--- title: "AI Training for Business, Up to 15 People | JustAutomate" description: "AI training on your own tasks, not slide examples. Up to 15 people per group, roles assigned by name, a banned-data list and a prompt library that stays." url: "https://justautomate.ai/en/services/ai-training-for-business/" lang: en --- Consulting, training and support # AI training for business that survives the first week back Training really starts on the day your team goes back to its own work. So we run the exercises on your real processes, every role leaves with its own scope, and the company leaves with a list of data that must not go into a model and a prompt library people can use the next morning. Groups of up to 15 people, so everyone gets through their own task instead of watching someone else's. Book a free consultation [See how the workshop works](https://justautomate.ai/en/services/process-workshop/) - up to 15 peoplein one group. Everyone works through an exercise on their own task, instead of watching the trainer click on a screen at the front - 4 to 8 hoursper session, depending on the format: fundamentals, prompts for one specific role, or no-code automation - 2 weeksof email support afterwards, because the real questions turn up on the first task someone does alone, not in the room The problem ## Why does a team go back to the old way of working after AI training? A company buys licences for everyone and books a training day. Three months later a salesperson uses the model to fix grammar in an email, marketing pastes in a finished post and asks for it to be shortened, someone translates a document. The work that eats hours every week, research on a company before a meeting, an offer from a template, the monthly report, still goes by hand. The reason is simple. The session showed a tool on an example from a slide, not on the task that one specific person has in their calendar on Wednesday. Nobody said who owns what either: who writes the prompts, who looks after the automation when it stops, who checks whether any of this actually shortens the work. Without that split, a week later everything lands back on whoever was most curious. There is also a layer that is easy to miss. [Article 4 of Regulation (EU) 2024/1689, the AI Act](https://eur-lex.europa.eu/eli/reg/2024/1689/oj), requires providers and deployers of AI systems to ensure a sufficient level of AI literacy among the people who operate those systems on their behalf. It has applied since 2 February 2025\. Separately, [the GDPR](https://eur-lex.europa.eu/eli/reg/2016/679/oj) applies whatever window someone pastes text into, and hardly any company has a written list of data that must stay inside. - Licences renew every month and a handful of people actually open them - Everyone writes prompts their own way and nobody knows which one works - Nobody checks the answer, so an invented number goes out to a client - Data that should never leave the company ends up in a model - The automation built during the session dies after a month with nobody to fix it Why now ## What changed, so that AI training can be judged on results? The model is no longer a separate browser tab. It sits inside the tools your team already keeps open for eight hours a day: Google Workspace, Microsoft 365, Slack, Notion, ClickUp. So the exercise happens in the same place the real work happens, and nobody has to change a habit to use it. The second change is no-code automation. A scenario in [n8n](https://docs.n8n.io/), Make or Zapier can be kept alive by a non-technical person, as long as somebody has shown them what to do when it stops: where the execution history lives, how to retry a failed step, when to call for help. That is a separate skill and a separate role, and usually nobody is put in it. The third thing is the AI agent. BOND, our company AI agent, answers from company data in 5 to 10 seconds, so the team stops assembling prompts and starts asking questions. But a question about data needs a different kind of care than a request to shorten a paragraph, because the answer always sounds credible. We teach that separately, on your numbers. What changes ## What can the team do after the training? - ### They practise on your process, not on a demo Before the session we collect 5 to 8 tasks the team repeats every week: research before a meeting, an offer, a product description, a call summary, a report. In the room, everyone runs their own task from start to finish. The exercise material is your own files and templates, prepared before the session - ### They know who owns what We split out three roles: the person writing prompts for their own job, the owner of the automation scenario, the manager who reports on the effect. Each role leaves with its own scope and its own list of next steps. Three roles filled by name before the session ends, written into the materials - ### They know the banned-data list We agree what may go into a model and what may not: client personal data, individual price lists, contracts, HR records, code under an NDA. We put it on one page you can pin above a desk. One page with the banned-data list and the line between a company account and a private one - ### They check the answer instead of trusting it A model invents most freely where it sounds most certain: numbers, dates, names of regulations and quotes. We show three ways to verify and a set of control questions for each of those four categories. An exercise on