Using Web Agents to Detect Fraud and Achieve Real-time Insights


Effective fraud detection necessitates understanding the meaning of streaming data in context, in real-time, and at scale.

The number of digital financial transactions taking place is growing rapidly, driven in part by customer expectations for instant transactions, the ease of accessing financial accounts electronically, the need for virtual transactions during the coronavirus pandemic, and the growing adoption of digital payment platforms.

In fact, McKinsey’s 2022 Digital Payments Consumer Survey reported that nearly nine out of 10 Americans are now using digital payments. As the volume of global digital activity continues to increase, fraud is becoming more common and difficult to catch. A major concern across all industries is the level of fraud detection needed to mitigate financial risk, preserve company reputation, and ensure high levels of customer satisfaction and brand trust.

Fraud detection and remediation must occur quickly, as cybercriminals are relentless. They continually adjust and invent sophisticated methods to avoid detection and launch new “attacks.” Automated fraud detection – which is in itself complex and technical – is the best option. However, combining data from multiple sources with historical and real-time/streaming data has proven difficult to achieve at low latency in a cost-effective manner.

Real-time insights for fraud detection

To successfully combat fraud, organizations need a system that can seamlessly combine data drawn from multiple data sources (streaming data and data at rest) and analyze it in real time to prioritize insights and generate rapid responses. If it takes hours (or even minutes) to take action after detecting potentially fraudulent activity, the damage is likely already done, and it could even become more severe.

Detecting and acting on fraud is a growing and more difficult challenge. According to the FTC, consumers lost $8.8 billion to fraud in 2022, which was more than 30% greater than 2021’s loss. This trend suggests a mounting urgency for companies to detect and act quickly on fraudulent anomalies.

To detect fraud in real-time, many organizations turn to streaming data, which is the continuous flow of information generated from multiple sources. The processing, storage, and integration of streaming data enables its movement throughout the application stack. Companies have the ability to analyze both the data being moved and the movement itself through the application stack.

Although insights gleaned from stream processing and analytics can come close to real-time, these methods still leave something to be desired when it comes to fraud detection. What if there was a way to access true, real-time insights, contextualize and visualize the data, and intervene immediately?

See also: Next-Gen Online Fraud Detection: The Journey is the Destination

The best, most effective way to detect fraud

Effective fraud detection necessitates understanding the meaning of streaming data in context, in real-time, and at scale while simultaneously programming autonomous action to respond based on meaning and context. The query-based approach of traditional stream processing architecture fails to meet all these requirements because it breaks streaming context at multiple points and increases latency. It also forces companies to make trade-offs regarding what data gets stored and what gets dropped based on cost and feasibility. More effort is needed to express true “real-world relatedness,” derive insights, and run decision automation.

However, there is a solution. For companies to understand what’s happening in real time, derive meaning from streaming data, and respond automatically, they need web agents. Web agents are web-addressable stateful objects – similar to actors in an actor model. They’re stateful because they preserve their data and apply context locally between operations, which keeps latency low. They don’t need to wait on database round trips as the required context is already accessible and updated continuously. Streaming APIs are leveraged to communicate state changes to other web agents and downstream applications like notification systems and real-time user interfaces.

When web agents are tasked with answering business-critical questions, they will continuously generate their own contextual KPIs and proactively update downstream systems continuously about the relevant information they need to know. They can even be programmed to take autonomous remediating action when appropriate. For example, Web Agents tasked with answering business-critical questions (such as whether fraud has occurred) continuously compute their own contextual KPIs (for example, fraud rate and final approval rate), proactively inform users about what they need to know at the right time (through an action like sending a subscriber an alert about suspicious activity), and even take autonomous remediating action, when appropriate (like freezing an account to prevent fraud).

Fraud detection use cases for Web Agents

Web agents have a number of applicable use cases, no matter the industry, from credit card fraud to malicious telecommunication activity to responding to rogue gamers.

Financial institutions must be able to detect and stop illegitimate transactions and credit card compromises to mitigate damage to customers. This requires enriching streaming data, such as information about real-time geolocation and current activities, with historical data, such as details of past transactions and spending habits. To do this, financial service providers can create a web agent for each customer and then load and model their historical data. Each financial transaction is then sent to the individual customer’s web agent, which can execute fraud detection logic by accessing local historical data. The resulting computations are stored as part of each web agent’s new “state.” This can be programmed to set off alerts if there’s suspicious activity or suspend a credit card if fraud is detected.

The gaming community isn’t immune to fraudulent schemes. Video game companies are using in-game actions and player data to spot and address suspicious behaviors in real time. By comparing live game data with a players’ past activities and account info, they can detect potential fraud. For instance, if someone is reported for in-game spamming, this system could immediately review their past actions and, if they are a repeat offender, depending on the circumstances, automatically issue a ban.

The power of real-time fraud detection

Real-time fraud detection can seem unrealistic, and settling for close-to-real-time isn’t the only option. Web agents simplify the process, offering true real-time insights from actual events and enabling automatic responses. From detecting fraudulent credit card activity or a bad gaming actor, web agents are key to identifying and acting on fraud instantly.

Aditya Chidurala

About Aditya Chidurala

Aditya Chidurala is Director of Product Marketing at Nstream. He joined the company after four years at Confluent. Aditya has a deep understanding of the open-source streaming data software ecosystem, and he is responsible for developing the positioning, messaging, and industry vertical and horizontal use cases enabled by the Nstream Platform. He holds a bachelor’s degree in electrical engineering from IIT Roorkee and an MBA from USC Marshall School of Business.

Leave a Reply

Your email address will not be published. Required fields are marked *