For years, AI in controlling has promised greater efficiency. In practice, however, controllers still gather data manually, identify variances on their own, and build reports by hand. This takes hours that could otherwise be spent on actual analysis.
Claude Duet is changing that. In October 2026, Anthropic released an AI agent that doesn’t wait for questions. It acts on its own initiative, retrieves data, detects anomalies, and suggests actions.
AI in Controlling refers to the use of artificial intelligence to automate planning, analysis, and reporting tasks. Claude Duet is an AI agent developed by Anthropic that operates autonomously on systems: it opens browsers, reads reports, identifies patterns, and acts without approval at every step.
What is behind the term AI in controlling?
Controllers have always been numbers experts. The nature of the work has changed: more data, more systems, more pressure, and faster reporting.
AI in controlling is intended to alleviate this pressure. Early approaches focused on automation: automatically generating standard reports and automatically importing actual figures. That helped, but it didn’t solve the actual problem.
The problem isn’t just data collection. It’s the interpretation, the analysis of deviations, and the assessment: What is relevant, what requires attention, and where does action need to be taken?
Recent studies show that controllers spend up to 60 percent of their time gathering and preparing data. Only 40 percent is devoted to the actual analysis. AI agents such as Claude Duet are designed to reverse this ratio.
What is Claude Duet, and how does the AI agent work?
Claude Duet is not a chatbot extension. In October 2026, Anthropic released a standalone AI agent based on Claude Opus 5.5.
The key difference from previous AI tools: Claude Duet runs on its own virtual machine with browser access. It opens programs, navigates websites, reads reports and writes results back. Without approval at every step.
That makes it a true AI agent. Not a copilot that makes suggestions. An agent that delivers.
For controllers, this means: Claude Duet can pull the latest plan-vs.-actual variances from your BI system at 6:00 a.m., identify critical items, and draft a comment—all before you even open your laptop.
This table shows how Claude Duet differs from previous approaches:
Are you using IBM Planning Analytics or TM1?
We’ll show you how Claude Duet works with your existing system environment—without data migration and without new infrastructure.
How does Claude Duet put AI in controlling into practice?
A realistic scenario from the month-end close:
- Claude Duet opens the BI system and reviews the latest cost center reports.
- It compares actuals with plan and prior year.
- Deviations above a defined threshold are highlighted and annotated.
- Claude Duet opens the ERP system and checks which entries caused the variance.
- It creates a structured analysis report with recommended actions.
- The report is saved in the specified location, ready for approval.
With Claude Duet, this process takes 15 minutes. Without it: several hours.
This isn’t a future scenario. It’s the current state of the art, available since October 2026.
Which tasks can Claude Duet take over in controlling?
Not every task is suitable for an AI agent. These are the areas that are:
- Variance analysis: Automatically detecting and commenting on plan-vs.-actual variances
- Month-end close support: Data preparation, posting review, report creation
- Rolling Forecast: Updating forecast models based on current actual data
- Liquidity planning: Daily monitoring of incoming and outgoing payments
- Benchmarking: Comparing a company’s own KPIs with industry or competitor data from public sources
- Reporting: Preparing management reports in a predefined format
What remains the responsibility of humans? Strategic assessment, internal and external communication, and deciding on courses of action. AI agents provide facts and analyses. You make the decisions.
Why is an AI chatbot no longer enough for controllers?
Many companies have already taken their first steps with AI: ChatGPT licenses, Microsoft Copilot, and internal chatbots. This has brought benefits, but it also has clear limitations.
An AI chatbot answers questions. It doesn’t perform any tasks. You have to provide it with data, formulate the question, and process the result manually.
An AI agent like Claude Duet works differently. You give it a goal. It plans how to reach that goal, carries out the necessary steps and reports back when it’s done or needs a decision.
The difference isn’t gradual. It’s fundamental. Chatbots are reactive. AI agents are proactive.
A chatbot helps you save time when drafting texts. Claude Duet helps you save time on your actual controlling work.
When is Claude Duet worth using in controlling?
Claude Duet is not a panacea. There are conditions that must be met.
Claude Duet is a good fit if these conditions are met:
- Your data is stored in digital systems, not in Excel silos or analog sources.
- You have recurring controlling tasks that follow a clear pattern.
- You can define what constitutes a good result. AI agents need clear goals.
- You’re ready to document processes. What is currently implicit in your minds needs to be made explicit.
Claude Duet is less suitable when controlling decisions are heavily influenced by political factors that are not reflected in the data. Or when systems are so fragmented that digital integration is not possible.
We can find out if that’s the case for you in a brief conversation.
How do you prepare your controlling for AI agents?
The biggest mistake when getting started with AI: trying to do too much at once. If you try to automate everything, you’ll end up automating nothing.
This approach has proven effective:
- Choose one process: Not the most complex one, but the one that is repeated most often.
- Check the data: Is the input data available in a usable form? Consistency, completeness, timeliness.
- Define the goal: What should the AI agent deliver in the end? Concrete, measurable, verifiable.
- Start a pilot: Small scope, tight time frame, clear success criterion.
- Learn and scale: What works gets rolled out. What doesn’t gets adjusted.
This approach typically yields the first measurable results within six to eight weeks. It’s not a major transformation project, but rather a concrete first step.
Why BI2run for getting started with AI in controlling?
BI2run has been advising companies for years on the use of BI and AI in finance and controlling. We are familiar with the systems that controllers use on a daily basis: IBM Planning Analytics, TM1, SAP, and various ERP solutions.
Claude Duet is the first AI agent tool that we can wholeheartedly recommend. Not because it’s perfect, but because it fills the gap we see time and again among our clients: too much effort spent on routine tasks, too little capacity for real analysis.
We’ll help you integrate Claude Duet with your existing systems. We’ll handle the technical integration and guide you through setting up your first automated workflows.
If the check shows that Claude Duet isn’t the right fit for you yet, we’ll tell you that too.
Glossary
FAQ
What is Claude Duet, and how is it different from ChatGPT?
Claude Duet is an AI agent developed by Anthropic that operates autonomously on systems. ChatGPT is an AI chatbot that provides answers to questions. The difference: Claude Duet performs tasks independently, while ChatGPT assists with phrasing. For controlling, this means that Claude Duet can automatically perform variance analyses. ChatGPT can only describe how to do them.
How secure is corporate data with Claude Duet?
Anthropic operates Claude Duet in accordance with enterprise security standards. Data does not leave the defined system environment without explicit configuration. For use in regulated industries or with particularly sensitive data, we recommend that you consult with your legal and IT departments.
Which systems does Claude Duet support in controlling?
Claude Duet is browser-based and can, in principle, work with any system that is accessible via a web interface. This includes BI solutions, ERP systems, planning tools, and reporting platforms. For direct API integrations with IBM Planning Analytics, SAP, or other systems, it is advisable to have specialists handle the configuration.
How long does it take to introduce Claude Duet in controlling?
The first automated workflows are productive within four to eight weeks. Prerequisites: a clearly defined process, a clean data foundation, a defined goal. A full rollout across several controlling processes typically takes three to six months, depending on system complexity and available resources.
How much does Claude Duet cost for businesses?
Anthropic offers Claude Duet in various pricing plans based on usage volume and company size. Specific pricing is available directly from Anthropic or through partners such as BI2run. More important than the list price is the ROI calculation: How much does the manual effort involved in routine tasks cost you today?
When does AI in controlling not make sense?
AI agents are not a panacea. If data sits in analog sources or is highly fragmented, the effort to clean it up often outweighs the benefit of automation. The same applies when controlling decisions depend on political factors that are not reflected in the data. An honest check at the start saves effort and frustration.