a real hallucination, triggered on purpose on your data during the session - ### They keep the automation alive when it breaks The scenario owner learns to read execution history in n8n, Make or Zapier, retry a failed step and spot when the input format has changed. Without that, the first error ends the automation. Everyone on this format leaves with one working scenario they built themselves - ### They have somewhere to go with a question After the session you keep the materials, the prompt library built during the exercises and the recording if the format was online. For 2 weeks we answer questions by email as the first real tasks come up. Materials, prompt library, recording of the online format and 2 weeks of email support Proof ## Have you got your own team to actually use AI? Yes, and it is the only adoption number we can show without asking a client for permission to publish their data. An AI agent inside our founders' agency, launched in March 2026\. Below is how real usage grew, together with what the number does NOT prove. Built for ourselves. These are not client deployments but implementations inside the founders' own companies. We say so plainly, because proof you cannot verify is not proof. - ![JustIdea Agency](https://justautomate.ai/klienci/justidea.png)![JustBot](https://justautomate.ai/agenci/justbot-80.webp)JustBottheir agent JustIdea Agency, a performance marketing agency of around 30 people 23 people from the company work with the agent 38 systems wired in permanently 112 processes run without a human 475 thousand lines of the agent’s code An agent wired into Google Ads, Meta, the CRM, the task system, the SEO tool, Analytics, Search Console and the data warehouse. First commit 1 March 2026, today 23 people use it. 8 of 38 systems it works in - Google Ads - Meta Ads - HubSpot - ActiveCollab - Ahrefs - GA4 - Search Console - BigQuery Before In March 2026, the first month after launch, 4 people used the agent and it ran 83 operations. The rest of the team knew the tool existed and worked exactly as before. After In August 2026, 23 people were using it and operations reached almost 63 thousand in the month. The growth came in steps, as each new team was brought into working with the agent. From 4 people and 83 operations in March to 23 people and 62,985 operations in August 2026 How much the team actually uses it Operations run by the agency agent month by month in 2026, from the first commit in March. From 83 operations in March to almost 63 thousand in August, with users up from 4 to 23\. The curve is here because a rollout nobody uses looks identical in a report to a successful one. Measurement: Agent call log in the data warehouse, March to August 2026, counted at the level of a single tool call. And in your company ## What if it does not stick here? Four sentences come up on almost every consultation. This is what we answer. - We already did AI training and nothing came of it. We know the pattern: four hours of showing a tool and no word on who does what the next day. That is why our exercises run on your tasks, and the session ends with three roles filled by name and a list of jobs for the first week. After 30 days we check which tasks really moved to the new way. - Half the team does not want to touch this. We do not try to convert everyone at once. We start with the people carrying the most repetitive work, because they see the difference on day one. The rest join when a concrete example is going round the office instead of a promise. A group of up to 15 people lets us sit down with each person separately. - We cannot put our data into an external model. Agreed, and that is part of the programme rather than an obstacle. During the session we write the banned-data list, agree which account is the company one, and show how to do the same task on anonymised data or on a model inside your own infrastructure. If a process cannot leave the company at all, we say so plainly. - We are 6 people, we cannot free a whole day. At six people the 4 hour format is usually enough and there is no point buying eight. We run the shorter session for everyone plus one extra session for whoever becomes the scenario owner. We teach on site at your office or online, so the day does not disappear into travel. Models and chat ChatGPT, Claude, Gemini, BOND as the company AI agent Team tools Google Workspace, Microsoft 365, Slack, Notion, ClickUp, Jira, Asana CRM and sales HubSpot, Pipedrive, Salesforce, LiveSpace No-code automation n8n, Make, Zapier, custom API integrations Data and reporting BigQuery, Google Sheets, GA4, SQL databases Commerce and finance Shopify, WooCommerce, BaseLinker, Xero, sevDesk ## Roles in the team, what each one practises and how to check it The split into roles matters more than the list of tools. Without it, a week later everything is back with one curious person, and when they go on holiday the work goes back to manual. The third column is how you tell after 30 days that a role is alive. | Role | What they practise in the session | How you know it is working | | ------------------------------- | --------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------- | | Prompt author, every job | writing a prompt for their own task, fixing it against the output, 3 ways to verify an answer | their prompts sit in the shared library, not in a private notebook | | Scenario owner | building a scenario in n8n, Make or Zapier, reading execution history, retrying a failed step | they restart a stopped scenario themselves, without raising it with us | | Data and reporting person | asking questions about data, automating a report, checking every number against the source | the report goes out with a date and a data source, not just a number | | Owner of outbound content | spotting hallucinations in names, dates, regulations and quotes | nothing reaches a client before the facts on the list are checked | | Team manager | reporting the effect: which tasks moved, how long they used to take, what is still manual | at the 30 day review they bring numbers, not impressions | | Accounts and permissions person | company accounts, privacy settings, the banned-data list, reporting an incident | the banned-data list is written down and the whole team knows it | In smaller companies one person takes two roles and that is normal. What is not normal is all six staying with the same person. How we do it ## Four steps from unused licences to work done the new way Each step ends with something you can judge before you decide on the next one. 1. ### Discovery before the session We collect 5 to 8 tasks the team repeats every week and ask how long each one takes. We check which tools you already have and what data moves through those tasks, because that decides what may go into a model. This is a 60 minute conversation, not an audit. 2. ### A programme built around roles, not around a tool We set the scope: the 4 hour format for the whole team, 8 hours for people writing prompts professionally or for scenario owners. Each role gets exercises on your tasks and examples from your industry. You approve the programme before the session. 3. ### Training on real files On site at your office or online, groups of up to 15 people. Everyone runs their own task to the end and sees the result. We trigger a wrong answer on purpose, so the team sees a hallucination on their own data, and we write the banned-data list. Roles are filled by name before anyone leaves the room. 4. ### Materials and a measurement after 30 days You keep the materials, the prompt library from the exercises, one page of safety rules and the recording if the format was online. For 2 weeks we answer by email. After 30 days we go through the list with the manager: which tasks moved, which went back to manual work and why. Before and after ## How the team's week changes ### Before - The model is used for fixing grammar in emails and for translations - Research on a company before a meeting takes half an hour every time - Nobody knows whether an answer is true, so to be safe nobody uses it - Client data goes into private accounts because nobody set the rules ### After - Every repeated task has its prompt in the shared library - Research before a meeting closes in a few minutes and lands in the CRM - Every number and name in an answer is checked against the control questions - The banned-data list is pinned above the desk and the whole team knows it The first prompts work on the day of the session. Whether the team still uses them is known after 30 days, and that is the only moment when a conversation about the effect makes sense. For the person signing this off ## What the team lead gets out of it You stop being the only person who knows how to approach this. The work spreads across roles, and after 30 days you hold a list of tasks that really moved to the new way. That is something you can put in front of a board, unlike an invoice for a training day and a feeling that it was interesting. The second thing is risk. A banned-data list and written rules for verifying answers take the question of who allowed a contract into a chat window off your desk. Article 4 of the AI Act requires AI literacy for the people operating these systems, so documented training is also evidence that the company dealt with it. - Three roles filled by name, instead of one curious person - One page of safety rules, known to the whole team - A measurement after 30 days: which tasks moved and which came back - Materials and the prompt library stay in the company, also after people leave What it costs ## The workshop has a published price, the training is quoted after it The process workshop costs **EUR 3,500 net per week** and runs from 1 to 7 weeks. Before a training, one week is usually enough, because the question is narrow: which tasks the team repeats and which of them suit a model. We quote the training after the workshop, once the number of people, formats and roles is clear. Model licences and accounts on the automation platforms go straight to the vendor, with no commission from us. A model query costs EUR 0.005 to 0.07 depending on the model. If your team already has ChatGPT, Claude or Gemini, you buy nothing new for the training. - ### Process workshop EUR 3,500 net per week, usually one week before a training A self-contained piece of work for companies that do not yet know what to teach. It ends with a task map and a programme you can run without us. - A map of 5 to 8 repeated tasks with the time each one takes today - Which of them suit a model and which suit an automation - The list of data that must not leave the company - A training programme split by role and by session length Start with the workshop - Most popular ### Workshop and team training Quoted after the workshop scope follows from the number of people, formats and roles The option most companies choose. The workshop, then training on your tasks, with roles filled by name and a measurement after 30 days. - Everything in the workshop - A 4 or 8 hour session, on site or online, groups of up to 15 people - Exercises on your own files, including a hallucination triggered on purpose - Materials, prompt library, recording of the online format, one page of rules - 2 weeks of email support and a review after 30 days Talk about a training - ### Training and an AI agent Quoted after the workshop for companies that want to ask about their own data, not about general knowledge Everything in the team training plus BOND, the company AI agent that answers from company data in 5 to 10 seconds. The agent takes about 2 months to go live. - Everything in the team training - BOND wired into your systems, with a pilot on real data - A separate session on verifying answers built from company data - Usage measured at 30 and 90 days Ask about the agent The workshop ends with a document that stands without us: a task map, a banned-data list and a training programme. You can run it with anyone. And if the workshop shows that your problem is not missing knowledge but messy data or a missing process, we say so plainly and we do not sell a training that would not fix it. ## Frequently asked questions How is this different from a ChatGPT course? A course shows the tool on examples the trainer prepared. Before our session we collect 5 to 8 tasks your team repeats every week and build the exercises on those. Everyone leaves with a finished prompt for their own task, not with notes about techniques they will never open again. Which training formats do you run? Fundamentals for the whole team in 4 hours, prompt engineering in 8 hours covering chain of thought, few-shot and the system prompt, sessions for sales and for marketing in 4 hours, and no-code automation in 8 hours on n8n, Make or Zapier. We pick the scope after the discovery call. How many people can take part in one session? 15 people at most, and that limit is hard. In a bigger group there is no time for everyone to run their own task to the end and get feedback, and the session turns into a presentation. For a larger team we run several rounds, usually splitting people by role. What must never be pasted into a model? We set the list during the session, because it depends on your industry and your client contracts. It usually holds client personal data, individual price lists, contract text, HR records and code under an NDA. We also show how to run the same task on anonymised data or on a model inside your own infrastructure. How is the team supposed to spot that a model made something up? A model sounds most certain exactly where it is most often wrong: numbers, dates, names of regulations and quotes. We teach three ways to verify and a set of control questions for those four categories. During the session we trigger a hallucination on purpose, on your data, so the team sees one with their own eyes. There is more in our guide on avoiding hallucinations. What stays in the company after the training? Materials for going deeper on your own, the prompt library built during the exercises, one page with the safety rules and the banned-data list, and the recording if the format was online. For 2 weeks we answer participants' questions by email as the first real tasks come up. Do we need the process workshop before the training? Not always. If you already know which tasks you want to move to a model, a 60 minute call and a programme are enough. The workshop earns its place when the team is still looking for things to automate, or when the training is meant to be the first step towards an AI agent. How will you check that the training changed anything? After 30 days we sit down with the team manager and go through the task list from discovery: which ones moved to the new way, how long they take now, which went back to manual work and why. When something goes back, the reason is almost always the same: nobody was given the role that covers it. ## Let us check what your team actually needs The free consultation takes 30 minutes. You tell us how many people you have, which tasks come back every week and whether anyone has tried doing them with a model. We tell you which format makes sense, and when training is premature because the data has to be sorted out first. You write directly to Paweł, with no reception and no tickets. We run at most two rollouts at a time, so that someone is actually sitting on each of them. [Book a free consultation](https://justautomate.ai/en/contact/) **PS.** If the model licences have been renewing for six months and three people use them, the tool is not the problem. The team does not know which of their tasks to do with it, or who answers for that. Training starts there, not with a list of techniques. ## Related - [AI consulting](https://justautomate.ai/en/services/ai-consulting/) - [Prompt engineering for business](https://justautomate.ai/en/guides/prompt-engineering-for-business/) - [AI hallucinations: how to avoid them](https://justautomate.ai/en/guides/ai-hallucinations-how-to-avoid/) - [AI in business: getting started](https://justautomate.ai/en/guides/ai-in-business-getting-started/) - [BOND: business AI agent](https://justautomate.ai/en/services/bond-business-ai-agent/) - [Process workshop](https://justautomate.ai/en/services/process-workshop/) --- **JustAutomate** · contact: contact@justautomate.ai · form: https://justautomate.ai/en/contact/