From Automation To Autonomy: How Agentic AI Is Transforming Accounting

_{This article was originally posted on Forbes.com, November 2024.}

I recently attended the EU Accounting Summit in Düsseldorf where the key topic was the role of AI in the future of accounting. The buzz was around agentic AI, and I was asked to define it numerous times.

I think of agentic AI as a supercharged assistant that has all the knowledge in the world at its fingertips. It has been trained on all publicly available, along with plenty of privately owned, data. You can feed it thousands of pages of instructions, millions of datasets of history, in any language, and it will understand all of it instantly.

The big difference with agentic AI, as opposed to generative AI, for example, is that it can act independently to do something with all that information.

For example, to get results, you just need to provide an intent, and the agentic AI can plan how to do the task. It can examine all the facts, access necessary data from your ERP system and then take action. It has the tools to do the tasks and can autonomously perform what you ask it to do.

The Impact Of Agentic AI

One thing that struck me at the conference is that people’s perception of what is possible in the future is still based on their current experience. People believe future tools will just be more efficient to drive more productivity. They are not imagining completely new scenarios that may be possible thanks to agentic AI.

There is already a lot of excitement about automation in accounting. Advanced technology can optimize invoice processing and improve overall data quality. It can process and analyze huge datasets and identify patterns and trends. It can also detect discrepancies and prevent illegal financial transactions. With all these advancements already in place, I believe it’s time to examine AI’s ability to handle complex tasks such as tax assessments and apply it to completely new realms.

What Will Agentic AI Look Like?

Agentic AI does not follow predefined rules or focus on creating new content. It emphasizes goal-oriented behavior and adaptive decision making, using advanced algorithms and sensory inputs to execute actions in real time. It’s always learning and optimizing its performance through continuous feedback.

Currently, agentic AI that can act autonomously, make decisions and execute tasks without requiring constant human input remains in its early stages for specific applications such as accounting. There are some solutions on the market that leverage AI for predictive analytics and process automation but are designed as assistive technologies, enhancing the work of accountants rather than replacing decision making entirely.

So, what are some accounting tasks that agentic AI could take over in the future?

For example, as a highly skilled tax expert, you may have to do a complex withholding tax assessment in Indonesia. You must take all the required Indonesian tax regulations into account, check the relevant business event, understand all the documents as well as all the tax regulations and adhere to all the company’s practices.

You might encounter several problems along the way. You may have to interpret tax treaties to apply the correct withholding tax rates for international payments such as dividends, interest and royalties. Tax treaties, however, often contain vague language or outdated provisions that may not align with modern business. In addition, tax treaties can conflict with domestic laws or regulations.

Or, you may have to provide documentation of withholding tax assessments and ensure timely filing and reporting to tax authorities. But, managing complex documentation such as forms to claim treaty benefits can be difficult.

The list of problems goes on, and there may even be some you haven’t thought of yet. Most people think a machine couldn’t possibly solve a problem that a human can’t conceptualize, but in reality, it can.

The caveat, of course, is that you need to provide all necessary information and tell the AI what it needs to do correctly. You can then harness its power. Once you understand that even complicated expert tasks can be fulfilled by an AI agent with ease, you can start thinking about what that actually means for your organization.

The Future Of Electronic Invoicing

Electronic invoicing has gained global attention because it enables greater transparency, especially in tax compliance. Many governments are even mandating businesses to submit invoices through government portals in a standardized digital format. This is particularly helpful for tax reporting, as it reduces errors and helps authorities monitor transactions in real time.

However, the transition isn’t always easy for companies. If you’re a global company, you may have to handle up to 60 different e-invoicing formats, as every country has its own rules and regulations. This complexity adds operational overhead, which traditional systems aren’t equipped to manage effectively. Agentic AI can help; it can seamlessly handle diverse input formats—whether PDFs, scans or global e-invoicing standards—by leveraging LLMs for flexible data extraction.

In essence, agentic AI is transforming accounting by introducing autonomy and intelligence into processes that were once manual or semiautomated. This shift allows for greater efficiency, improved accuracy and the capacity to handle complex tasks that goes beyond the reach of traditional automation.

Best Practices For Implementing Agentic AI

Some best practices for using agentic AI in accounting include:

• Establish clear human oversight and control mechanisms with human-in-the-loop processes to reduce risks and maintain complete compliance.

• Ensure robust data governance and security to deal with vast amounts of sensitive financial data.

• Maintain traceability and auditability to build trust and facilitate regulatory scrutiny.

Starting an AI implementation should be tackled in the same way as any major software implementation. Begin by establishing clear goals and objectives for what you want to achieve, with a detailed project scope and engagement plan. Next, make sure you have the highest data quality. Low-quality data can have serious repercussions on financial processes, leading to inaccuracies that harm your business. AI algorithms are powerful tools, but their performance is strictly related to the information they receive. And lastly, you need to assemble a skilled project team of trained experts and develop a clear change management plan.

‍