In many teams, the monthly closing process still takes just as long as it did five years ago, even though AI has long been in use. According to the EY AI Readiness Check 2026, 69 percent of companies use AI, but only 8 percent have scaled it across multiple departments. AI is only effective when it’s integrated into the process, not just used alongside it.
This article compares the traditional and AI-powered processes step by step. No futuristic scenarios – just the monthly routine you’re familiar with.
A quick Explanation
Controlling AI describes the use of AI functions directly within controlling processes: data reconciliation, reporting, commentary, forecasting, and ad hoc analysis. The difference from traditional controlling lies not in different key metrics, but in the sequence of tasks. Traditionally, the controller generates the figure first, followed by the analysis.
With Controlling AI, a draft is generated automatically, and the controller’s work begins with the review. In both cases, responsibility, goal-setting, and approval remain with humans.
What distinguishes AI-powered controlling from traditional controlling?
A step-by-step comparison along the typical process chain.
The right-hand column contains the actual findings. The challenging part of the work remains; the repetitive part is eliminated. Which key figures should be included in the report is also a professional judgment call.
How exactly will the monthly schedule change?
A typical closing ceremony at a medium-sized company, shown here in both versions.
Profit isn’t generated at a single point, but throughout the entire chain. That’s why a pilot project adds little value beyond the process itself. It shortens one step, while the others continue unchanged.
“Automated Commenting in Reporting” demonstrates how automated commenting works in practice. “Self-Service for Business Units” covers ad hoc analysis in Controlling.
Which roles are shifting within the team?
We see three shifts in almost every project.
First, the role of data maintenance is growing. Whoever is responsible for definitions and master data suddenly finds themselves in a central position, because every automated design depends on it.
Second, a verification role emerges. Someone must determine how a result is verified and documented. Only 23 percent of companies do this systematically, according to the EY AI Readiness Check 2026.
Third, the work of the business units is shifting. When ad hoc questions are answered directly within the model, Controlling becomes less of a service provider and more of a discussion partner. The skills required for this are covered in AI training for Controlling.
Clarify roles before the technology is in place
Who maintains definitions, who reviews, who signs off? These three answers have a bigger impact on the value you get than the choice of tool.
What remains unchanged even with AI-powered controlling?
Four things never change, and that’s not a limitation – it’s the foundation for its use;
- Responsibility. A system does not represent a target figure to management and is not liable for anything.
- Context. The AI knows the data in the system, not the conversation with the sales manager from last week. A significant portion of the relevant knowledge isn’t captured anywhere in the model.
- Objectives. A budget is a decision, not a forecast. Confusing the two leads to discussions instead of effective management.
- Data quality. It determines the significance of every number, both before and after.
We therefore advise against its use in two situations. If the data foundation is not in place, Controlling AI automates existing inaccuracies and thereby accelerates the production of incorrect values. And if a process runs only twice a year, it does not justify the implementation effort. In both cases, the more honest answer is a pilot project – or no project at all.
That leaves us with the point we consider most important. Controlling AI does not replace controlling expertise; it simply shifts the point at which it is applied. Instead of generating numbers and then evaluating them, the work begins with the assessment of a design. Teams that consciously embrace this shift gain time for analysis. Teams that ignore it get the same reports faster – and nothing else.
Why BI2run looks at the process first, not the tool
We build planning and reporting models and support finance teams in implementing AI agents for finance and controlling. From this work, we know that the benefits arise across the entire process chain, not at a single point.
If our review of your monthly process reveals that the data structure needs to be addressed first, we’ll say so. A faster but inaccurate report doesn’t help anyone.
Fact Sheet
- 69 percent of companies use AI, 8 percent have scaled it across several areas. 23 percent validate results systematically. Source: EY AI Readiness Check 2026.
- 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, and 48 percent plan to do so. Source: Bitkom, April 2026, 604 companies surveyed.
- Only 31 percent of AI-using companies apply AI to data analysis. Source: Bitkom 2026.
- 47 percent of companies cannot assess the economic value of their AI applications. Source: EY AI Readiness Check 2026.
Glossary
Frequently Asked Questions about Controlling AI
How does Controlling AI differ from a chatbot in the financial sector?
A chatbot answers questions based on text. Controlling AI operates using the company’s data model and generates values that are incorporated into reports and planning. As a result, the requirements for verification and approval are more stringent.
Which processes are a good place to start?
Highly repetitive processes with clean data – typically involving comments and standard reporting. These deliver the fastest measurable results.
How much time does an AI-powered monthly closing save?
That depends on the proportion of manual work in the current process. The statement is only valid if the effort required for each step was measured before the changeover.
What role will the controller continue to play?
Evaluation, contextualization within the business context, and approval. These tasks are becoming increasingly important because production costs are falling.
What happens when there’s a disruption in business operations?
That’s when history-based recommendations lose their value. In such situations, manual judgment matters more than any model.
