AI Future in Tax Integrates Human Expertise and Machine Learning

Artificial intelligence technology in tax is advancing rapidly. In just a year, KPMG expects to move from chatbots to advanced systems that can manage entire processes.

AI “personas"—customized templates for different tax scenarios—are helping consultants work more efficiently. These tools are facilitating the advance of next-generation AI systems poised to significantly enhance the efficiency of consulting teams and tax departments.

The integration of human expertise and machine intelligence in tax operations has commenced.

AI at Scale

Today, AI models already handle a significant portion of the work in selected tax projects. They help by selecting legal rules, extracting data, and doing initial analyses. The rest of the work involves using data-processing tools, allowing tax professionals to focus on complex legal issues and refine AI suggestions.

Large companies are adopting similar strategies, using AI to simplify high-volume tax and legal tasks. This approach requires only basic AI knowledge but effectively streamlines complex processes.

One such example can be found in KPMG in Germany, which implemented automation for the annual creation and updating of local files and master files for a client that is a company in the European real estate sector.

The client was able to streamline the traditionally labor-intensive transfer pricing documentation process, revealing that generative AI was effective for 40% of tasks that had previously caused significant bottlenecks.

‘Think Tank’ Agents

In the next year and a half, KPMG observes a tendency for AI in tax and legal services to evolve beyond simple task automation or chatbots to more advanced “think tank” agents. These innovative agents will operate with an emphasis on achieving comprehensive objectives.

Rather than adhering solely to predefined instructions, these agents will be equipped to manage entire workflows autonomously. Their responsibilities will include gathering information, interpreting legislation, analyzing various scenarios, and drafting essential documents.

Automating these processes enables professionals to concentrate on complex matters that demand human judgment. In addition to streamlining routine tasks, such agents leverage data analytics to identify emerging trends and evaluate evolving organizational processes against anticipated outcomes, enhancing the quality of tax and legal decision-making.

However, achieving the necessary autonomy and precision in these domains goes beyond relying on broad internet data.

Because even a single percentage-point error can have serious financial consequences, agents must be built using AI models that are carefully fine-tuned with situation-specific tax and legal content. But because merging and maintaining these datasets can be complex, many organizations will seek support from expert partners to keep their AI agents both reliable and effective over time.

The demand for tailored oversight and knowledge has also created new roles. For example, KPMG already assigns subject matter experts to serve as “AI quality evaluators,” who are responsible for implementing partially automated but primarily manual, human-driven checks to ensure the accuracy of AI responses.

In addition, we expect that “data and knowledge curators"—enhanced knowledge managers with AI duties—will collect, label, and regularly update the information used by AI.

With some initial workflows already in motion—and fine-tuning now recognized as crucial for superior performance—broad adoption of these specialized agents is likely by late 2026. At that time, both routine operations and complex issues within the tax and legal sectors will be managed with unprecedented accuracy and efficiency.

Targeting Turnkey Automation

Currently, AI automates some tax work, but heavy data tasks still require classic coded software. Because both AI prompts and classic software code are inflexible, each new project or workflow change requires major updates, holding back wider AI adoption. Furthermore, this approach limits AI’s ability to learn and improve over time.

In contrast, an agent-based model starts with a broad objective—rather than a fixed prompt—such as preparing a specific tax document. It leverages AI to learn from all available data, handle multiple tasks (even involving heavy data) iteratively, and improve efficiency over time.

Early tests show that this method can significantly reduce the need for human oversight and streamline workflows.

Until major technology providers offer universal AI platforms, the focus will remain on developing AI agents and supporting solutions. While local AI models will improve, cloud-based solutions are likely to dominate. Organizations should focus on designing and scaling robust AI agents. It’s important that the agents have access to the relevant knowledge and tools.

This article does not necessarily reflect the opinion of Bloomberg Industry Group, Inc., the publisher of Bloomberg Law, Bloomberg Tax, and Bloomberg Government, or its owners.

Author Information

Christian Stender is head of AI for Tax and Legal, KPMG International.

Eduard Seregin is manager, AI and product management, with KPMG Deutschland.

Write for Us: Author Guidelines