Before you spend on automation, you should be able to defend the return in a single page. Here's a simple, honest way to estimate it — one that survives scrutiny because it doesn't over-claim.
Start with the hard, countable savings
Multiply it out:
(hours saved per week) × (fully-loaded hourly cost) × 52 = annual labour saved
“Fully-loaded” matters — include salary, benefits, and overhead, not just take-home pay. Be conservative on hours: count the time the task actually takes, not the worst week.
Add the cost of errors
Manual work has a defect rate, and defects cost money — reworked invoices, wrong shipments, compliance findings, refunds. Estimate the frequency and the average cost to fix one, and add the avoided total. This number is often larger than the labour saving and almost always ignored.
Account for the value you can't put on a spreadsheet
Some returns are real but harder to price — faster response times, work that now runs 24/7, staff freed for higher-value tasks, and the ability to grow without adding headcount. Don't inflate these, but don't pretend they're zero. List them qualitatively alongside the hard numbers.
Subtract the honest costs
Against the gains, put the build cost, any ongoing tooling or model usage, and maintenance. A fair business case shows the payback period — how many months until the savings cover the build.
A simple rule of thumb
If a process is frequent, manual, and error-prone, automation usually pays back within months, not years. If your estimate shows payback longer than a year for a simple workflow, either the process is more complex than it looks or it's the wrong first project.
Keep the model conservative and transparent. A modest, believable ROI you can defend beats an enormous one nobody trusts.