One-third of companies that use AI are spending more than planned. Thirty-three percent report higher-than-expected costs, according to a Bitkom survey conducted in April 2026 among 604 companies with 20 or more employees. The biggest risk of AI in financial controlling isn’t the technology itself. It’s implementing it without a business case.
Nevertheless, the opportunities are real and measurable. This article compares both sides, using facts rather than opinions.
A quick explanation
AI in controlling means that systems consolidate data, draft reports and comments, suggest forecasts, and flag variances. The opportunity lies in the time this frees up and in shorter cycles. The risk lies in results that no one reviews, in unclear data processing, and in projects without solid proof of benefit. Both aspects hinge on the same question: How robust are the company’s data foundation and review process?
What measurable opportunities does AI offer in controlling?
Adoption is growing rapidly. In May 2026, 54.4 percent of German companies were actively using AI, compared with 40.9 percent a year earlier. That represents an increase of 13.5 percentage points over twelve months, according to the ifo Institute’s monthly business survey.
In Controlling, the effects are particularly evident in four areas.
The common denominator is listed in the right-hand column. Every opportunity depends on the data set, not on the model.
Where does AI stand in your controlling?
We look at your monthly process and identify the biggest lever, and whether using AI can make a measurable contribution today.
What risks are underestimated in practice?
The Bitkom survey identifies three obstacles that companies themselves cite most frequently: a lack of AI expertise within the team (53 percent), data protection and legal uncertainty (41 percent), and unclear costs (37 percent). In financial control, there are two additional risks that are rarely included in the project plan.
Risk 1: Results without verification
Only 23 percent of companies systematically validate AI results, according to the EY AI Readiness Check 2026. A forecast without verification is an estimate for which no one takes responsibility. That’s not enough to manage a company.
Risk 2: Benefits that no one can quantify
47 percent of companies are unable to assess the economic value of their AI applications. For management accounting teams, this is a two-sided issue. It represents a risk within their own projects and, at the same time, a task that management accounting can take on for the entire company.
Risk 3: Data processing without a clearly defined framework
Only 24 percent of companies have implemented governance measures. Anyone entering budget data or personnel costs into a model should know in advance where that data will be processed.
Risk 4: Skills gap in the team
Only 25 percent of companies believe their employees are capable of correctly interpreting AI results. This makes the verification requirement under Risk 1 a bottleneck.
Risk 5: Pilot without a connection
69 percent of companies use AI, but only 8 percent scale it across multiple departments. The most common reason is not technical. The pilot runs alongside the process, rather than within it.
How do you assess the opportunities and risks for your own process?
These five steps will lead to a decision that you can justify to management.
- Select a recurring task from the monthly workflow and measure the time it takes today in hours. Without a baseline, you won’t be able to demonstrate the benefits later.
- Review the data foundation for this one task. Are the definitions, history, and responsibilities accurate?
- Start in read-only mode – that is, without write access to planning data. The risk remains low, and the learning curve is steep.
- Compare three results against the manual solution and document the discrepancies.
- Calculate the benefits against the costs, including implementation, licensing, and testing expenses. Only this figure will determine the outcome.
Comparing the results for each use case makes the trade-offs clear.
The Controlling offerings demonstrate how the first scenario can be implemented in practice. For the third scenario, it’s worth taking a look at AI agents in Controlling.
When is AI not worth it in controlling?
There are three situations in which we advise against getting started.
First, if the data foundation isn’t in place. If you manually map accounts month after month, you’re essentially automating inaccuracies. The correct order is: data model, then AI.
Second, if no one is responsible for verifying the data. Without clearly defined responsibility for approval, there’s no benefit – only an additional risk.
Third, if the process rarely runs anyway. A task that takes two hours twice a year isn’t worth the effort of implementation. The effort then goes into the project rather than into the results.
The honest conclusion, therefore, is this: AI in controlling is not a risk to the controller’s role. It is a risk to projects that launch without a solid data foundation. Establishing precisely this data foundation is the craft of controlling. Anyone who can reliably assess the benefits of AI in their own company is making decisions that go beyond their own reporting.
Why BI2run is the right partner for this decision
We’ve been building planning and reporting models in IBM Planning Analytics for years and understand both sides of the equation. We see where AI adds value in controlling and where clean modeling delivers more benefits than any model could. That’s why our conversations start with your process – not with a product demo.
If our assessment shows that your data foundation needs attention first, we’ll tell you. That answer saves more money than any pilot project.
- 54.4 percent of German companies actively used AI in May 2026, up from 40.9 percent the year before. Source: ifo Institute, Business Survey, June 2026 survey.
- 41 percent of companies with 20 or more employees actively use AI, double the 20 percent of the previous year. Source: Bitkom, April 2026, 604 companies surveyed.
- Obstacles from the companies’ perspective: lack of AI skills 53 percent, data protection and legal uncertainty 41 percent, unclear costs 37 percent. Source: Bitkom 2026.
- 33 percent report higher costs than expected. Source: Bitkom 2026.
- 69 percent use AI, 8 percent have scaled it across several areas, 47 percent cannot assess the economic value, 23 percent validate results systematically. Source: EY AI Readiness Check 2026.
Glossary
Frequently Asked Questions about the opportunities and risks of AI in Controlling
Which risks associated with AI in controlling are the most significant?
Two risks stand out above all others: results that no one systematically reviews, and projects without a feasibility analysis. While technical risks can be isolated, these two directly influence decision-making.
How much time does AI actually save in controlling?
That depends on the process. Recurring tasks that involve a lot of writing – such as providing comments and generating standard reports – have the greatest impact. This statement is only valid if the current workload was measured before the project began.
What requirements must a company meet?
Clear definitions, a consistent history, and clearly assigned responsibility for the audit. If any one of these three requirements is missing, the benefits are delayed.
Who is liable for an inaccurate AI forecast?
The responsibility lies with the person who approves the figure. A system does not represent a target figure to management. That is why every assignment must be accompanied by a documented approval.
Why do many AI projects in controlling fail after the pilot phase?
Because the pilot is running alongside the process. As long as the results aren’t incorporated into the monthly workflow, there’s no measurable benefit – and therefore no budget for the next step.
