AI training for companies: how to build team capability

AI training for companies builds team capability when it starts from the company’s own processes and context, splits groups by role and has each person practice on the tasks they perform every week. The generic beginner AI course, with prompt tricks and live demos of generated text, produces enthusiasm on Friday and little use the following month, because participants return to their desks without knowing how to apply it to the stalled deal, this week’s campaign or the critical ticket. The training that changes the work combines practice with real cases from the operation, usage rules that make clear people delegate execution and remain responsible for the result, a deliverable that stays with the team and metrics agreed before the first session. I wrote this for the people who hire that training: HR and L&D, the executive team and the sales, marketing and Customer Success leaders.

At Iasbeck&Co we implement CRM and AI agents in B2B operations, and I have trained more than 10,000 professionals. In the workshop I ran in October, a participant asked which course the team should take to operate the agents. I answered that the platform’s own courses teach the tool, and that connecting it to the business calls for training on the company’s process inside the CRM, because the agent will work there like a new employee. I design AI training starting from that connection.

Why doesn’t a generic AI course change how the team works?

Follow a salesperson on the Monday after an open AI course. He learned to ask the chat to “act as an expert”, saw examples of emails and summaries and left with a handout of prompts. When he opens the CRM, he finds a deal stalled for three weeks, with an email history the chat cannot see, and the prompt from the handout returns a follow-up that would fit any customer of any company. He rewrites everything by hand and goes back to the old way, with no answer to the question that matters: how to use this in my task, with my data and within the company’s rules.

The 2024 Work Trend Index from Microsoft and LinkedIn, which surveyed 31,000 people in 31 countries, shows the gap between use and training. According to the report, only 39% of people who use AI at work have received AI training from their company, and 78% of AI users bring their own AI tools to work. The same study describes AI power users as 35% more likely to have received training on how to use AI for their role or function. For HR, the takeaway is that the team already uses AI, often with personal accounts and no rules.

The generic course leaves two gaps. Context is missing, because the model only knows what someone gives it, and nobody taught people to give it the ICP, the stage criteria and the company’s tone. Rules are missing too, and someone pastes the customer list into a personal chat to ask for a segmentation, with the best of intentions.

	Generic AI course	Training on the company’s process
Starting point	The vendor’s syllabus	The company’s processes, tools and policy
Group	Everyone together, from intern to director	Split by level and by area
Practice	Fictional examples	Real tasks from the week, with treated data
Usage rules	General privacy tips	Company policy, data protection law and who approves what
What stays	Handout and certificate	Prompt library by area and agent architecture
How it measures	Satisfaction survey	Adoption, quality and incidents against a baseline

AI literacy: what does everyone in the company need to know?

AI literacy is the common foundation, and it applies to the executive team, the leaders, the revenue teams and the back office. People leave understanding what a language model does, why it states with confidence something that is not in the data and which information cannot go into each tool. They also learn to write an instruction with input, rules and output format, and to check the result before using it. The literacy session fits in a few hours, as long as the exercises use the company’s own documents.

Literacy on its own reproduces the generic course with the company logo on the slide. It pays off when it works as the entry point to the area sessions and when it presents, with examples, the company’s AI usage policy: which tools are approved, what data can go into each one and what requires human review before reaching the customer. Emailing the policy PDF does not teach anyone to apply it on a Tuesday with a full queue.

How do you organize AI training by level and by role?

I organize the training in three levels, each with its own audience, practice and deliverable. Level 3, for the people who create, configure and answer for the agents, usually gathers few people.

