The End of the Tax Dispute? Technology, Data and the Limits of Automation
The idea that tax disputes might diminish, or even disappear, has accompanied each major development in tax administration.
The introduction of self-assessment was intended to improve accuracy by placing greater responsibility on taxpayers. HMRC’s Connect system, in use since 2007, promised to identify discrepancies through large-scale data analysis. More recently, third-party reporting obligations and the rollout of Making Tax Digital have been designed to close information gaps and enable near real-time intervention.
Each of these developments has been driven by technology. Each has improved visibility and enhanced detection. None, however, has eliminated the conditions in which disputes arise.
With the expansion of real-time reporting regimes and the rapid development of artificial intelligence (AI), that longstanding expectation has regained momentum. Tax authorities now have access to volumes of data that would have been unthinkable even a decade ago. Patterns can be identified more quickly, anomalies flagged earlier, and discrepancies addressed closer to the point at which they arise.
It is tempting to conclude that the conditions for traditional tax disputes are beginning to erode.
The difficulty with that conclusion is not that technology has failed, but that tax disputes are not primarily driven by missing information. They arise because the same information can be interpreted in different ways.
Technology can reduce information asymmetry. Tax disputes, however, arise from interpretive asymmetry.
The availability of increasingly granular data may also create an appearance of precision that exceeds what the law itself can support. Tax outcomes often depend not only on factual accuracy, but on evaluative concepts that remain inherently contestable.
The promise of real-time visibility
The potential impact of improved data and analytics should not be understated. Real-time reporting and digitalisation reduce reliance on retrospective review and provide tax authorities with a more immediate view of taxpayer activity. They allow for earlier engagement, which in some cases may prevent issues from escalating.
For taxpayers, there is a corresponding benefit. The temporal gap between commercial activity and tax reporting is reduced, allowing discrepancies to be identified closer to source and addressed more efficiently.
From the tax authority perspective, the ability to process large datasets and identify patterns across taxpayers allows AI-driven tools to detect anomalies that may previously have gone unnoticed. This has the potential to increase consistency of approach and reduce the scope for purely administrative error.
However, increased visibility does not eliminate institutional constraints. The ability to process data at scale does not necessarily translate into an equivalent increase in capacity to exercise judgment.
Automation and the UK penalties insight
The practical limits of automation are perhaps most clearly illustrated by HMRC’s approach to penalties.
In the UK, more than 9 million automated penalties, penalty points and surcharges were issued in 2024–25. These decisions were generated at scale, with minimal human intervention. Accuracy is introduced downstream through statutory review and appeal processes. The effect is striking. A substantial proportion of automated penalty decisions do not survive review. In cases involving automated penalties and default surcharges, approximately 33% of statutory reviews upheld the original decision, 67% resulted in cancellation, and fewer than 1% were varied once human judgment was applied.[1]
This is not evidence of a fundamentally flawed system. Rather, it reflects a different institutional design choice.
Automation prioritises speed, scale and consistency. Human review reintroduces judgment only once the taxpayer engages. In that sense, the review process operates less as an exceptional safeguard and more as a form of large-scale quality control.
This has two important implications.
First, technology may change when errors are identified rather than eliminate them altogether. Issues are surfaced earlier, but not necessarily resolved more accurately at the outset.
Secondly, increased data and automation may generate more disputes, not fewer. Where automated processes produce large volumes of initial decisions, a corresponding volume of challenges is likely to follow.
The result is a system in which visibility increases, but the need for engagement, and therefore the potential for dispute, remains.
Data does not eliminate disagreement
Many of the most significant tax disputes arise in situations where the facts are broadly understood, and the disagreement lies in how those facts should be interpreted.
This is particularly evident in areas such as:
the classification of transactions or supplies;
the application of special regimes or reliefs;
transfer pricing and the attribution of value;
the application of anti-avoidance principles; and
questions of economic substance and purpose.
In each of these areas, additional information does not necessarily reduce uncertainty. In some cases, it may increase the number of plausible interpretations.
A real-time reporting system may identify that a particular transaction has taken place. It is far less well placed to determine, as a matter of law, how that transaction should properly be characterised.
The distinction is important. Data can establish that an event occurred; it cannot, by itself, resolve competing interpretations of its legal or economic meaning.
Technology is particularly effective at automating compliance processes. It is materially less effective at resolving legal ambiguity.
The persistence of interpretive risk
Tax law, particularly in complex or cross-border contexts, often requires evaluative judgment. Courts and tribunals are frequently asked not simply to apply a rule, but to determine how a rule should be understood in light of its purpose, context and interaction with other provisions.
This is not a purely mechanical exercise. It involves:
weighing competing interpretations;
assessing the relevance of different factual elements;
considering the broader statutory framework, including its legislative history and policy intent; and
in some cases, balancing legal form against economic substance.
