineer is about $15,000 a month, which is the default the build vs buy engine starts from.
The benefit side is where the lying happens
"Saves each person five hours a week" is the sentence every vendor sells and every buyer repeats. Two people, five hours, fifty-two weeks, $60 loaded: $31,200 a year. The formula is right. The question is whether the hours are real.
An hour saved is only worth $60 if one of three things happens to it: the person does billable or revenue-producing work in it, the company needs one fewer hire because of it, or the overtime it replaced stops being paid. An hour that becomes a longer lunch, a slower afternoon, or a meeting that expands to fill it is worth nothing to the business — it is still a real hour, but it is not $60. The honest practice is to haircut time savings by 50% unless you can name the redeployment, and by 100% for hours under about two a week per person, which disappear into the day.
| Benefit assumption | Annual benefit | Year-one ROI | Payback |
|---|---|---|---|
| Vendor's claim: 10 hours/week saved, all redeployed | $31,200 | 239% | 3.5 months |
| Honest default: 50% of hours redeployed | $15,600 | 70% | 7 months |
| Only one person's hours are real | $7,800 | −15% | 14 months |
| Replaces a $45,000 contractor outright | $45,000 | 389% | 2.4 months |
The last row is the strongest business case in software — a tool that removes a named, paid cost — and it needs no haircut because the saving is a cancelled invoice, not a theory about hours. The third row is the one to show the budget meeting first, because if the case survives it, it survives anything.
Payback, ROI and NPV — which number for which decision
Payback period answers "how long until this has paid for itself". It is the right number for a startup or any business where cash is the constraint, because a 400% ROI that pays back in 18 months can still kill you in month 9. Under six months is strong; under twelve is normal; past eighteen needs a strategic reason. For a company with investors, every month of software spend is burn — the burn multiple engine shows what a $6,000 line does to the ratio investors check before growth.
ROI answers "how much do I get back per dollar" over a fixed period. State the period: 239% in year one becomes a different number over three years once setup is amortised and the subscription keeps running. It is the right number for ranking several candidate purchases against each other.
Net present value discounts future benefits to today's dollars. It matters for a five-year commitment at a 10% discount rate, where year five's benefit is worth 62% of year one's; it is theatre for a month-to-month subscription. Most software decisions do not need it.
A tool that changes revenue rather than cost — a checkout that converts better, a pricing tool that lifts margin — is a different calculation: the benefit is the margin on the extra revenue, not the revenue, and the right engine for it is the breakeven calculator, which shows how many extra units the margin change needs to cover the cost.
Where a build beats a subscription
The same arithmetic applies to building it yourselves, with one line that never makes the engineering estimate: upkeep. Take the defaults our build vs buy engine carries — three engineer-months to build at $15,000 a fully loaded engineer-month, one engineer-month a year of upkeep — against the $500-a-month tool.
| Three-year cost of ownership | Build | Buy |
|---|---|---|
| Up-front | $45,000 (3 engineer-months) | $2,000 setup |
| Running cost, per year | $15,000 (1 engineer-month of upkeep) | $6,000 |
| Three-year total | $90,000 | $20,000 |
| Breakeven vendor price | — | about $2,450 a month |
The build costs 4.5 times as much to own, and the $45,000 estimate was only half of it. The answer flips when the vendor's price passes about $2,450 a month — three years of that subscription costs what the build does — or when the thing being built is the product, in which case the comparison is not about cost at all. At $4,000 a month the identical build is $56,000 cheaper over three years and has paid for itself by month 16. The engine takes your own engineer cost, estimate, upkeep and vendor quote and finds the crossover; the companion post walks through the two cases where building still wins.
The three mistakes that inflate every software ROI
Counting list price, not loaded cost. The implementation hours, the training hours and the renewal that goes up 8% in year two are all cost. Year one is routinely 50–70% above the invoice.
Counting hours that are not redeployed. Five hours a week is $15,600 a year only if those hours become something. Name the something, or halve the number.
Counting the benefit forever and the cost for a year. A subscription runs until someone cancels it, and nobody does. Compare three years of cost to three years of benefit, and put a review date in the calendar for the month the payback period ends — that is the one moment the decision gets re-opened honestly.
Frequently asked questions
What is a good ROI for software? There is no universal number, because the period matters. Over one year, anything positive after honest haircuts is defensible; over three years, well-chosen tools commonly return 100–300% on the figures above. A claimed 1,000% almost always means the hours were not haircut.
What payback period should I expect? Under six months is strong, under twelve is ordinary, and past eighteen the purchase needs a reason beyond the arithmetic — a compliance requirement, a strategic capability, a customer who demands it.
How do I put a dollar value on time saved? Loaded hourly cost — salary × 1.3, divided by about 2,080 hours — times the hours, times the share of those hours that become billable work, an avoided hire or cancelled overtime. If you cannot name which of the three, use 50%.
Should my own time count in the cost? Yes, at your loaded rate. An owner spending 20 hours implementing a tool has spent $1,200 to $3,000 of the most expensive labour in the company. Leaving it out is the most common reason small-business software "pays back in a month" on paper and never does in practice.
How do I calculate ROI on a tool that makes money rather than saving it? Benefit is the gross margin on the extra revenue, not the revenue. A checkout tool that adds $50,000 of sales at a 40% margin is worth $20,000 a year — and the pipeline you expect it to feed should be modelled, not assumed, which is what the ARR forecast engine is for.
Does it matter whether the tool is paid monthly or annually? For ROI, no — use the annual figure. For payback and cash, yes: an annual upfront payment moves the whole year's cost to month one, which lengthens payback and is why a startup on a tight runway should resist the 15% annual discount until the tool has proven itself month to month.
Sources and as-of dates. Build-side figures ($15,000 engineer-month, three-month build, one engineer-month a year of upkeep, $500/month and $2,000 setup to buy) are the published defaults of the Investor Sam build vs buy engine, October 2026, and the three-year totals, the $2,450 crossover and the $4,000-a-month figures are its output. Loaded-cost multiplier (salary × 1.3) and the 2,080-hour year are standard cost-accounting conventions. The subscription example and the hours-saved scenarios are hypothetical, chosen to show the shape of the calculation.
This article is educational and illustrative. It is not financial or purchasing advice; your own loaded costs, hours and vendor quotes decide the answer.