MMOOZZGG.com

AI department

How to build an AI team: where to start and in what order

5 min read

How to build an AI team when nobody in the company is responsible for AI yet: start with one responsible person and one process, not with a new unit. Below is a three-month order, week by week: a review, an AI use case map, a pilot with a metric, training and a report to the chief executive.

On this page · 5 sections

Weeks 1–2: an owner and a review

The chief executive appoints a person responsible for AI. It can be an in-house AI director, a part-time outside AI director or a manager who has the time and the authority to change how departments work.

The owner gathers the heads of departments and asks three questions: where time and money go, which AI services the company has already bought and who really uses them. The answers are written down for each department.

Separately, they record how many people do routine work in each department and what their time costs with taxes. These numbers are needed to calculate the effect in money at the next stage.

In parallel, the IT department lists the systems: CRM, accounting, phone line, messengers. The lawyer marks where the company processes customers’ personal data. Without these two lists you cannot build the map.

If you need an outside view, start with a 30-minute review; it is free. In it we tell you which process to start with and whether you need a full map.

AI due diligence goes deeper: a full check of the company with one question, where AI will bring money and where it will bring risk. It covers processes, data, systems and the cost of people and contractors.

Weeks 3–6: the AI use case map

The AI use case map shows where AI will bring money, which solution will do it, what it costs and in what order to introduce it. This is the team’s main document: it works from it for the whole next year.

The map has eight parts:

1. Processes: where time and money go, for each department.

2. Losses in money: what each process costs now.

3. Solutions: a ready service, an AI agent, a local model or nothing, if AI is not needed here.

4. Market: which solutions already exist in Russia, what they cost and how they differ.

5. Effect: savings and growth in money, the investment and the payback period.

6. Risks: data, 152-FZ (Russia’s personal data law), dependence on one vendor, AI errors.

7. Order: what to introduce first, what later and what to leave alone.

8. People: who is responsible for each solution and who needs to learn what.

Result of the stage: a list of processes with calculations and one process to start with. You pick the one where the result is easiest to measure and fastest to see. Most often it is leads and replies to customers.

Weeks 7–10: the pilot

A pilot is one AI agent on one process with a metric written down before the start. For example, the share of leads answered within a minute or the number of bookings made without an administrator.

The process owner writes the rules for the agent: what to answer, what it must not promise and when to call a person. The IT department gives the agent a separate account with access only to its own process. The lawyer checks which data the agent sees.

Our pilot costs from $1,499 and takes 14 days, with the metric written into the contract. The rest of the stage goes to the agent’s work with real customers and a comparison with the number before launch.

While the agent works, the owner reviews its answers once a week together with the process owner. Errors turn into rule changes at once and do not pile up until the end of the pilot.

If the metric is not reached, the solution is not scaled. The team looks for the cause: the agent’s rules, incomplete records in the CRM or the wrong choice of process.

Weeks 11–13: training and the report

The department where the agent works learns to accept its work: what the agent does by itself, what it passes to a person and which number shows that things are going well. Managers learn to choose the next process.

All staff get short rules: which information they may give to neural networks and which they may not. This closes the risk that people send contracts and customer lists to outside chats on their own.

At the end, the owner brings the chief executive a one-page report: the pilot metric before and after, the money the agent saved or earned, and the next process from the map.

An AI director can train departments and run the report: with us this role costs from $999 a month, 5 hours a week, with no new hire.

What the company gets after 90 days

1. A person responsible for AI who reports to the chief executive.

2. An AI use case map with calculations in money and an order of rollout.

3. One process on AI with a metric before and after.

4. A department that can work with an agent, and rules for the other staff.

5. A decision on the next step: a second process, an AI department of 3–5 agents from $4,999, or a pause if the numbers did not add up.

This is a minimal AI team. It then grows with the number of processes: process owners in other departments, a regular report and training as AI reaches new departments.

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