Tax Compliance with AI: How artificial intelligence supports an effective TCMS

What you can expect from this article

To many companies, a Tax Compliance Management System – or TCMS for short – initially sounds like extra paperwork, consultancy costs and new obligations. Where should you start? Who will be responsible for setting it up? And how can you even manage this alongside your day-to-day business? These and other questions lead many decision-makers to an obvious line of thought: Can AI help me? hsp Managing Director Paul Liese discussed this with Torsten Stockem – himself a partner and tax adviser at the firm Energiesozietät GmbH – on the programme „hsp live at 11“.

AI has made its mark in the tax sector – but practitioners remain cautious

In many sectors, AI tools have long been part of everyday working life. AI services are already widely used for research, drafting texts, marketing and sales. However, in the tax sector, concrete, productive use cases are still somewhat rarer. Whilst some decision-makers do test a chatbot here and there, their use often comes to an end quite quickly when it comes to prompting, tax analysis and actual integration into workflows. There are understandable reasons for this reluctance. Particularly in the public sector and where security is a concern, data protection, technical stability and the long-term reliability of providers are key considerations. Added to this is the fact that the AI market is developing at a rapid pace.

A tool that is considered the gold standard today may look very different tomorrow. On the one hand, start-ups with AI ideas are springing up everywhere; on the other, the tech giants are just waiting to copy good ideas. For the tax compliance management system (TCMS) sector, this means relying on AI tools that do not damage the TCMS itself if they are deactivated. Or to put it more simply: companies and law firms should not build their TCMS using an AI tool, but rather with specialist TCMS software that offers AI support.

A TCMS is more than just a means of avoiding liability

Tax compliance is often reduced to a means of avoiding potential liability. The focus is then frequently on Section 153 of the German Fiscal Code (AO), i.e. the correction of tax returns in the event of errors identified retrospectively. However, an effective TCMS should not only provide assistance once an error has already occurred or an external audit is imminent, but before that happens. In this way, a TCMS creates transparency regarding tax-relevant processes. It makes it clear who is responsible for which tasks, where decisions are made, what data is required and where errors may occur. From this, clear lines of responsibility, escalation procedures and control measures can be derived.

A good TCMS therefore helps to ensure legal certainty in tax matters. And it improves the tax function as a whole:

• Processes are described in a transparent and consistent manner
• Responsibilities and handover points are becoming clearer
• Tax risks are identified at an early stage rather than being discovered retrospectively
• Audits provide reliable insights into actual error rates
• The workload involved in external audits and recurring reconciliations may be reduced

This makes the TCMS a living system that needs to be maintained, adapted and further developed.

The most common mistake: looking at processes in too broad terms

Take purchasing as an example. „We have a purchasing process“ is a start, at least, but it is by no means a sufficient basis for tax compliance. After all, a process name does not explain what actually happens within the company. For tax assessment purposes, the process must be broken down into individual steps. After all, when it comes to procurement, it is not enough simply to look at the receipt of invoices. Instead, the entire chain should be considered: from the purchase requisition, through the purchase order, goods receipt and invoice verification, right through to posting and payment. Only when these steps are transparent can it be identified at which points tax risks may arise.

The same applies to sales. A complete process chain can begin with a customer enquiry and extend through to the quotation, order, delivery, invoicing and handover to the accounts department. Risks frequently arise, particularly at the interfaces between specialist departments, sales, logistics, accounts and the tax function. This is because information is often missing at these points, or responsibilities are not clearly defined. Whilst many companies do have general process documentation, for example as part of an ISO certification, However, tax-related aspects are often not adequately covered in such documentation. A tax process description must therefore take a closer look: What business transactions occur? Which types of tax may be affected? What information must be available? And who ensures that it is processed correctly?

How AI can help with process documentation

But how can artificial intelligence help to ease the workload in this area? For example, by enabling the specialist to use an integrated AI tool to generate an initial structured draft of a process description. For instance, one could have the invoice receipt process described and specifically enquire which roles are typically involved, which steps are included, and at which points tax-related issues may arise. This means nobody has to start from scratch anymore. But be careful: the result is not a finished set of procedural documents, but a working basis. This is because the content must always be checked against the reality within the organisation by the specialist. After all, no purchasing, sales or accounting process is completely identical to a model process.

If you want to achieve a more accurate result, provide the AI tool with more specific context. This may include specific business models, supply chains or contractual arrangements. Rather than simply asking „Describe the purchase“, the prompt should be as specific as possible:

• What are the steps in the process, from identifying a requirement to making a booking?
• Which people, departments and systems are involved?
• What tax issues might arise?
• What specific features apply to your own business model?
• What supporting documents and data must be available, and where?

AI therefore helps to establish an initial structure more quickly and to highlight blind spots. However, it does not replace the scrutiny carried out by real people.

From the process, through risks, to control measures

Once a process has been described in sufficient detail, the crucial step follows: the risk analysis. AI can also provide support here. For instance, it can scrutinise existing risk analyses, suggest typical risk areas for a process and structure potential control measures. In addition, data extraction and analysis tools can be used to test assumptions against actual data. A simple example: the VAT registration number for new customers. One possible control could check whether the VAT registration number is recorded for the relevant customers and is plausible. In the event of irregularities, the company can make corrections, investigate the causes and improve the process. If controls are carried out regularly, this also creates a robust data basis reflecting the actual error rate.

It all sounds impressive, which gives some experts pause for thought. Yet the fear of being replaced is unfounded. This is because AI generates results based on probabilities, but it can neither assess nor weigh up options, nor make decisions. Nor can it walk around the office to check whether all the descriptions match reality. AI support will, however, mean that the role of tax professionals changes. In future, far fewer drafts will be prepared manually. Instead, there will be more time for technical review or the development of alternative solutions. Even though the work will become more efficient, responsibility for the correct configuration of the TCMS will remain with the relevant specialists in future.

No stand-alone solutions: tax compliance must be part of everyday life

Anyone wishing to create and implement an effective TCMS should avoid stand-alone solutions. An Excel spreadsheet with risk assessments here, a BPMN process visualisation there, and process descriptions scattered all over the place. Chaos is inevitable. And AI tools can only ever access individual fragments of the TCMS, which inevitably leads to errors. If a TCMS is to be effective holistically and in the long term, it must be integrated into the existing system and process landscape. This includes regular risk assessment, the documentation of controls, and the identification and rectification of vulnerabilities. If a business model, an IT system, a supply chain or a legal requirement changes, the TCMS must incorporate this change.

It is understandable that companies are initially put off by the costs. However, Paul Liese and Torsten Stockem recommend taking a long-term economic view. The costs involved in the initial roll-out – for design, documentation, consultancy and technical implementation – are investments. These will pay for themselves many times over in the long run: through reduced time spent searching for information, clearer lines of responsibility, better data quality, fewer discussions and more efficient external audits. And anyone who chooses not to invest in this should bear one thing in mind: their competitors will not be passing up these advantages.