How to Implement AI in Business: Where I Start With a Team

To implement AI in business, start with one recurring task, map how it runs today, write the data rules, train the people who do that task on their own documents, and measure the result before you pick the next one. A tool licence and a company-wide announcement give you accounts. One task with one owner gives you a habit.

I run AI training and company programmes for teams with ChatGPT, Claude and other tools, and I build my own companies with the same tools. In Lithuania the work runs through my company Retos galimybės (opens in a new tab), which also builds process automation and custom AI tools. Below is the order I follow with a new team, with the first month laid out.

You are not late. Eurostat counted 20.0% of EU enterprises with 10 or more employees using AI technologies in 2025, up from 13.5% in 2024. In Lithuania the share rose from 8.76% to 21.3% in the same year, one of the three biggest jumps in the EU.

How to implement AI in business, step by step

1. List the recurring work for one week

Ask each person on the team to note every task that repeats: replies, reports, proposals, summaries, data entry, translations. Next to each task, they write how long it takes and how often it comes back. One week of notes beats any workshop on "AI opportunities", because the list comes from your work and carries real hours.

2. Map one process from start to finish

Pick the task with the most hours behind it and draw how it runs today: who starts it, which documents go in, which decisions people make, what comes out and who receives it. The people who do the work draw the map; a manager or an outside expert only runs the session. I explain the method in how to do process mapping.

On the map, mark each step with one of three labels:

AI drafts, a person checks: writing, summarising, sorting, translating
A person decides: prices, approvals, anything with legal or client risk
Software moves it: copying data between tools, notifications, filing

The third label is automation, and it often needs no AI at all. How to automate business processes covers that part.

3. Pick the first workflow

The first workflow has to win, because the team judges AI by it. Good first picks share four traits:

It repeats every week.
It runs on text, documents or exports.
One person owns it and wants it to get easier.
A colleague can check the output in a few minutes.

Replies to inbound requests, meeting notes turned into tasks with owners, and a monthly summary from a spreadsheet export fit. Pricing decisions and anything that goes to a court, a regulator or a bank do not belong in round one.

4. Write the data rules before anyone pastes a file

Decide which tools are approved and which account type the team uses. The account type changes the data rules. In a personal ChatGPT account, conversations can be used to train OpenAI models until the user switches off "Improve the model for everyone". In ChatGPT Business and Enterprise workspaces, OpenAI does not use content for training by default.

Then write one page: what may go in, what never goes in (client personal data, contracts under NDA, passwords), and whom to ask. Hand it out on the training day.

5. Train the team on that workflow

People learn AI on their own task or they forget it by Friday. In my team workshops each group brings its own documents and clients and leaves with prompts and workflows it uses the next day. The agenda I use is in AI training for employees, and the daily habits are in how to use ChatGPT at work.

6. Measure, then pick the next task

Before the rollout, write down how long the task takes today. Two to four weeks later, ask the owner again and look at the outputs. If the time dropped and the quality held, move to the next task on the list from step 1. If people stopped using the workflow, find out at which step they dropped it and fix that step first.

An AI implementation plan for the first month

This is how a company programme with me runs:

When	What happens	What you have at the end
Before we start	A 30-minute call: you tell me what slows the team down, I tell you whether training will fix it	A yes or no, and the reason
Agreement	We agree in writing what I deliver, by when and at what price	A written scope
Weeks 1-2	Audit of how work runs today: the task list, one process map, the data rules	The first workflow, chosen with its owner
Rest of month one	The first workflow goes live; I run the sessions and build it with the team	One process running with AI, checked by a person
After month one	I come back until the team uses the workflows without me	A team library, an owner, the next task

The audit stays short on purpose. A two-month study of "AI readiness" teaches the team nothing about AI, while one live workflow in month one teaches them a lot.

Where AI implementation stalls

Watch for these five causes:

No owner. A workflow that belongs to everyone belongs to no one. Name one person per workflow.
No data rules. People either paste everything or paste nothing. One page fixes both.
Generic training. A demo on a poem proves nothing to an accountant. Use the team's documents.
No return date. Set the review date before the training day ends.
Fear for the job. Say in clear words what AI takes over and what stays with people. My post on whether AI will take your job helps that conversation.

The law also points this way. Article 4 of the EU AI Act asks companies that use AI systems to take measures to support the AI literacy of their staff, and since 27 July 2026 it asks for measures, with no guaranteed level. The European Commission says there is no need for a certificate and suggests an internal record of trainings. Training people on the workflow you roll out covers the duty and the habit at once. The details are in AI Act Article 4.

Training, a consultant, or both

A company that knows its first task needs training and someone who builds the workflow with the team. A company that has no idea where AI fits needs the audit first. I compare the two roles in what does an AI consultant do.

Start with one task

Send me the task that eats the most hours in your team and how many people do it. On a 30-minute call you hear whether training fixes it or something else does. Sessions run in Lithuanian or English, on site in Vilnius or remote. See the formats on my AI training page or book a call.

Questions and answers

How do you implement AI in a business?

Start with one recurring task that eats hours every week. Map how it runs today, decide which steps AI drafts and which a person checks, write one page of data rules, then train the people who do the task on their own material. Measure the time before and after, and only then pick the next task.

How long does AI implementation take?

In my company programmes the audit of how work runs today takes one to two weeks, and the first AI workflow goes live in month one. After that I coach the team until it uses the workflows without me. Company-wide change is a series of these cycles, one process at a time.

What should a company automate with AI first?

A task that repeats every week, runs on text or documents, has one person who owns it, and produces an output a colleague can check in a few minutes. Replies to inbound requests, meeting notes turned into tasks and monthly summaries from an export are typical first picks.

Do we need an AI strategy before we start?

You need three decisions: which tools are approved, which data may go into them, and who owns the first workflow. A longer strategy document helps later, once you know from one working process what AI does in your company.

Is AI training required by law in the EU?

Article 4 of the EU AI Act requires companies that use AI systems to take measures to support the AI literacy of their staff. Since the Digital Omnibus took effect on 27 July 2026, the law asks for measures and no guaranteed level. Training people on the workflow you roll out is one such measure; keep a record of who attended.