How to Calculate ROI on AI Automation
Every automation pitch promises savings. Few show the arithmetic. This is the simple model we build with clients before they commit, and it often talks them out of the wrong project.
- Published
The three value sources
- Time saved: hours per task × tasks per month × loaded hourly cost of the person doing it today.
- Errors avoided: error rate today × cost per error (rework, refunds, penalties) × volume.
- Revenue recovered: leads answered after hours, faster quotes, fewer missed calls.
Then subtract the build cost, the running cost (LLM usage, hosting) and the human review time the automation still needs.
A worked example
An accounts team processes 3,000 supplier invoices a month. Each takes about 6 minutes of manual entry and checking — 300 hours a month. Automated extraction with human review of exceptions cuts that to about 1.5 minutes per invoice on average.
| Line | Monthly |
|---|---|
| Hours saved | ~225 hours |
| Value at loaded cost | Hours saved × your hourly cost |
| Errors avoided | Fewer duplicate and mis-keyed payments |
| Running cost | Small per-invoice model and hosting cost |
| Build cost (one-time) | $12k – $25k for a single-process automation |
At typical loaded costs, projects like this pay back within a few months. If your volume were 300 invoices instead of 3,000, the same build might take years to pay back — and we would tell you not to do it.
Keep the model honest
- Measure today’s baseline for two weeks before building — guesses are always optimistic.
- Count only hours that turn into something: reduced overtime, avoided hires, or redeployed work.
- Include the review time — automation at 95% accuracy still needs someone on the other 5%.
Frequently asked questions
What payback period is good?
Under 12 months is a strong case. Under 6 is excellent. Over 24, look for a cheaper approach or a bigger process.
What if savings are hard to measure?
Pick a proxy you already track — tickets closed per person, days to invoice, calls answered — and baseline it before launch.
Can you help build the business case?
Yes. We build the ROI model with you in discovery, before any build commitment.
