Artificial intelligence is rapidly reshaping the fintech landscape. Yet amid the excitement surrounding automation, it is important to separate genuine opportunity from unrealistic expectations. The question is no longer whether AI can add value within regulated financial markets, but where it can do so safely, effectively and sustainably.
From my perspective working in withholding tax (WHT) technology, the evidence increasingly points towards a more measured conclusion: AI delivers the greatest value when it augments human expertise rather than attempts to replace it entirely.
The automation temptation – and its hidden fragility
On the surface, the business case for AI appears compelling. Faster processing, reduced manual intervention and lower operational costs are attractive objectives for any financial institution, and are fuelling adoption[i].
However, many highly regulated financial processes operate within environments characterised by fragmented data, evolving regulation and stringent evidential requirements[ii]. WHT reclamation is one example, but the same challenges appear across client onboarding and KYC, trade surveillance, corporate actions processing, collateral management, fund accounting and complex credit decisioning.
The underlying technical challenge is often data quality[iii]. WHT workflows typically combine information from custody systems, portfolio accounting platforms, transfer agents and large volumes of unstructured documentation, including tax vouchers and residency certificates. These sources rarely share consistent formats or standards.
As a result, AI systems remain vulnerable to the classic “garbage in, garbage out” problem. Even highly sophisticated models can generate confident but incorrect conclusions when they are trained on incomplete ownership records, inconsistent documentation or outdated tax information. In highly regulated environments, accuracy is only ever as good as the underlying data.
Regulatory complexity moves faster than static models
A second challenge is the extraordinary complexity of cross-border regulation.
In withholding tax alone, eligibility for relief or reclaim can vary according to jurisdiction, security type, investor status, treaty provisions and local administrative practice. Similar complexity exists throughout financial services, where rules frequently differ across markets and continue to evolve.
The challenge for AI developers is that regulatory frameworks are dynamic rather than fixed. New initiatives, including the European Union’s FASTER framework and national reforms across multiple jurisdictions, continue to reshape reporting requirements, documentation standards and evidential expectations.
This creates a model-maintenance challenge that is often underestimated. Building an AI model is one task; ensuring that it remains aligned with continuously changing legal and regulatory requirements is another. In many cases, the cost and governance burden associated with maintaining fully autonomous systems can significantly reduce the economic benefits originally anticipated. Especially given that in this sphere, mistakes come with very severe consequences.
Explainability remains the make-or-break factor
For many financial institutions, explainability may ultimately be the most important constraint on AI adoption[iv].
Tax authorities, regulators, auditors and courts generally expect organisations to demonstrate how a decision was reached. A conclusion alone is rarely sufficient; there must also be a clear and defensible audit trail.
This becomes problematic when organisations rely on opaque or “black-box” AI models[v]. If a system determines treaty eligibility, identifies a tax risk or recommends a filing position but cannot clearly demonstrate which data points, rules or legal provisions informed that conclusion, the result may struggle to satisfy governance, audit or regulatory requirements.
International organisations such as the OECD have highlighted transparency, explainability and accountability as central requirements for trustworthy AI systems, particularly where decisions affect legal rights or obligations[vi]. The ability to understand and challenge automated decisions remains fundamental to due process and good governance.
Low-frequency errors can create high-impact outcomes
Unlike many consumer fintech applications, withholding tax operations involve a different kind of risk.
Although errors are rare, their impact is substantial. Research found that mistakes and inefficiencies in withholding tax recovery can have a material impact on the flow of capital across markets and investment outcomes.
For custodians and asset managers overseeing large institutional portfolios, even a very low error rate can result in important financial exposure when applied across millions of positions.
Over-claiming can lead to penalties, regulatory attention and reputational damage. Under-claiming can be just as costly, resulting in unrecovered tax, lower returns for investors and potential fiduciary challenges.
Data governance and cross-border constraints
Data governance presents another important consideration.
Tax information is among the most sensitive categories of financial data. As financial institutions increasingly explore AI-enabled workflows, they must also navigate complex requirements relating to privacy, data protection, operational resilience and outsourcing oversight.
For organisations operating across the UK and Europe, transmitting investor-level tax information to external AI providers may create additional governance obligations under GDPR and other regulatory frameworks. These considerations do not prevent AI adoption, but they do reinforce the need for careful architectural design, strong controls and appropriate human oversight.
Where AI delivers measurable value today
None of this suggests that AI lacks an important role in financial services or withholding tax operations. In fact, some of the most valuable use cases are already proving their worth.
Intelligent document processing is perhaps the clearest example. AI can classify incoming documentation, extract key fields and convert unstructured information into standardised data for human review. This can significantly reduce operational effort while preserving appropriate controls.
AI also shows considerable promise in data-quality management. Anomaly detection techniques can identify missing residency information, treaty inconsistencies, duplicate claims and other exceptions that would otherwise require extensive manual review.
Knowledge augmentation represents another powerful application. By combining curated treaty libraries, regulatory guidance and internal expertise, firms can create research tools that help specialists navigate complex requirements more efficiently without removing human judgement from the process.
Operational optimisation offers further opportunities. AI can support workflow prioritisation, identify claims at risk of approaching statutory deadlines, forecast recovery timelines and model the potential impact of regulatory changes.
These applications enhance human capability rather than replacing it, which is precisely why they often achieve stronger results[vii].
A pragmatic path forward
The fintech industry’s conversation around AI should move beyond simplistic narratives of full automation.
In highly regulated environments, fragmented data, evolving rules, explainability requirements and the potentially significant impact of errors all create meaningful barriers to fully autonomous decision-making.
That does not diminish AI’s value. On the contrary, organisations that deploy AI selectively and strategically – focusing on data ingestion, quality assurance, expert support and workflow optimisation – can unlock substantial efficiency gains while maintaining the governance, transparency and accountability that regulators and clients increasingly expect[viii].
The most successful applications of AI in fintech may not be those that seek to replace experts, but those that make experts significantly more effective.
[i] https://www.bankofengland.co.uk/report/2024/artificial-intelligence-in-uk-financial-services-2024
[ii] https://www.imf.org/en/news/articles/2024/09/06/sp090624-artificial-intelligence-and-its-impact-on-financial-markets-and-financial-stability
[iii] https://www.nist.gov/itl/ai-risk-management-framework
[iv] https://www.oecd.org/en/publications/governing-with-artificial-intelligence_795de142-en/full-report/ai-in-tax-administration_30724e43.html
[v] https://www.nature.com/articles/s41599-025-06099-7
[vi] https://oecd.ai/en/gov/issues/tax-administration
[vii] https://arxiv.org/abs/2504.20086
[viii] https://www.bankofengland.co.uk/financial-stability-in-focus/2025/april-2025
