Free AI ROI Calculator

How the AI ROI Calculator Works

The formulas, in full

1. Current annual cost. Team size × hours per week × hourly rate × 52. Use a fully loaded hourly cost — salary plus benefits, payroll tax, and overhead — not the salary line, or you will understate the baseline by roughly a third.
2. Gross annual saving. That cost × an automation percentage for your industry, between 45% and 65% (table below). This is the share of the hours you named that a competent automation can realistically take over — not the share of the job.
3. Annual cost of the automation. $2,400 for platform and model access, plus $600 per person per year for per-seat tooling, plus one-off build cost of $1,500 per person spread over three years. For a team of ten that is $2,400 + $6,000 + $5,000 = $13,400 a year.
4. Net saving and ROI. Gross saving minus that cost. ROI is the net saving as a percentage of the cost, so 100% means the automation returned its own cost once over. The five-year figure compounds the annual net at 5% a year, on the assumption that both wages and the scope of what you automate drift upward.

Most automation calculators stop after step two, which is why their numbers look so good. Showing the cost side is what makes the ROI figure mean anything — and for small teams doing few manual hours, it is also what turns an apparently attractive number negative.

Automation assumptions by industry

These percentages are our own planning assumptions, not measured results. They encode a simple observation: the more of a sector's routine work is already digital and rule-shaped, the more of it automates cleanly. Treat them as a starting point and override them with what you know about your own processes — if your work is judgement-heavy or your data lives in people's heads, the real figure is lower.

Where this estimate will be wrong

A model this simple is useful for deciding whether a project is worth scoping properly, and not for signing a budget. Four things it does not capture, all of which move the answer:

• Automation is rarely all-or-nothing. Most workflows end up part automated with a human reviewing edge cases, so the hours saved are real but the headcount does not change. Savings show up as capacity, not as payroll, unless you decide otherwise.
• Build cost varies enormously. $1,500 per person is a reasonable figure for wiring up existing tools. A bespoke integration against a legacy system can be several times that, and the model has no way to know which you are facing.
• Maintenance is ongoing. Models get deprecated, APIs change, and an automation nobody owns quietly degrades. The per-seat figure covers tooling, not the attention of whoever keeps it working.
• The first project is the expensive one. Early automations carry the cost of learning. Later ones reuse it, which is why the five-year view flatters a programme and the first-year view flatters nothing.

If the net figure comes out marginal, that is information — it usually means the hours are too few or too scattered to be worth automating yet. The AI readiness assessment is the better next step in that case: it scores whether the groundwork is in place before the money question is worth asking.