At DevRev’s Effortless Mumbai 2026 event in March, over 600 leaders from India’s leading enterprises gathered to explore how AI works for teams. The conversations centered on making work more meaningful and giving people their time back. 

A fraud prevention executive from one of India’s largest banks stood on stage and described her team’s daily reality. They stay ahead of people who wake up every morning with one goal, which is to cheat the system or the people using it. After two decades at the bank, she’s watched fraud evolve from a technical problem into a human psychology problem.  

The scale she operates at stops most fraud teams cold. India processed 21.6 billion UPI transactions in December 2025 alone, more than half the world’s digital payment transactions concentrated in a single country.  

Against that backdrop sits the detection challenge. The Indian Cyber Crime Coordination Center reported 220,000 to 230,000 fraud cases during the period, and those numbers only capture victims who noticed they’d been defrauded and reported it. Most don’t realize for weeks, and many never report.  

Traditional fraud prevention focused on flagging anomalous payments, blocking suspicious transfers, and reacting to signals. The executive pointed to the shift that’s breaking that model. 

Why Behavioral Signals Matter More Than Transaction Flags  

Technology got harder to compromise, so fraud moved ahead as well. Instead of exploiting software vulnerabilities, attackers target emotions like greed, fear, urgency, and the desire to help. They layer social engineering techniques to get people to compromise their own credentials, share their own OTPs, authorize their own fraudulent transactions. When a legitimate customer voluntarily enters their password, shares their one-time code, and clicks confirm, every technical signal reads normal. The fraud lives in the context surrounding the transaction. 

The executive described her team’s work as finding needles in a constantly moving haystack. They monitor millions of transactions daily, each one connecting to hundreds of behavioral signals. Device fingerprints, location patterns, IP addresses, typing rhythms, swipe patterns, and pause durations. Fraud doesn’t announce itself with a single wrong number. It hides in subtle deviations from established behavioural baselines. The fraudulent transaction is the last step in a much longer story that begins days earlier with account takeover attempts, credential testing, device changes, and location shifts. A successful fraud prevention system reads that story in real time and intervenes before the final transaction happens.  

The Coordination Problem  

DevRev demonstrated this behavioral approach during the event with a simulated fraud case. A customer, Arjun – reported a fraudulent transaction. The support agent, Riya, opened the ticket and asked the AI a single question – :Help investigate this dispute”. The system pulled signals from multiple internal platforms. Transaction history, device patterns, location data, account behavior, and customer communication history. It analyzed relationships between data points that would take a human analyst hours to compile, then surfaced recommendations with context and evidence.  

Should the transaction be marked fraudulent? The AI presented its case but kept Riya in control of the decision. Once she confirmed, the system executed the full response. Processed the refund, locked the compromised card, initiated a new card request, asked whether to close the ticket and notify the customer. The entire workflow stayed within a single interface.

Fraud investigation at scale is a coordination problem disguised as an analysis problem. Multiple systems hold pieces of the puzzle. Different teams own different response actions. Fraud analysts traditionally spend most of their time gathering data rather than making decisions. The demo showed how AI teammates can compress hours of coordination work into minutes while keeping human expertise in the critical decision points. 

AI as a Vaccine  

The banking executive framed the future of fraud prevention around a specific concept: AI as a vaccine. Vaccines work by using the same virus in a different form to build immunity. Fraud prevention now requires AI versus AI because fraudsters already use deepfakes, synthetic identities, and AI-generated phishing content. They collaborate across borders, analyze which banks handle cases in which ways, and train their teams on those patterns.  

The question is whether we can deploy AI that understands behavior and intent for fraud detection. She described the future as systems that identify genuinely unusual behavior patterns within millions of normal-looking transactions,  moving beyond “this transaction is unusually large” or “this location is new” to understanding that this sequence of micro-behaviors indicates a compromised account ,even though every individual transaction appears legitimate.  

This challenge extends beyond fraud. Any organization operating at scale faces the same tension between customer experience and security. Add friction and customers abandon legitimate transactions. Remove friction and fraud increases. AI can resolve this by moving from rule-based gates to behavioral understanding. The right transactions flow through, the suspicious ones get stopped, and legitimate customers barely notice the protection layer.  

From Detection to Resolution  

Many organizations already have sophisticated fraud detection models that identify suspicious patterns. The output is a score or a flag. Then the analyst still needs to investigate, multiple teams need to coordinate, and actions span different systems. Intelligence without operational integration creates alerts rather than outcomes.  

QualityKiosk’s DevRev implementation practice focuses on connecting these pieces. The platform demonstrated at Effortless shows how AI teammates handle data gathering, cross-system coordination, and repetitive workflow steps without replacing the fraud analyst’s expertise or judgment. Running a sophisticated fraud model in a sandbox is different from deploying it into production operations where response time and coordination determine outcomes.  

The banking executive mentioned her team has been deploying AI, behavioral biometrics, and LLM models while acknowledging the challenges. Defining the right use cases. Getting models into production. Building confidence in AI-driven decisions. Most organizations still combine AI signals with traditional rule-based checks because trust hasn’t fully scaled yet.  

What Scale Requires  

Processing 50% of global digital transactions means operating in an environment where small failure percentages translate to massive absolute numbers. A fraud detection system that’s 99.9% accurate still generates tens of thousands of false positives daily at that volume. The speaker emphasized that fraud prevention requires connecting real-time detection to post-incident analysis and systemic improvement. When fraud succeeds, teams need to understand exactly how the system was compromised and ensure it doesn’t happen again.  

Most organizations treat these as separate functions. The fraud detection team runs models. The incident response team investigates cases. The risk management team updates policies. AI can close this loop if the operational systems support it. DevRev’s approach integrates conversational AI with support workflows and internal system connections. The platform surfaces patterns in context, recommends actions, executes approved steps, and learns from outcomes. AI value compounds when the system can act rather than just alert. 

Organizations succeeding with AI-powered fraud prevention invest in systems that connect detection to action while keeping human expertise in the loop for critical decisions. They track fraud rates alongside operational efficiency gains. The demo showed a fraud case resolved in minutes that would traditionally take hours or days. The customer received notification of resolution while still in the session. That speed came from integrated intelligence and execution. 

Fraud techniques evolve faster as attackers coordinate across geographical boundaries, share intelligence, and move quickly. Defence needs to match that pace with operational systems that can deploy new detection logic, update response procedures, and coordinate across teams without weeks of planning.  

QualityKiosk’s work with DevRev implementations focuses on this operational layer:  

• How AI integrates with existing workflows 

• Which systems need connections 

• What data needs to flow where 

• How teams maintain oversight while AI handles coordination 

The Questions That Determine Success  

Customers expect instant transactions with zero friction while fraudsters exploit that expectation. Behavioral AI changes the calculus by understanding normal patterns for each customer, spotting anomalies without adding blanket friction. The legitimate transaction flows through instantly, the suspicious one triggers additional verification, and customer experience improves for most people while security strengthens. This only works if the system can act on its understanding rather than producing scores for human review that add latency and limit scale.  

India’s digital payment volume makes it a natural laboratory for these approaches. The scale demands solutions that work in production rather than theory. The insights from DevRev Effortless reflect lessons learned at that scale. Organizations operating in other markets or domains face similar challenges even if the specifics differ. The operational principles transfer.  

For companies evaluating how AI fits into their fraud prevention strategy, the questions center on operational integration, decision authority, response speed, and systemic learning rather than model accuracy alone. The technology exists. Whether the operational systems can support it determines the difference between AI that delivers value and AI that creates new operational complexity. 

To watch the full event coverage, click here: DevRev Effortless Mumbai 2026 – YouTube