Why full AI automation in tax reclaims remains risky

AI has become one of the most discussed technologies in financial services, with expectations growing that it can automate not just administrative tasks but increasingly complex decision-making too

In capital markets, that promise looks attractive at first glance, offering lower costs, faster processing and greater scalability. Yet financial services remains one of the most heavily regulated and risk-sensitive industries globally, meaning the assumption that AI can simply take over critical processes deserves closer scrutiny

Insight from TaxTec’s development of its withholding tax (WHT) reclaim platform illustrates both the promise and the pitfalls of AI adoption. WHT reclaim processing draws on fragmented, non-standardised data from custody systems, portfolio accounting platforms, transfer agents and supporting documentation such as tax vouchers and certificates of residence

Where an AI model is trained on incomplete or inconsistent inputs, it risks producing confidently wrong reclaim calculations, a modern echo of the old “garbage in, garbage out” warning, now amplified by the scale at which AI can generate errors

Compounding this is the complexity of the global tax landscape, where withholding rules vary by jurisdiction, security type and investor classification, and are constantly reshaped by initiatives such as the EU’s FASTER framework and Germany’s digital filing reforms

Explainability presents a further obstacle, since regulators and tax authorities expect decisions to be backed by clear legal reasoning and an auditable trail, something opaque AI models struggle to provide

With regulators including the FCA and supervisors under the EU AI Act sharpening their focus on model governance and customer outcomes, and with reclaim errors carrying the risk of substantial financial exposure, sensitive tax data also demands strong governance rather than casual connection to public cloud-based models

Despite these constraints, AI still has a meaningful role as a decision-support tool rather than a decision-replacement one. It performs strongly in document ingestion and data extraction, classifying tax vouchers and identifying details such as ISINs, payment dates and tax rates for human review

It can also flag data-quality issues, including missing beneficial ownership information or treaty rate mismatches, mirroring how tax authorities themselves use AI to spot irregularities

Elsewhere, AI acts as a research assistant navigating treaty text and regulatory guidance, supports workflow prioritisation based on claim value and limitation periods, and helps teams manage increasingly complex electronic filing portals such as Germany’s BZSt BOP interface

Read the daily FinTech news

Copyright © 2026 FinTech Global

Investors

The following investor(s) were tagged in this article

Source: fintech.global