How to calculate the return on investment of an AI agent

An AI agent is not an expense for the sake of being modern: it is an investment, which must bring in more than it costs. Here is a simple method to estimate it before signing, and check it afterwards.

Step 1 — Put a figure on the time spent today

For the work to be handed over, three figures:

• frequency: how many times a week?
• duration: how many minutes each time?
• the fully loaded hourly cost of the person doing it.

Example — customer reminders: 20 reminders a week, 15 minutes each, i.e. 5 hours a week. Over 46 working weeks, 230 hours a year. At a loaded cost of €35 an hour (assumption), that is about €8,000 a year.

Step 2 — Estimate what the agent takes over

An agent does not do 100% of the work: it handles routine cases, and hands over the others. Be cautious:

• the agent prepares, a person reviews and approves: some human time remains;
• special cases go back to a person;
• the first period requires adjustments.

Example (continued): if the agent takes over two thirds of the reminder time, the gain is about 150 hours a year, i.e. about €5,000.

Step 3 — Compare with the total cost

The cost of an agent is made up of:

1. set-up, paid once;
2. running costs: AI model usage, hosting, licences;
3. monitoring, if you choose it, monthly.

The calculation: annual gain − annual running and monitoring costs gives the net gain per year; set-up cost ÷ monthly net gain gives the number of months to pay it back.

Gains that are not time

Some agents pay off in other ways:

• cash flow: regular, graduated reminders shorten payment times;
• revenue: answering a request in a few minutes rather than two days changes the conversion rate;
• quality: fewer data-entry errors, fewer duplicates, fewer oversights;
• risk: a contractual deadline that is no longer missed.

These can be measured too, provided you have the starting figure.

The rule that keeps the calculation honest

Agree on the indicator before signing, and record it before starting. Without a measured starting point, no gain can be demonstrated: all that is left are impressions. We set two indicators at most per project — for example the time spent and the processing time —, recorded before going live then thirty days after.

It is also what protects the buyer: with us, the last payment instalment depends on the final approval; if the agreed indicators are not met, it is not invoiced.

Running the calculation on your case

The free Flash Diagnostic quantifies 3 to 5 leads in time saved, with an order of magnitude of the budget. If you want to compare several processes, with figures to back it up — gains, costs, return on investment —, that is what the Business Process Audit is for. For prices: How much does an AI agent cost?

Frequently asked questions

Which hourly cost should we use?

The fully loaded cost of the person doing the work today: gross salary plus employer contributions, divided by the hours actually worked. Without a precise figure, take a low estimate: the calculation will stay cautious.

How quickly does an AI agent pay for itself?

It depends on the volume handled. A task that takes several hours a week often pays for itself within a few months. A rare task never does: it is not a good candidate.

Is saving time enough to justify a project?

Not always. Some gains are not time: more regular reminders improve cash flow, a faster answer converts more requests. They must be measured too.