Level	Who takes part	What people leave able to do	Practice	What stays
1. Literacy	Everyone, from the executive team to the back office	Use approved tools, write instructions with input, rules and format, check the result and recognize data that cannot go in	Exercises with company documents, without personal data	Known usage rules and acceptance of the policy
2. Advanced users by area	Salespeople, marketing analysts, support, CS and the managers of each area	Apply AI to the area’s recurring tasks, with company context and a check before logging or sending	Each participant’s real tasks from the week	The area’s prompt library
3. Agent owners and administrators	RevOps, the CRM administrator and the leaders who approve agent actions	Configure an agent, its context, credit limits and approval rules, and review errors with the team	Build or review an agent from the pilot	Agent architecture and review routine

Sales: prepare, log and follow up with the customer’s history

In the sales session, each salesperson brings two or three deals from their own pipeline. They practice preparing a meeting from the history, summarizing a call with the fields the sales process requires and drafting a follow-up that respects what the company allows them to say about price and timing. I describe these uses in AI in revenue operations, and the session lets each salesperson run them on their own deals, with someone alongside pointing out what to check.

Marketing: personalize and segment within the brand voice

The marketing analyst practices rewriting a nurturing email for real contacts, keeping the offer and the call to action of the original, and compares the base version with the personalized one. They build segments by describing the audience in plain language and check whether the result matches the written ICP. The marketing leader defines in the session what AI can change in a piece and what only changes with approval, such as promised results, deadlines and price.

Support and Customer Success: classify, summarize and anticipate risk

In support, practice revolves around real tickets that are already closed: the analyst asks for a classification using the company’s categories and compares it with what the team decided at the time. In CS, the account manager prepares a renewal meeting from the period’s tickets and notes and flags what the records left out. The post-sale session usually ends with a list of categories and fields to fix, because AI exposes on the spot what the team used to write as free text.

Leaders and executives: decide uses, approval and budget

In the leadership session, leaders choose the pilot tasks, define who approves each agent action, set the spending cap and agree on the metrics with HR. The executive team joins at least this session, because it answers for what the company tells customers and for the data it keeps.

Delegate execution, never responsibility: how do you teach usage rules?

Picture the nurturing email the agent personalized for a customer in negotiation, promising an integration the product does not have yet. The analyst approved the batch without reading it, and the salesperson found out about the promise when the customer asked for the integration in the meeting. Whoever approved answers for the email, and the training makes that rule clear before the first agent goes into production.

I teach this rule with a checking exercise. I give the group AI outputs with planted errors: a call summary that attributes to the customer a sentence they never said, a decision maker field filled in without evidence, a follow-up that offers an unauthorized discount. Each participant looks for the errors and learns to ask, for every output, where the information came from and whether it falls within their authority.

The same session covers personal data. The team needs to know which customer and lead data can go into each approved tool, for what purpose, and what to do when a request involves an entire contact list. I cover the data protection side in AI and LGPD, and in the training I bring these rules into each area’s cases, because marketing and support run into them at different points of the day.

What does AI training leave in place at the company?

The first deliverable is the prompt library by area, which participants build in the level 2 sessions from their own tasks. Each prompt comes with a card, and the library lives somewhere the team opens every day, such as the CRM or the internal knowledge base, with an owner per area who removes what nobody uses.

The second deliverable is the agent architecture, designed with level 3. For each agent, it records what it reads, what it executes on its own, what it executes and holds for approval, who approves, who reviews errors and how much it can spend per month. Without this document, each area turns on agents on its own and the company discovers the cost and the error later.

Agent	What it reads	What it executes on its own	What waits for approval	Who approves
Proposal follow-up	Deal history and proposal	Message draft	Sending to the customer	Account executive
Nurturing personalization	Contact data and engagement	Personalized version within the defined percentage	Extending to a new segment	Marketing leader
Ticket triage	Conversation, account history and knowledge base	Category, priority and summary	Customer reply on a critical ticket	Support coordinator

How do you measure the results of AI training?

The satisfaction survey measures the class. To measure the operation, HR and the leaders agree on the metrics before the first session, record the baseline and pick a few metrics per area.

