Every Friday at 4pm, someone in accounting opens the invoice register, opens the billing system, and re-keys forty invoice totals from one into the other. Three hours, give or take. Nobody calls it a cost center, because the person doing it is salaried, and salaried hours are invisible.

Your CFO has a different name for it. A $40,000-a-year problem with no budget line.

An automation ROI calculation is three numbers: what the manual work costs per year, what the fix costs once, and how many months until the fix pays for itself. Hours per week times loaded hourly rate times fifty weeks, compared against the build cost. If payback lands under twelve months, the automation is usually a buy.

The copy-paste tax

Manual work feels free because it's already paid for. The salary runs whether your person spends Friday re-keying invoices or doing actual accounting. But the formula is merciless: hours per week × loaded hourly rate × 50 weeks. Loaded means salary plus benefits plus overhead, typically 1.3 to 1.5 times base pay.

The tax hides in aggregates. Three hours here, ninety minutes there, twenty minutes of "quick" report assembly every morning. One client itemized a single week and found 19 hours of pure glue work spread across four people. Nobody thought they had an automation problem. They had four small ones sharing a trench coat.

A quick aside on the loaded rate, because it's where the math usually goes soft. Use the real number: salary plus payroll taxes, benefits, and a share of overhead. For a $50k administrator that lands near $65 an hour; for a $80k analyst, closer to $100. Using base salary flatters the manual process by 30% or more, which is a polite way of saying it cooks the books against your own case.

A worked automation ROI calculation

Take the Friday invoice ritual. Say it's 12 hours a week across two people once you count the checking and the error-chasing. Loaded rate, $65 an hour. 12 × $65 × 50 = $39,000 a year. That's the annual cost of doing nothing, and it repeats forever.

The fix, a script that reads the register, validates totals, posts to the billing API, and logs everything, quotes at $15,000. Payback: $15,000 ÷ ($39,000 ÷ 12) ≈ 4.6 months. Year one: $39,000 saved minus $15,000 spent leaves $24,000, roughly a 160% return. Every year after that is close to pure margin.

Run the same math on your own Friday and you'll see why fixed-price FDE engagements are structured the way they are: the build cost is knowable up front, and the savings side of the ledger is sitting in your own payroll data.

One more example so the pattern sticks. A sales manager assembles the Monday pipeline report by hand: exports from the CRM, a pivot, pasted charts, forty-five minutes a week plus the monthly version. Call it 5.5 hours a month at a loaded $95, so roughly $6,300 a year. An auto-generated report emailed at 7am Monday costs maybe $4,000 to build. Payback inside eight months, and the report stops depending on one person's Sunday evening. Small numbers still compound.

The costs nobody puts in the model

The simple math understates the case, because manual re-keying has an error rate. The rule of thumb is 1–3% of hand-entered records carry a mistake. At forty invoices a week, that's a wrong invoice every week or two, and each one costs a correction cycle, a mildly annoyed customer, and occasionally a credit memo.

Then there's the rework loop, where errors found downstream cost multiples of errors caught at entry. And the morale line item nobody models: the person spending Friday as a human API knows it's below their pay grade, and eventually acts accordingly. Replacing them costs more than the re-keying did.

Opportunity cost is the quiet giant. Those twelve hours could be collections calls, vendor negotiations, or anything with a higher return than transcription. The full cost breakdown of an automation build looks even better when the freed hours get redeployed instead of absorbed.

When the math says no

Honesty keeps the model credible. If the task takes two hours a month, automating it is a hobby. A $10,000 fix for a $3,000-a-year problem never pays back, and the CFO running your numbers will notice. Say it first; it buys you the right to be believed on the real ones.

Also, don't automate a broken process to make it a faster broken process. If the invoice register is a swamp of special cases and tribal exceptions, fix the process first, then automate the survivor. Automating chaos just produces chaos on schedule, every Friday at 4pm.

There's a third no, quieter than the other two: the task that's about to disappear. If the whole workflow gets replaced by a system migration next quarter, automating this quarter's version of it buys you three months of savings and a new piece of technical debt. Check the roadmap before you run the math.

The one-page business case

CFOs approve one page, not decks. The template that works: current annual cost with the hours and rate shown, the one-time build cost, payback in months, year-one ROI, and a risk line describing what happens if the automation breaks. If you've scoped it right, that line reads "someone does it by hand for a day."

Two presentation tips earned the hard way. First, bring the person who does the manual work to the meeting and let them describe their Friday; thirty seconds of that beats any chart you can draw. Second, anchor the ask against the cost of doing nothing for one more year, not against the build price. $15,000 sounds like money. $15,000 versus $39,000 a year, forever, sounds like a decision.

Keep the arithmetic visible. A CFO who can check your multiplication in the meeting is a CFO who says yes in the meeting. If you need backup for the pitch itself, the worked forward deployed engineer ROI calculation starts with exactly this page.

Then do the fun part: find the most expensive Friday in your building and put a number on it.