PayU Doubles Down on GenAI to Boost Fraud Detection and Productivity

PayU, a Prosus-owned company, is into payments and fintech business. In an aggressive emerging tech push it is upping the ante on its Gen AI-related initiatives. Whether it is about enhancing customer experience, merchant onboarding and servicing, fraud detection, or boosting employee productivity, PayU is doubling down on its efforts to leverage the full potential of Gen AI in 2025 and scaling it beyond.

In India, PayU offers advanced payment gateway solutions to online businesses. It supports over 4.5 lakh merchants with more than 150 payment methods and is the go-to payment partner for leading e-commerce platforms and most major airlines.

In an exclusive conversation with CiOL, Koushik Kadidal, Chief Data Officer and Head of the Insights Business at PayU, shares insights into the company’s strategic use of Generative AI to tackle business challenges and drive efficiency. He explains how PayU has implemented Gen AI models to streamline merchant onboarding and due diligence, helping reduce fraud and chargebacks, and how AI is being used to filter spam and duplicate support tickets, saving valuable time for care agents. Koushik also discusses Toqan, PayU’s internal Gen AI platform designed to enhance employee productivity, and outlines the roadmap for upcoming Gen AI initiatives through 2025 and beyond. Excerpts.

PayU has been leveraging Gen AI across various functions. What have been the biggest successes, and where have you faced challenges or limitations in its deployment?

At PayU, we are strategically deploying Generative AI across our businesses to drive innovation and operational efficiencies. Our key successes demonstrate the tangible impact of such initiatives:

For instance, in our payments business, Gen AI has significantly enhanced fraud prevention capabilities, streamlined merchant onboarding, and transformed customer service. We have achieved an 80% reduction in fraud and chargeback losses while accelerating the onboarding of top-category merchants to just 2–3 days. Our Gen AI-powered email classification system now automatically resolves 30–40% of merchant inquiries instantly, with spam and duplicate ticket filtering saving our care agents 7% of their processing time.

If you look at our credit business, we have made significant strides in democratizing Gen AI access. All employees now have secure access to leading Gen AI models like OpenAI, Meta, Anthropic, and others within a robust InfoSec and Compliance framework. Notably, PayU Finance employees can now access backend databases without requiring prior SQL knowledge, which has substantially improved data accessibility and decision-making across the organization.

Meanwhile in order to boost the productivity of our developer teams, we have implemented Gen AI tools for code reviews, security assessments, and text-to-code generation with 65% accuracy, yielding 15–20% time savings. Furthermore, our internal Gen AI platform, Toqan, has revolutionized how our employees access information, draft reports, and make data-driven decisions through secure access to leading AI models within our robust compliance framework.

However, as with any transformative technology, there are a few present challenges. Balancing innovation to achieve future successes with Gen AI, all while ensuring compliance with regulatory and governance standards, is somewhat an opportunity and a limitation. We are putting in exponential effort to ensure the ethical use of Gen AI.

While Gen AI is streamlining merchant onboarding and due diligence, how do you ensure accuracy and fairness in decision-making, especially in preventing false positives or biases?

Ensuring accuracy, fairness, and transparency in our AI-driven decision-making processes is paramount, particularly for critical functions like merchant onboarding and due diligence. We have implemented a comprehensive framework to address these concerns.

First, our AI models are built on high-quality, representative data that undergoes rigorous preprocessing to minimize inherent biases. We continuously fine-tune these models to improve accuracy and reduce false positives that could unfairly impact legitimate merchants. Our proprietary risk assessment engines employ advanced authentication and verification tools that adhere to stringent security standards, including PCI-DSS, GDPR compliance, and Global Prosus Standards.

Second, we have established robust governance mechanisms that ensure human oversight of AI-generated decisions. For merchant onboarding, our systems flag potential discrepancies between provided information and public data, but final decisions always incorporate human judgment. This human-in-the-loop approach allows us to catch edge cases that pure automation might miss while maintaining the efficiency benefits of AI.

Third, we have invested in building audit trails and explainability tools that provide transparency into how decisions are made. These systems allow us to trace the rationale behind specific outcomes, which is essential for both internal reviews and regulatory compliance. We are also implementing more sophisticated fairness testing protocols to proactively identify and eliminate any discriminatory patterns in our models.

Finally, our commitment to ethical AI is embedded in our long-term governance strategy. We continuously align our practices with evolving regulatory frameworks while investing in research and development to enhance our capabilities. This approach ensures that as we scale our Gen AI applications, we maintain the highest standards of fairness and accuracy in all our decision-making processes.

Automating SPAM and duplicate ticket filtering sounds efficient, but has there been any pushback from customers or agents about AI-driven decisions? How do you balance automation with the human touch in customer support?

Our implementation of Gen AI for automating spam and duplicate ticket filtering has been receiving a positive response from both our customers and support agents. This positive reception stems from the tangible improvements in service quality and efficiency: our turnaround time for query resolution has significantly improved, with 30–40% of inquiries now addressed instantly through AI-powered responses.

Specifically speaking, