ESL.EDU.PL Guide
7 traps of AI implementation in a company
and how to avoid them
Most companies can purchase licences for AI tools. It is more difficult to ensure that staff use them and that management can see the results. Below are seven recurring pitfalls in implementations and how to avoid them.
This guide is based on the methodology we use in ESL.EDU.PL implementation programmes.
- Trap 1
Licences purchased but not used
The core issue
Everyone gets access at once, the training lasts one day, and after a month, only a handful of people are using the tool. No one is measuring this, so no one is taking action.
How to avoid it
Measure how many people are working with AI every week, not how many licences have been purchased. Roll out licences in waves, once you have confirmed that the employee knows how to use them. Set a success metric before making the purchase.
Check your situation: Do you know today how many employees are using AI on a weekly basis?
- Trap 2
Overly broad file permissions
The core issue
An AI assistant shows the user everything they have access to. If a folder with salaries is open to the entire company, the AI will find it. This is not a tool error, but the exposure of existing disorganisation.
How to avoid it
Before assigning the first licence, perform a permission audit in SharePoint, OneDrive, and Teams. Label sensitive documents and lock folders currently open to everyone.
Check your situation: Who in the company currently has access to documents containing salaries and contracts?
- Trap 3
Lack of rules, leading to data being fed into private tools
The core issue
Without clear guidelines, employees use private AI accounts and paste company data into them. They do this in good faith, as they want to work faster.
How to avoid it
Prepare a one-page AI policy that can be understood in two minutes. The three-colour data classification system works best: green (public, allowed in any approved tool), yellow (commercial, financial, internal, only in company tools on company accounts), and red (special category personal data, salaries, recipes, only in processes approved by the responsible person). Add a list of approved tools.
Check your situation: Does the employee know what data must not be entered into AI tools?
- Trap 4
One-off training instead of habit formation
The core issue
Training provides knowledge, but habits are only formed through daily work. Without post-training support, people revert to old ways of working.
How to avoid it
Apply the 70-20-10 principle: 70% of learning through real work, 20% from other people, 10% through training. Give everyone one weekly task from their own workflow and discuss the results for 10 minutes during team meetings. Appoint individuals within departments to support their colleagues.
Check your situation: What happens to the employee in the week following the training?
- Trap 5
Lack of clear expectations for roles
The core issue
'Please use AI' is not enough. Sales, finance, and HR need different things, and employees do not know what is expected of them.
How to avoid it
Build an AI competency matrix: areas of competence and proficiency levels, ranging from knowledge of policies, through working independently with the tool, to designing solutions for the team. Set a target level for each job group. This makes it clear who has which developmental task.
Check your situation: Does every job group have a defined expected level of AI proficiency?
- Trap 6
Leaders do not lead by example
The core issue
If the management and managers do not use AI, teams will dismiss it as a passing fad. Decisions about the pace of change are made in leaders' offices.
How to avoid it
Start with the leaders. Have them work on their own tasks for two days, establish rules for their teams, and select processes to redesign. Each leader should leave with a plan for their department.
Check your situation: Will the leaders undergo training before their teams?
- Trap 7
Lack of measurement makes it impossible to justify the budget
The core issue
Without numbers, 'we are saving time' remains just a statement. When the question about return on investment comes up, there is no answer.
How to avoid it
Before starting, measure 2–3 tasks per person. After a few weeks, measure them again and compare with a group not using the tool to distinguish the AI effect from seasonal variations. Record use cases with time spent before and after, and convert these into financial value. Calculation example: 100 people × 1 hour per week × 44 working weeks = 4,400 hours per year. Substitute your own figures and your own hourly rate.
Check your situation: How and when will you measure the implementation effect?
Ten questions before deciding on implementation
- 1Who on the board is the programme sponsor and has time for monthly decision-making?
- 2What success metric will you establish before purchasing licences?
- 3Have the file permissions been reviewed?
- 4Do you have a one-page AI policy?
- 5Which three processes would you like to optimise first?
- 6What level of AI proficiency do you expect from each job group?
- 7Who will act as the support contact for colleagues in each department?
- 8How has post-training support been planned?
- 9Will leaders undergo training before their teams?
- 10How and when will you measure the impact?
Unsure where to start? Take our 5-minute organisational AI readiness assessment or schedule a 30-minute consultation. No strings attached.