AI has become one of the most discussed technologies in financial services, with growing expectations that it can automate not just admin tasks but complex decision-making too.
In capital markets, that promise looks especially attractive, offering lower costs, faster processing and greater scalability. But financial services remains one of the most heavily regulated, risk-sensitive sectors globally, and the idea that AI can simply take over critical processes deserves scrutiny.
Building TaxTec’s withholding tax (WHT) reclaim platform has exposed both the opportunities and dangers of AI adoption in this space.
The first challenge is data. WHT reclaim processing pulls information from custody systems, portfolio accounting platforms, transfer agents and documents including tax vouchers and residency certificates, much of it fragmented and inconsistent.
Feed an AI model incomplete or inaccurate data, and it can produce confidently wrong reclaim calculations, an old IT warning that now carries far greater scale and risk.
Complexity compounds the problem. Withholding tax rules vary by jurisdiction, security type and treaty entitlement, and are constantly evolving, as seen with the EU’s FASTER framework and Germany’s new digital filing requirements. An AI system trained on today’s rules can quickly become outdated.
Explainability is another concern: regulators expect decisions backed by clear legal reasoning and an audit trail, something AI models often cannot reliably provide. With regulators including the FCA and frameworks like the EU AI Act increasingly focused on model governance and conduct risk, fully autonomous AI agents determining treaty eligibility remain, in practice, a pipe dream.
Given the scale of sums involved, even small error rates create significant exposure, whether through over-claiming penalties or under-claiming tax leakage. Data governance adds a further layer of difficulty, given how sensitive tax information is under privacy and cross-border transfer rules.
None of this means AI has no role. Used as a decision-support tool rather than a decision-maker, AI performs well at document ingestion, extracting details such as ISINs, payment dates and tax rates from reclaim documentation.
It can also flag data quality issues, missing beneficial ownership information, mismatched custody and tax voucher records, and inconsistent treaty rates, mirroring how tax authorities themselves increasingly use AI to detect anomalies.
As a research assistant, AI helps specialists navigate treaties and regulatory guidance faster, while in workflow management it can prioritise claims by value and deadline. As digital filing systems such as Germany’s BZSt BOP interface expand, AI will also help map data to required formats and interpret rejection notices, with human teams retaining final sign-off.
Read the full TaxTec post here.
Copyright © 2026 RegTech Analyst
Copyright © 2026 RegTech Analyst
