The Honest Math of AI Automation ROI
Most automation ROI claims are marketing. Here's the actual framework we use to decide which workflows are worth automating — and which will never pay back.
Every automation pitch includes a number like '80% time saved.' Almost none of them include the denominator: saved on what, for whom, at what build and maintenance cost? After building automation systems across dozens of companies, we use a simple model — and it kills about half the automation ideas that come to us. That's the point.
The four-variable model
Automation value = (hours saved × loaded hourly cost) + (errors prevented × cost per error) − (build cost) − (ongoing maintenance). The first term is what everyone calculates. The second is usually bigger than the first in document-heavy workflows. The fourth is what vendors hope you forget.
Where the model says yes
High-volume workflows with structured decisions and expensive errors: invoice processing, lead routing, intake triage, report assembly. These pay back in weeks because the volume multiplies small per-unit savings into real money.
Where the model says no
Low-frequency workflows, tasks requiring genuine judgment on every instance, and processes that change monthly. If the process isn't stable, you're automating a moving target — maintenance eats the return.
The step most teams skip
Baseline measurement. If you don't know what the workflow costs today — actual hours, actual error rates — you'll never prove the automation worked. We spend the first week of every engagement measuring before we build anything.