Home / Guides / Measuring AI agent ROI

Is the agent paying back? How to tell.

Most AI business cases are written once and never checked. Measuring return is simple if you decide the number before the build and compare like with like afterwards.

Short answer

Measure an AI agent's return by choosing one outcome metric before the build (for example hours of manual work per week, reply time or bookings), recording a baseline, and comparing the same metric 30 and 90 days after go-live. Return per month is the value of hours saved plus revenue you can honestly attribute, minus running costs such as maintenance, model usage and channel fees. Divide the build cost by that monthly figure to get the payback period.

Method

Five steps, in order.

BeforePick one metric

The thing the agent should change. One number, not a dashboard.

BeforeBaseline

Two to four weeks of the current numbers, from the system where possible.

Day 30First read

Same metric, same sources; check quality on a sample of the agent's work.

Day 90Real read

Seasonality and teething problems smoothed out; decide to extend, change or stop.

OngoingMonthly line

One line in the monthly report, so the number stays honest.

Worked example

Assumed numbers, real method.

LineAssumptionPer month
Manual work removed10 hours a week removed × €40 an hour, fully loaded≈ €1,733
Attributable revenueNot counted in this example, to stay conservative€0
MaintenanceRun & Care−€490
UsageModel and channel fees, assumed−€150
Net monthly value≈ €1,093
Payback€3,900 build ÷ €1,093≈ 3.6 months

Every number except Oravel's prices is an assumption for illustration. It is not a client result or a promise; replace it with your own baseline.

Common mistakes

Where ROI maths goes wrong.

01Counting time not freed

If the hours saved are spread as minutes across ten people, nothing changes on the payroll. Say what the time will be used for.

02Attributing every sale

A faster reply helps conversion; it does not create every booking that follows. Count only what the baseline did not have.

03Forgetting running costs

Maintenance, model usage and per-message fees are monthly. One published estimate (Searchlab, March 2026) puts typical SME agent running costs at €300–€1,500 a month.

04Ignoring quality

An agent that is fast but wrong costs money later. Check a sample of its work every week at the start.

Note

Not legal advice.

This guide is general information, checked against the sources listed, and not legal advice. Rules change; check the current text or ask a lawyer before you rely on it. Checked November 2026.

FAQ

Common questions.

What is a good payback period?

Many owners look for payback within a year. That is a business choice, not a rule; a lower-risk, smaller system is easier to pay back quickly.

Can revenue be counted?

Yes, when you can compare like with like: for example, the share of enquiries that became bookings before and after, on the same lead sources.

What if the target is missed?

With Oravel, the target is written into the plan. If it is missed, we keep working at no cost; if it is still not met after 90 days, the setup fee is refunded.

Do we need a BI tool to measure this?

No. Most of the time the CRM, helpdesk or a simple log of the agent's work is enough.

Free 30-minute audit

Want this in your business?

Tell us how you work today. We'll tell you honestly what AI can take off your team's plate, what it costs, and whether we are the right fit.