AI can assist in analysing large volumes of material. It can identify patterns in previous decisions and suggest possible lines of argument. It may also improve consistency in routine or highly standardised cases. But it does not remove the need for judgment. Nor does it eliminate the possibility that different decision-makers will reach different conclusions on the same set of facts. The consequence is that a significant category of tax risk remains inherently resistant to automation.
The current state of AI in practice
The gap between the theoretical potential of AI and its current practical application is also significant.
Recent experience across the broader legal profession has highlighted limitations at the intersection of human oversight and machine output. AI systems can produce plausible but incorrect legal content, including fabricated authorities or mischaracterised - or even hallucinated - statutory provisions.
Similar challenges arise within the tax sphere. Automated outputs may lack coherence, and draft communications generated by AI tools often require substantial revision before they can be relied upon. Outcomes can also vary significantly depending on how queries are framed and interpreted.
These may be transitional issues, which will no doubt improve over time. However, they reinforce a broader point. Technology does not operate in isolation. Its effectiveness depends on the quality of inputs, the design of systems and the judgment of those using it.
Where those elements are imperfect, automation can amplify inconsistency rather than resolve it.
Behavioural dynamics remain unchanged
Even if the technical limitations of AI were fully overcome, a further constraint would remain.
Tax disputes are not determined solely by the availability of information or the mechanics of analysis. They are also shaped by behaviour.
Tax authorities make decisions about:
which issues to pursue;
how extensively to investigate them;
when to escalate a matter; and
whether to abandon, settle or litigate.
Those decisions are influenced by resource allocation, policy priorities and the perceived importance of particular issues. Improved data does not eliminate those institutional dynamics. Automated systems inevitably reflect the assumptions embedded within them, including which behaviours are treated as indicators of risk. As a result, technology may influence not only the efficiency of enforcement, but also the profile of disputes that emerge.
From the taxpayer perspective, decisions about whether to engage, concede or contest are shaped by commercial priorities, risk appetite and the wider context in which the dispute arises.
In that context, tools that help taxpayers manage the cost and uncertainty of disputes become increasingly important. Tax insurance, for example, can enable taxpayers to engage more confidently with HMRC processes rather than defaulting to early concession in response to procedural pressure. By mitigating financial exposure and reducing downside risk, insurance may influence not only the economic consequences of a dispute, but also how it is approached and navigated.
What is likely to change
Technology is likely to change the profile of disputes in several important ways.
Issues will be identified earlier. The lag between transactions and regulatory scrutiny will continue to narrow.
The volume of enquiries may also increase. With more data available and more anomalies identified, authorities are likely to engage more frequently, even if individual issues are narrower in scope.
The early stages of disputes may become increasingly standardised. Automated processes are likely to shape how issues are raised, how risks are categorised, and how initial information requests are formulated.
These developments may shorten some disputes, particularly those arising from straightforward errors, omissions or inconsistencies. They are, however, less likely to affect the complex interpretive disputes that tend to define the upper end of tax controversy. In those cases, where uncertainty persists and outcomes remain finely balanced, tools such as tax insurance may become increasingly relevant as mechanisms for managing risk rather than eliminating it.
What is unlikely to change
At a structural level, the underlying drivers of significant tax disputes remain largely unchanged.
Disputes arise where:
the law is capable of more than one interpretation;
the facts can be characterised in different ways;
the financial or precedential stakes justify detailed scrutiny; and
the parties involved have different incentives in resolving the issue.
None of these conditions is removed by improved data or more sophisticated technology.
In some cases, increased transparency may instead bring a greater number of issues into focus, particularly in areas where legal frameworks have not fully kept pace with commercial or technological developments.
In those contexts, the limiting factor is not information. It is the absence of settled legal principles capable of resolving uncertainty with consistency and predictability.
A familiar conclusion
The recurring prediction that tax disputes will disappear has not yet proved correct.
The current wave of technological change is more advanced than those that preceded it. Data is more abundant. Analytical tools are more powerful. Automation is more deeply embedded. But these developments change when disputes arise, and how they are managed. They do not eliminate the underlying reasons why they occur.
In a system where uncertainty cannot yet be engineered away, the ability to manage that uncertainty, through governance, strategy and, increasingly, mechanisms such as tax insurance, becomes as important as resolving it.
AI will nevertheless have an important role to play. Tax authorities, advisers, taxpayers and insurers are all increasingly using AI-assisted tools to streamline analysis, improve efficiency and support strategic decision-making. The future of tax controversy is unlikely to be dispute-free. It is more likely to be faster, more data-driven and more strategically complex.
—
Paula Ruffell
Senior Underwriter
Toremis Specialty
[1] HMRC Annual Report and Accounts 2024–25.