The biggest hurdle in AI is not a technical one. Fifty-three percent of companies cite a lack of AI expertise within their teams as the main obstacle, ahead of data protection (41 percent) and unclear costs (37 percent). These are the findings of a Bitkom survey of 604 companies with 20 or more employees conducted in April 2026.
For controlling teams, this means the software is often already in place. What’s missing is training in AI for controlling – the kind that turns a tool into a robust process.
A quick Explanation
Training in AI for controlling is not intended to turn controllers into developers. It covers five areas: verifying AI results, understanding one’s own data set, formulating precise questions, the legal framework for data processing, and translating results into decisions. The fastest way to build these skills is by working on a real-world task from the monthly workflow, not in a seminar room.
Why is AI expertise becoming a bottleneck in controlling right now?
Adoption is growing faster than the ability to interpret the results. In May 2026, 54.4 percent of German companies were actively using AI, up from 40.9 percent a year earlier, according to a survey by the ifo Institute. At the same time, according to the EY AI Readiness Check 2026, only 25 percent of companies believe their employees are capable of correctly interpreting AI results.
This gap is the real story. A forecast proposal that no one can verify does not provide relief; it merely shifts the work to a later stage in the process.
Search behavior also reflects this trend. According to SE Ranking, the search volume for AI-related continuing education in controlling in Germany rose from 20 to 50 queries per month between October 2025 and September 2026. The need is emerging right within the teams themselves.
Specifically, what skills does a controller need?
Five fields are key. None of them requires a degree in data science.
Audit expertise is deliberately prioritized. It is the skill that gains value through AI, while the mere generation of numbers becomes less expensive. The reason why the risk increases without it is explained in “Opportunities and Risks of AI in Controlling.”
Which AI training program in controlling is best suited for which goal?
The formats differ more in terms of their benefits than in price. This overview categorizes them.
In our experience, the guided use case is the most effective. It builds expertise and delivers a usable result in the same amount of time. A good starting point is a task such as automated commenting in reporting or ad hoc analysis in controlling.
How do you build those skills in 90 days?
A process that works without an additional budget.
- Days 1 through 10: Select a recurring task from the monthly workflow and measure today’s effort in hours.
- Days 11 through 30: Define the scope. Which data can be processed in which tool, who approves it, and what needs to be documented?
- Days 31 through 55: Solve the task in read-only mode – that is, without write access to planning data. Each team member checks at least one result against the manual solution.
- Days 56 through 75: Record the discrepancies and use them to develop a verification routine. This routine is the actual learning outcome.
- Days 76 through 90: Based on the hours tracked, decide whether the task will continue to be handled this way permanently. Only then does the second use case follow.
The difference from a traditional seminar lies in Step 4. The verification routine remains with the team, even if the tool is changed later.
A guided use case instead of a slide deck
We guide a real use case from your monthly cycle and build the verification routine together with your team. In the end, you’re left with a result you keep.
When is AI training a waste of time?
Three situations in which we advise against it.
- If no system is in place after the training. Knowledge without application is lost within a few weeks.
- If the data foundation is inadequate. Then the team learns how to work with results it cannot trust. The order is: data model, then building expertise.
- If training is used as a substitute for a decision. A training budget is no substitute for an answer to the question of which processes should be implemented in the future and how they should be run.
The second most important point in this article is at the end because it often gets overlooked. Building expertise in management accounting is currently the most cost-effective investment a company can make in its AI capabilities. Forty-seven percent of companies are unable to assess the economic value of their AI applications. This assessment is precisely what management accounting is all about. Those who master it will be needed within the company for more than just their own reporting.
Why BI2run focuses on building competence through Real-world processes
We don’t train using sample data – we train using your model. From our experience working with IBM Planning Analytics and AI applications in the financial sector, we know that expertise is gained where a task is actually completed.
If our discussion reveals that you need to work on your data model first, we’ll let you know and won’t suggest a training package.
- Lack of AI skills is the most frequently cited hurdle at 53 percent, ahead of data protection (41 percent) and unclear costs (37 percent). Source: Bitkom, April 2026, 604 companies with 20 or more employees.
- 54.4 percent of German companies were actively using AI in May 2026, up from 40.9 percent the year before. Source: ifo Institute, business survey.
- Only 25 percent of companies trust their employees to interpret AI results correctly. 23 percent validate results systematically. Source: EY AI Readiness Check 2026.
- 47 percent are unable to assess the business value of their AI applications. Source: EY AI Readiness Check 2026.
- Search volume for training on AI in controlling in Germany rose from 20 to 50 queries per month between October 2025 and September 2026. Source: SE Ranking, Germany database.
Glossary
Frequently Asked Questions about AI Training in Controlling
What prior knowledge does a controller need regarding AI topics?
No programming skills required. What is needed is an understanding of your own data model and a willingness to systematically verify results. Both of these are already present in Controlling.
How long does it take to develop the most important skills?
About 90 days are sufficient for a robust testing routine for a single use case if the team sets aside two to three hours per week. Testing across multiple processes takes longer.
What role does the EU AI Act play in continuing education?
It requires companies to train employees on how to use the AI systems they deploy. Our analysis of this is included in our article on the EU AI Act.
When is an external certification a better option than in-house training?
A certification is worthwhile for roles that are responsible for a strategy or represent the organization externally. For day-to-day work on one’s own model, internal implementation is more beneficial because it is based on real data.
How do you measure the success of an AI training program?
Two figures: the number of hours spent on the assignment before and after implementation, and the percentage of AI-generated results that had to be corrected during the exam.
