Artificial intelligence offers meaningful efficiency in withholding tax recovery, but expertise remains critical. Both elements can (and must) work together.
AI is reshaping how financial services manage information and process data. In the complex world of withholding tax recovery, its appeal is obvious. The reclaim process involves fragmented documents, multiple jurisdictions and short statutory deadlines. For fund managers with significant cross-border exposure, these factors can quickly become a source of hidden value leakage.
Where AI proves useful
AI is well-suited to tackling the volume and variety of data that slows recovery. Optical character recognition (OCR) can allow thousands of client and custodian documents to be processed in seconds, across formats that previously required painstaking manual entry. Natural language processing can read and interpret clauses in treaties, certificates or tax authority notices, highlighting relevant points for further review. This not only accelerates preparation but reduces standard error rates associated with manual handling.
AI also supports better oversight. Automated systems can scan global case law, regulatory changes and treaty amendments, flagging developments that may affect portfolios. In practice, this means reclaim opportunities can be identified earlier, reducing the risk that they expire unnoticed. For investors operating across multiple jurisdictions, this proactive intelligence can help transform recovery from a reactive task into a structured part of portfolio management.
Why human expertise still matters
The efficiency gains are real, but the limitations are equally clear. Whether or not a Luxembourg SICAV qualifies under a treaty exemption, for example, or how to frame a comprehensive response to a tax authority query, is not best determined based on large language models (LLM) and their probabilistic outcomes; deterministic outcomes prevail. Queries, for example, often arrive with extremely short response deadlines and require precise, justified and auditable documentation and argumentation.
Across the industry, claims that are pushed quickly through pro forma automation modules can end in rejection if responses do not meet procedural requirements. In withholding tax recovery, rejection risk is one of the most significant threats to value preservation.
An applied model
A leading example of how I’ve seen AI effectively integrated with domain expertise is in the development of a claim maximization engine for withholding tax recovery. This system evaluates thousands of potential reclaim scenarios for each dividend event, optimizing outcomes through advanced algorithmic analysis. These outputs are then subjected to human reviews. When effectively combined with humans in the loop, AI can enhance processing speed while maintaining high standards of accuracy and oversight.
The lesson for managers
The real question is not whether AI should be used or not, but how it should be used. Which tasks are automated, how errors are prevented and where human review enters the process will determine if this technology delivers genuine improvements or simply shifts the risk elsewhere. In a market where governance standards, audits and detailed reporting carry weight, transparency around process is as important as speed.
From my point of view, AI has already changed what is possible in withholding tax recovery. Yet, as with investing itself, technology achieves the best results when it supports, rather than replaces, informed human judgment.
Reuben John is managing director, DACH, UK & IE, at WTax. The firm is a member of Investment Officer’s panel of experts.