MMOOZZGG.com

AI department

AI KPIs: how to measure the results of AI in your company

5 min read

AI KPIs, the key performance indicators of AI, are counted in money: how much a solution saved or earned in a month and how much it costs. An example from the calculator on our home page: 5 employees on routine work with a salary of $950, AI takes 60% of their work, and savings are $3,220 a month after paying for support.

On this page · 5 sections

Metrics in money

An AI department has one main metric: money. It means savings on staff work, extra revenue from leads that nobody had time to answer before, and the cost of the solution.

Savings follow the calculator formula: employees x salary x 1.3 (payroll taxes) x the share of work AI takes, minus support of $480 a month.

For example: 5 employees, a salary of $950, a share of 60%. Savings are $3,220 a month. An AI agent team at a price of $4,999 pays back in 2 months, and $33,600 remains after the first year.

You can calculate it on your own numbers in the calculator on the home page. Use the actual salary, and estimate the share of routine by how much time staff spend on typical tasks.

If the share of routine is unknown, measure it for a week: staff note how much time typical tasks took.

Count growth separately: multiply the leads that used to be lost at night and on weekends by the conversion to a sale and the average order value. Do not mix this number with savings, so you can see where the money came from.

Process metrics

Money does not show in the first week, so each process has its own number. Write it down before launch and compare it with what was there before.

1. Response speed: how many minutes it takes a customer to get a reply to a request.

2. Share of tasks without people: how many requests the agent closed alone, without passing them to an employee.

3. Conversion: how many requests turned into a deal or a booking.

4. Quality: the share of answers that need no changes in a spot check by the process owner.

5. Use: how many employees work with AI every day. A solution that nobody uses brings no money.

For each number, write down three values: the level before launch, the target by the end of the pilot and who checks it. Without the level before launch, there is nothing to compare with.

Two or three such numbers are enough for one process. More metrics blur responsibility: the process owner should remember them without a report.

Cost of the solution

The cost of a solution includes the one-time rollout price, monthly support and model fees by the volume of work. If you count only the one-time price, after a year the solution will turn out to cost much more.

Our prices for the calculation: a pilot from $1,499 in 14 days, an AI agent team of 3–5 agents from $4,999 in 4–6 weeks, and support from $480 a month that you can cancel in any month.

The cost also includes your people’s time: the process owner checks the agent’s answers, and the person in charge of AI prepares the report. In the first weeks this takes more time, later less.

Divide the cost by the result: what one handled request, one booking or one closed document costs. Compare this number with the price of the same work done by an employee, taxes included.

Look at the model bill month by month. If requests grow, the bill grows too, and that growth must pay back through a better result.

Report to the top manager

The report to the top manager fits on one page: each solution’s metric before and after, savings and growth in money, cost for the period and a decision on each line.

The manager dashboard collects the daily numbers: leads, deals and money in one window every morning. Once a month, the person in charge of AI puts them into a report and proposes the next process.

For each process, the report has one of three decisions: expand, keep as is or switch off. It has no general words about digital change, only numbers from the metrics.

Keep each report with the past ones to see the trend. A solution that shows the same number three months in a row is not growing, and that is also a reason to talk.

When to switch a solution off

Switch a solution off if, by the end of the pilot, its metric has not reached the target written down before launch. Extending it for another month without a new reason means paying for hope.

Also switch off a solution that people stopped using: the share of employees who work with it every day falls month after month.

The third reason: one operation costs more with AI than with an employee, and the price does not fall. Then the money moves to the next process on the map.

Switching off does not mean throwing away. The rules written for the agent stay with the company and will help when the process or the market changes.

You can cancel our support in any month, so switching off does not bring a long contract with it. What happens when there is no metric from day one is covered separately.

Publications

Mike Mozg

Agentic AI use cases at AT&T: nearly 500 processes now run on agents

AT&T no longer uses AI only to help staff work faster: it is rebuilding whole business processes around agents. SiliconANGLE reported on 14 November 2026 that nearly 500 agentic AI use cases at AT&T are already in production. Here is how the company rebuilds them, where a human still checks the agent, and what other large companies can take from it.

News4 min
Contact us