Artificial intelligence promises to transform financial services compliance by replacing periodic reviews with continuous risk monitoring. However, Suresh Sankaran of NatWest Group warns that operational barriers, weak governance and excessive reliance on automation could undermine its effectiveness. He argues that successful AI adoption requires stronger accountability, reliable validation and a clear distinction between machine intelligence and human judgment.
Artificial intelligence could
fundamentally transform how financial institutions identify and manage
compliance risks, but the greatest obstacles to effective implementation are
organizational rather than technological, according to Suresh Sankaran, Prudential
Regulation Lead at NatWest Group.
As banks move from periodic
compliance reviews toward continuous monitoring, AI offers opportunities to
detect emerging threats earlier, improve controls testing and strengthen
regulatory oversight.
However, Sankaran argues that
implementing these capabilities requires more than deploying sophisticated
technology.
"The biggest challenge is rarely
the technology itself. It is the operating model and the mindset shift,"
he explains.
Traditional compliance and assurance
functions rely heavily on periodic reviews, sample testing and retrospective
assessments. Continuous monitoring challenges these established practices by
providing near real-time visibility into risks and control effectiveness.
Yet fragmented control environments,
inconsistent data, unclear ownership of alerts and remediation activities, and
excessive notifications can undermine implementation.
Cultural resistance presents further
difficulties, particularly when employees fear automation could replace
professional judgment or expose weaknesses previously obscured by periodic
assessments.
Sankaran believes successful
organizations must position AI as a tool that strengthens professional
capabilities rather than replaces them, supported by clear accountability and
effective data governance.
The most immediate benefits are
emerging in activities involving large volumes of information that require
rapid analysis and prioritization.
These include regulatory horizon
scanning, policy and obligation mapping, transaction and communication
surveillance, anomaly detection, evidence gathering and regulatory reporting
quality assurance.
However, Sankaran cautions that more
complex compliance decisions remain unsuitable for autonomous systems.
"AI can highlight potential
issues, but humans should remain responsible for final decisions," he
says.
Conduct assessments, regulatory
interpretation, disciplinary decisions and final assurance opinions require
contextual understanding that automated systems cannot yet reliably provide.
Evaluating AI effectiveness also
demands a broader approach than measuring operational efficiency or cost
savings.
Sankaran argues that institutions
should assess accuracy through false positive and false negative rates,
precision and recall, while examining how extensively automated monitoring
covers business activities.
Equally important are improvements in
risk identification, regulatory findings and remediation outcomes.
Governance standards must incorporate
explainability, auditability, reproducibility and independent validation.
"AI outputs should be regularly
validated, challenged and back-tested, with standards comparable to established
model risk management principles," he emphasizes.
Looking ahead three to five years,
Sankaran expects AI to facilitate closer integration between compliance, risk
management and assurance functions.
A shared risk intelligence capability
could continuously monitor exposures, test controls, collect evidence and
generate management information from a common information base.
Nevertheless, he rejects the prospect
of complete convergence because the three disciplines serve distinct governance
objectives.