Metric	How to measure	When to look	Who tracks it
Adoption	People who use the approved tools and the library prompts every week	30, 60 and 90 days	HR and the area leader
Automated tasks	Library or architecture tasks that now run with AI, by area	60 and 90 days	Agent owner or RevOps
Time saved	Time for a task measured before and after, in the same sample of people	30 and 90 days	Area leader
Quality	Drafts approved without edits, corrected classifications, fields filled in with evidence	Monthly	Area leader
Incidents	Wrong messages that reached customers, personal data in unapproved tools, spending above the cap	Monthly	Agent owner, legal or DPO

Time saved is the metric the executive team asks for first and the easiest to inflate. Measure a few tasks with the same people and avoid adding up estimates each participant makes off the top of their head. The financial return comes later, with licenses, credits and team hours in the calculation, as I detail in AI ROI. The logic of measuring behavior before revenue is the same as in sales training that changes behavior, with incidents on the dashboard as well.

How does HR hire in-company AI training?

HR recognizes the need through signs that appear before the formal request. People use personal AI accounts for company tasks, leaders ask for “an AI course” without saying what it should solve, purchased licenses sit idle and some area turned on an agent that nobody knows who approves. When two of these signs show up together, the company already uses AI without training, and L&D needs to step in before the next incident.

Before signing with a provider, HR can use the list below.

• Does the provider ask to understand the processes, approved tools and usage policy before proposing the program?
• Does the practice use the company’s real tasks, with treated or anonymized data?
• Does the program cover the usage policy, data protection and responsibility, with a checking exercise?
• Does the proposal describe the deliverables that stay, such as the prompt library by area and the agent architecture?
• Does the proposal define metrics, a baseline and follow-up after the sessions?
• Has the person leading it implemented AI in sales, marketing or support operations, beyond teaching the tool?

A roadmap for leaders: before, during and after the sessions

Two weeks before: survey current use and choose the tasks. Ask each area which AI tools the team already uses, with which account and for what, without the tone of an audit. Each leader picks three to five recurring tasks, high in volume and cheap to fix when wrong, for the practice.

Before the first session: policy and baseline. Publish the usage policy, even in a short version, with the approved tools and the data that cannot go in. Measure the chosen metrics on the pilot tasks and record the numbers with a date.

During the sessions: practice on the week’s work. Each participant brings cases from their own pipeline, campaign or ticket queue, and the area leader validates on the spot the cards that go into the library.

Thirty days later: review the library and the behavior. Remove the prompts nobody used, fix the ones that caused rework and compare adoption and quality with the baseline. Take the incidents to the area meeting as case studies, without hunting for someone to blame.

Ninety days later: decide on expansion. With the numbers on the table, the leaders decide which tasks become agents and which areas join the next round of sessions.

The training holds up through the work around the sessions: choosing the tasks, writing the rules, keeping the deliverable alive and measuring. That is the design of the in-company applied AI training I run with leadership, sales, marketing and CS groups, built on the company’s process and tools, which delivers each area’s prompt library and the agent architecture, with what each agent reads, what it executes on its own and who approves the rest.

Frequently asked questions

What is the difference between an AI course for companies and in-company training?

An open course teaches the tool with generic examples. In-company training starts from the company’s processes, tools and policy, splits groups by role and leaves deliverables the team keeps using.

How long does AI training for teams take?

It depends on the number of areas and levels. Literacy fits in a few hours, the area sessions need more practice time and the agent owner level continues in short meetings throughout the pilot, with follow-up at 30 and 90 days.

Who should take part in AI training first?

Leaders and process owners, because they choose the tasks, define who approves agent actions and agree on the metrics with HR. Then come literacy for everyone, the area sessions and the session for those who configure and administer the agents.

Does the company need an AI usage policy before the training?

It needs at least a short version, with the approved tools and the data that cannot go into each one. Without it, the training teaches people to use AI without saying where the limit is, and the team keeps deciding on its own.

How should we handle people who already use AI on their own?

Treat them as a source of cases and as candidates for advanced users in their area. Ask what they already do, bring the tasks that work into the library and show them which tools and data are authorized.

How do you measure the return on AI training?

Record the baseline before the sessions and track adoption, automated tasks, time saved on measured tasks, output quality and incidents. The financial return comes later, with the cost of licenses, credits and team hours in the calculation.