Resources · For operators and finance · 5 min read
The math behind automation ROI
How to put a defensible number on repetitive work, why most estimates run low, and the recovery rate we use when we scope a build.
Every automation pitch ends with a big number. Most of those numbers are made up. This guide shows you the arithmetic we actually use to scope engagements, so you can run it yourself and trust what comes out.
The base formula
The annual cost of a repetitive workflow is four numbers multiplied together: the people who touch it, the hours each spends per week, their loaded hourly cost, and the working weeks in a year.
Three people, six hours a week each, at a loaded cost of $38 an hour, over 48 working weeks: that's 864 hours and roughly $32,800 a year. Not for the whole business. For one workflow.
Loaded cost, not wage
The most common error is using the wage. A $25-an-hour employee does not cost $25 an hour. Payroll taxes, benefits, equipment, software seats, and the management time that rides on every role push the real figure 30 to 50 percent higher. For most SMB admin roles the loaded cost lands between $30 and $55 an hour.
Owner time is its own category. Price it at what an hour of your selling or managing time produces, not at what you pay yourself. For most owners that number is uncomfortable, which is why the repetitive work an owner does personally is usually the most expensive leak in the building.
Why estimates run low
Ask someone how much of their week is repetitive and they'll report the blocks they notice: the Friday invoicing, the Monday scheduling. They won't report the interstitial work, because nobody experiences two minutes of copy-paste as a task.
When businesses actually track it for a week, the measured figure typically comes in at one and a half to two times the estimate. If you want the honest version, have two people tally for five working days. If you want the quick version, take your gut number and add half.
The recovery rate
Automation does not recover 100 percent of a workflow's cost, and anyone who models it that way is selling something. Edge cases stay manual. Someone still reviews output. The system itself needs occasional tending.
We model at a 35 percent recovery rate. Well-scoped automations routinely return 50 to 70 percent, but 35 is the floor we're comfortable putting in writing before we've walked the workflows. If the project only clears the bar at an 80 percent assumption, the project doesn't clear the bar.
What the formula can't see
Hours are only the first of four places leverage shows up. The formula says nothing about the lead that never got a reply, the renewal that lapsed unwatched, or the compliance date nobody owned. Those layers are harder to put on a slider and often worth more. We cover them in what an AI assessment actually finds.
For the hours layer, though, the arithmetic is enough to act on. The calculator runs this exact model with your numbers. If the annual figure it shows you is small, close the tab with a clear conscience. If it isn't, you now know what the problem costs, which is the only sane starting point for deciding what to spend fixing it.