New Delhi, Oct. 27 -- Traditional merchant risk assessment, which relies heavily on checks conducted during onboarding, may not be sufficient as businesses change their transaction patterns, operations and online presence, according to Rajiv Beri, chief technology officer at NPST (Network People Services Technologies Ltd).
The Thane-based fintech firm offers digital payment technology and infrastructure solutions to banks, financial institutions and payment service providers.
"Banks and payment providers need to move towards continuous monitoring to identify emerging risks, including unusual transaction volumes, rising chargebacks and discrepancies between a merchant's declared business and its actual activities," Beri told Mint CIO Circle.
A merchant that passes initial identity and compliance checks may present a different risk profile later. Traditional rule-based systems can miss such changes when they assess individual indicators in isolation, making it important to evaluate transaction behaviour alongside business information and other risk signals.
Artificial intelligence (AI) can help financial institutions identify patterns across these indicators and flag anomalies for investigation, Beri said. Unlike conventional systems that rely on predefined thresholds, machine learning models can establish behavioural baselines for individual merchants and business categories, helping identify deviations that may otherwise go undetected.
However, unusual activity does not necessarily indicate fraud. AI-generated alerts require contextual assessment and verification before institutions decide on further action.
Merchant risk assessment involves more than transaction monitoring. Identity records can help verify business information, while behavioural analysis can identify changes in transaction frequency, refunds and chargebacks. Website monitoring can flag prohibited products or services, domain-related risks and inconsistencies between a merchant's declared category and its online activities.
Taken together, these signals can help banks decide whether to seek additional verification, revise transaction limits or escalate a case for investigation.
Beri said AI-based systems could also help identify potential payment-network compliance violations before they develop into larger problems. Such systems, however, cannot independently establish that fraud or a regulatory breach has occurred.
Data quality remains a significant challenge. Merchant records, transaction histories and website information may be incomplete, inconsistent or updated at different intervals, affecting the reliability of risk assessments. Models must also distinguish suspicious activity from legitimate changes in business operations, while limiting false alerts that add to compliance teams' workloads.
Financial institutions must additionally address data privacy, security, explainability and scalability as monitoring expands across larger merchant ecosystems.
AI-led risk systems need to work alongside existing fraud management processes rather than replace regulatory checks or human judgement. Risk teams must be able to understand why a merchant has been flagged and determine whether the evidence warrants intervention.
NPST's Risk Intelligence Decisioning Platform (RIDP) combines identity, behavioural and website-risk signals to support merchant underwriting and transaction monitoring. The company said its models are trained on more than 750 million transactions. Beri did not provide comparative data on fraud reduction or the platform's impact on false-positive rates.
Over the next three to five years, AI is likely to make merchant risk management more continuous and predictive, Beri said. Greater use of risk intelligence across the payments ecosystem could also help identify recurring risk patterns, subject to data-sharing arrangements, privacy safeguards and regulatory requirements.
The challenge for banks and payment providers will be to detect emerging threats earlier without imposing unnecessary friction on legitimate merchants.
Published by HT Digital Content Services with permission from MINT CIO CIRCLE.