Did the automation pay for itself? Hours saved is the wrong answer
Six months in, the workflow runs, everyone likes it, and nobody can say whether it was worth the money.
The standard answer is hours saved times hourly rate. It produces a number, the number is always flattering, and in a business of one to twenty-five people it is usually not true. Worth understanding why before deciding whether the next one is worth building.
The hours were real. The money was not
Say a task took four hours a week and now takes twenty minutes. Nearly four hours a week, freed. At any plausible rate that is thousands a year.
Ask where those thousands went and the answer, most of the time, is nowhere. Nobody was let go. Nobody's salary changed. The owner who spent Friday afternoon on reconciliation now spends Friday afternoon on something else — probably something more useful, possibly something more pleasant, but not something that sent an invoice.
Saved time turns into money in exactly four ways, and it is worth being honest about which one applies:
You sold the time. The hours went into billable work, or into selling, and revenue moved. This is the only case where the multiplication is straightforwardly true — and it only holds while there is demand waiting to absorb the time.
You did not hire. You were about to take somebody on, or bring in a freelancer, and now you are not. Here the saving is real and dated: it is the salary or invoice that never started.
You stopped paying for mistakes. The duplicate order, the missed deadline, the refund, the late fee. Errors have prices and most businesses know theirs roughly.
Capacity you already had got used. More appointments in the same day, more orders shipped by the same people. The time did not become money directly; it raised the ceiling on what could be sold.
If none of the four applies, the automation may still have been an excellent idea — it can buy back your evenings or remove work that made people miserable, and both are legitimate reasons to spend money. Just do not put a currency symbol in front of it and call it a return.
The costs that do not appear in the quote
The other half of the arithmetic is usually incomplete in the same direction.
Subscriptions are the visible part, and they are not fixed. Zapier's public pricing, read in August 2026, charges by task volume: on the Professional plan billed annually, 750 tasks a month is $19.99, 1,500 is $39, 2,000 is $49 and 5,000 is $89. The Team plan starts at 2,000 tasks for $69 and reaches $249 at 20,000. Prices and structures change — check the vendor's page — but the shape is the point.
You are billed for operations, not for outcomes. An automation that processes every incoming form to catch the occasional important one costs proportionally more as volume grows while delivering the same benefit. Growth does not make that automation cheaper per useful event; it makes it more expensive per useful event.
Then the invisible costs. The hours somebody spent building and rebuilding it. The maintenance when a vendor changes something. The occasional half-day of untangling a run that went wrong. And the one nobody counts: the time spent checking that it worked, which is real and continues indefinitely if the system never tells you when it fails.
What to record, and when
The measurement problem is that the "before" number stops existing the moment you automate. Nobody remembers accurately how long the old way took, and estimates made afterwards drift in whichever direction supports the decision already made.
Write the before down before you build. One line: what the task is, how often it happens, roughly how long it takes, and who does it. Ten minutes now, and it is the only version of that number you will ever have.
Say which of the four returns you expect. In advance, in a sentence. "We expect not to hire the part-time person we were about to" is a claim you can check in six months. "It will save time" is not.
Note the error rate if errors are the point. How many duplicates last month, how many things missed. This is the most checkable number available and the one most often skipped.
Then look again at ninety days. Long enough for the novelty to wear off and for the first maintenance to have happened, short enough to abandon it cheaply.
The question at ninety days is not "does it work". It is: did the thing we predicted actually happen? If you expected to stop hiring and you hired anyway, the automation may still be worth keeping — but you have learned that your model of your own business was wrong, which is worth more than the workflow.
The ones that pay and the ones that do not
Patterns worth knowing before you commit money.
Automations that pay reliably tend to be frequent, boring and error-prone: the task done fifty times a week, where mistakes cost real money and the rules do not change. Fifteen minutes saved fifty times beats two hours saved once a month, and the second is where people usually start because it feels more painful.
Automations that disappoint tend to be rare, judgement-heavy or unstable. If the process changes every few months, you are not automating a process, you are subscribing to rebuilding one. If the automation needs a human to check every result, you have added a step rather than removed one.
And the quiet failure mode: an automation that works perfectly for a process that should not exist. Speeding up a report nobody reads returns exactly nothing, faster.
When the answer is worth more than a spreadsheet
For one or two workflows, the exercise above is a page in a notebook.
It gets harder when you have a dozen, built at different times by different people, some of which have been quietly failing for months, and the subscription line on your card is a single number covering all of them. Then "did it pay for itself" cannot be answered per automation at all, because nobody knows which ones are still doing anything.
Establishing that — what runs, what it costs, and what would break if you turned it off — is one of the things a process audit produces: $299, three business days, and a list you can actually make decisions from.
Sources
I read this page in August 2026. Pricing changes; check the vendor's page.
Task-based pricing tiers used in the arithmetic above: Zapier's pricing page.
Related: what automation costs a small business, and why quotes differ covers the other side of this — what you are paying for in the first place, rather than whether it came back.