AI in Tax Advisory: Why Disruption Comes from the tax advisory firm

Accounting will automate, prices will fall, and the firm as we know it will disappear. Something like that is the standard narrative when people talk about the future of tax advisory. It draws on Clayton Christensen's theory of disruptive innovation and likes to point to Uber, Airbnb, or Amazon as evidence that no profession is safe from technological disruption.

The theory is convincing. The problem is its transferability. Because tax advisory has structural peculiarities that make textbook disruption significantly more difficult. And the question nobody answers satisfactorily is: through which platform is this supposed to happen in concrete terms?

Why tax advisory is not a taxi industry

Disruption according to Christensen typically works like this: a new entrant offers a simpler, cheaper solution that is initially attractive only to the lower end of the market. Over time, the product improves and displaces the established providers. Uber needed no taxi license. Airbnb needed no hotel building. Both used technology to take an existing market apart from the outside.

In tax advisory, that is more difficult for several reasons. First: the service is subject to the German Tax Advisory Act. Reserved tasks such as preparing tax returns and annual financial statements may only be performed by licensed tax advisers. A pure software company can offer bookkeeping, but not tax advice in the narrower sense. Second: the client relationship is not a transaction business. A tax adviser supports a company over many years and knows the ownership structure, industry specifics, and the management's personal circumstances. This depth of context cannot be packed into a software interface. Third: the German market is shaped by DATEV as the system of record. More than 90 percent of firms work with DATEV. That creates an infrastructure that cannot simply be bypassed, but one into which every technological solution must fit.

That does not mean nothing will change. It means that change will look different from the familiar disruption narratives.

Scenario 1: full-stack companies that bypass DATEV

The most ambitious vision is pursued by companies that want to offer both software and service from a single source. They build their own bookkeeping platforms, onboard clients, and promise a fully digital experience without DATEV dependency.

The concept has appeal. If you control the entire value chain, efficiency can theoretically be maximized. In practice, however, hard limits become apparent. Most of these providers have no licensed tax advisers on their team, or only a few. That means: for pure bookkeeping and preparatory tasks, that is enough, but for everything that falls under the reserved tasks, tax returns, appeals, structuring advice, the client additionally needs a tax adviser. The full-stack provider becomes an intermediate step, not a replacement.

There is also a practical problem: if the client eventually does need or want a traditional tax adviser, they have to migrate out of the proprietary system. That creates lock-in, which many entrepreneurs deliberately want to avoid.

Scenario 2: AI tools at the client with DATEV integration

A more pragmatic approach is tools that introduce AI-supported bookkeeping directly at the client, but remain connected to DATEV. Providers like Countful, with their AI agent Tess, are pursuing exactly this path: documents are captured, booking entries are created automatically, follow-up questions are sent to the client by AI, and the finished data flows into DATEV Kanzlei-Rechnungswesen.

That is technologically impressive and solves a real problem: the time-consuming preparation of documents, which in many firms still runs manually or semi-automatically. The bottleneck lies elsewhere. These tools introduce a new frontend at the client. The client works in a new interface, uploads documents there, and interacts with the AI. That works well for clients who do not yet have a functioning digital system or are willing to switch.

But the hundreds of thousands of clients already connected via DATEV Unternehmen Online, whose upstream systems (bank accounts, invoicing software, credit card management) deliver clean data through interfaces, have little reason to add another system in between. For them, there is no added value, only additional complexity.

The adoption question is therefore similar to the client portal: the tool is only as strong as the client's willingness to use it. And that willingness declines the better the existing setup already works.

Scenario 3: AI within the firm

The third scenario is the least visible, but possibly the most effective. Instead of moving the client to a new frontend, AI is used where the data already comes together: inside the firm itself.

With the Automatisierungsservice Rechnungen (ASR) and the Automatisierungsservice Bank, DATEV already has its own AI solutions on the market. ASR analyzes digitized documents and generates booking proposals based on historical booking data. More than 5,500 firms use the service for more than 56,000 client portfolios. The numbers show that the approach works at scale.

However, ASR works with classic machine learning based on pattern recognition in historical data. What is possible with that has limits. Complex cases that require contextual understanding, such as distinguishing between a down payment invoice and a final invoice, interpreting an unusual payment reference on the bank statement, or recognizing that an invoice must be treated differently for tax purposes than it appears at first glance, cannot be solved by pattern matching alone.

That is exactly where the potential of large language models lies. LLMs can not only classify documents, but understand them semantically. They can read a payment reference and interpret it, place a document in the context of earlier business transactions, and formulate follow-up questions that go beyond a simple "document missing." Tools like Finmatics already combine LLM-based line-item recognition with DATEV integration and show what is possible in this layer.

The decisive advantage of this approach: the clients' existing upstream systems remain untouched. The bank account, invoicing software, credit card management, everything continues to deliver data through the configured DATEV interfaces. The AI sits between the arrival of the data in the firm and the final booking in the accounting system. The client hardly notices, but benefits from faster processing and more proactive follow-up questions.

Fully automated bookings are only the beginning

In the debate about AI in bookkeeping, people often act as if automatic booking were the end state of automation. That is a thinking error. A booking is a deterministic assignment: document X belongs to account Y with tax code Z. That can be automated, with ML or with LLMs.

But after booking, the non-deterministic decisions begin. Is this client's cost structure plausible? Should a provision be recognized? Are there tax optimization opportunities that can be seen from the booking pattern? Has the VAT treatment of a specific type of transaction changed due to a recent BMF letter?

These questions require professional judgment, client knowledge, and tax expertise. AI can support here: flag conspicuous patterns, show deviations from the previous year, surface relevant legal changes. But the tax adviser makes the decision. That is exactly why the direct contact person for the client remains essential. German tax law is too complex, too dependent on the situation, and too consequential for a client to rely solely on software.

The SME is connected in the firm

There is a reason why none of the three options has transformed the market so far: the existing infrastructure is stronger than any individual tool.

The overwhelming majority of German SMEs are connected to DATEV through a tax firm. The upstream systems deliver data, the firm processes it, and the client receives its management report, annual financial statements, and tax return. This system is not elegant, but it works. And it has a property that is often underestimated: it is extremely sticky. A client rarely changes tax advisers, and even more rarely changes accounting systems.

Anyone who wants to use AI broadly in tax advisory must therefore start where SMEs are already connected: in the firm. Not with a new portal that the client is supposed to use. Not with a new frontend that replaces existing systems. But with technology that speeds up and improves the processing workflow inside the firm, without requiring the client to change their setup.

Where it is actually going

The future will probably not be either-or. For existing clients with functioning systems, the change will happen inside the firm: better automation, AI-supported review, faster processing, more proactive advice. The client will notice the difference not in a new login, but in shorter turnaround times and more substantive analyses.

For new clients who do not yet have a setup or are willing to switch, client-side AI tools will play a role. A startup founded today that does not have a pendafolder can start directly on a modern frontend. But these clients also still need a tax adviser in the end who reviews the data, prepares the annual financial statements, and handles tax structuring.

The actual disruption in tax advisory will therefore not come from a single technology company rolling up the market from the outside. It will come from firms that integrate AI into their processes in such a way that they can serve significantly more clients with the same team size, and better than before. That is less spectacular than the Uber analogy. But it is realistic. And it has already begun.