Automate or Hire? The Break-Even Arithmetic
Using March 2026 federal employer cost data, a workflow costing about $2,600 to build and run for a year breaks even at roughly 1 hour 20 minutes of freed clerical work per week. Under an hour a week, doing it by hand stays cheaper, and the setup and maintenance time you pay in your own hours is 88% of the bill.
Automating beats hiring when the task is repetitive, high-volume, rule-based and stable. At a small US private-sector employer, using March 2026 federal compensation data, a workflow that costs roughly $2,600 to stand up and run for a year breaks even at about 1 hour 20 minutes of freed labor per week. Below that, hiring or simply doing it by hand is cheaper. Above it, automation pulls away fast.
The arithmetic is below, along with the part most comparisons leave out: the setup and maintenance hours you pay in your own time.
What an hour of employee time actually costs
The Bureau of Labor Statistics publishes Employer Costs for Employee Compensation, which measures total employer cost per hour worked — wages plus benefits, not just the wage. For March 2026, private industry workers averaged $46.60 per hour worked in total compensation: $32.60 in wages and salaries and $14.01 in benefits (BLS, Employer Costs for Employee Compensation, March 2026).
Two refinements matter for small businesses.
Size. Costs are much lower at small establishments. In private industry establishments with 1–49 workers, total compensation averaged $37.36 per hour worked — $27.68 in wages and $9.68 in benefits. At establishments with 500 or more workers, it was $68.03 (BLS ECEC Table 6, March 2026). If you run a 12-person company, the large-employer figure is not your number.
Occupation. The work most people automate first is clerical. Office and administrative support occupations in private industry averaged $36.42 per hour worked in total compensation — $25.00 in wages and $11.41 in benefits (BLS ECEC Table 4, March 2026).
Those two figures land in the same place, so we will use $37.36 per hour worked as the loaded cost of the labor being displaced.
Benefits are about 26% of total compensation at establishments with 1–49 workers ($9.68 of $37.36), which means the multiplier from wage to true cost is roughly 1.35×. For civilian workers overall, BLS breaks the $15.60 benefits figure into paid leave $3.72, supplemental pay $1.74, insurance $4.18, retirement and savings $2.56, and legally required benefits $3.39 per hour worked.
One technical note that works in automation's favor: ECEC is measured per hour worked, not per hour paid. Paid leave is already spread across productive hours, so you do not need to gross the number up again for vacation.
What an automation actually costs
Three line items, and the subscription is the smallest.
Subscription. Mid-market workflow platforms are cheap by comparison. Make's Core plan is $12/month for 10,000 credits and n8n's Starter plan is €20/month billed annually for 2,500 executions, both verified September 2026 (Make pricing, n8n pricing). Call it $300 a year for one meaningful workflow, allowing headroom.
Build time. A genuinely useful workflow — trigger, filter, lookup, two or three destinations, error handling, testing against real data — is rarely under 15 hours of someone's attention and often 25 or more once edge cases surface. We will use 20 hours.
Maintenance. This is the line item people omit and then regret. APIs change, credentials expire, a vendor renames a field, someone edits the spreadsheet the workflow writes into. Budget 1.5 hours a month, or 18 hours a year, and treat that as a floor rather than a ceiling.
Build and maintenance hours are paid at your rate, not the displaced worker's — this work is usually done by an owner, an ops lead or a contractor. We will use $60 per hour, which you should replace with your own honest number.
The break-even formula
Annual cost of the labor you remove:
- Weekly hours removed × 52 × $37.36
Annual cost of the automation, year one:
- $300 subscription + (20 build hours × $60) + (18 maintenance hours × $60) = $300 + $1,200 + $1,080 = $2,580
Year two onward, the build cost disappears: $300 + $1,080 = $1,380.
Setting those equal gives the break-even. In year one: $2,580 ÷ (52 × $37.36) = 1.33 hours per week. In year two: $1,380 ÷ $1,942.72 = 0.71 hours per week, about 43 minutes.
Break-even at different volumes
| Hours freed per week | Annual hours | Annual loaded labor cost at $37.36/hr | Year-1 automation cost | Year-1 net |
|---|---|---|---|---|
| 0.5 | 26 | $971 | $2,580 | −$1,609 |
| 1 | 52 | $1,943 | $2,580 | −$637 |
| 2 | 104 | $3,885 | $2,580 | +$1,305 |
| 5 | 260 | $9,714 | $2,580 | +$7,134 |
| 10 | 520 | $19,427 | $2,580 | +$16,847 |
Change the inputs and the break-even moves, sometimes a lot. If your build takes 40 hours instead of 20, year-one break-even rises to about 1.9 hours a week. If you value your own time at $100/hour, it rises to about 2.1 hours a week. Run it with your numbers, not ours.
The subscription is 12% of the year-one cost. Your time is the other 88%. Choosing a platform on price is optimizing the smallest line item on the page.
When automating plainly isn't cheaper
- The task is under an hour a week. The table above is unsentimental about this. Doing it by hand is cheaper, and it stays cheaper.
- The rules change often. Every rule change is a rebuild. If the process changed three times last year, maintenance will not be 18 hours.
- Judgment is the actual work. If a human has to review the output anyway, you have automated the typing and kept the thinking, which is usually the expensive part.
- The source data is messy. Automating an unreliable input produces reliable garbage faster. Fix the input first; that fix is often the whole win.
- The volume is seasonal. A workflow that earns its keep for six weeks a year and needs maintenance for fifty-two rarely pays back.
- Nobody owns it. An automation with no named owner breaks quietly, and the failure cost lands on whoever discovers it.
The honest caveat about "hours saved"
Everything above assumes the freed hours are hours you genuinely stop paying for. Frequently they are not.
If you remove five hours a week from a salaried employee who stays salaried, the $9,714 in the table is not cash — it is capacity. That is real value only if the capacity gets used for something that produces revenue or prevents a hire you would otherwise have made. If those five hours turn into a slightly less pressured week, the automation still paid for itself in stress and error rates, but not in dollars, and you should not put it in a budget as if it were.
The clean cases where the savings are genuinely cash: work you currently pay a contractor or agency for by the hour; overtime you actually stop paying; a seasonal temp you do not rehire; a vacancy you decide not to backfill. Everything else is capacity, and capacity should be argued for on its own terms.
A short checklist before you decide
- Time the task for two weeks with a timer rather than estimating it. Recalled effort and measured effort are rarely the same number, and the business case is only as good as this input.
- Write down what percentage of cases the automation will actually handle end to end. If it is 70%, only count 70% of the hours.
- Use your establishment size band from the BLS data, not the national average.
- Price your own build hours at what you would pay someone else to do it.
- Ask whether the freed hours convert to cash or capacity, and say which in the business case.
Bottom line
For a small US employer, automation clears its cost at roughly one and a half hours of freed clerical work per week in year one and under an hour a week thereafter — but only if you count your own build and maintenance time at a real rate, and only if the workflow handles the cases without supervision. Under an hour a week, hire the hours or do it yourself. Over five hours a week on stable, rule-based work, the arithmetic stops being close.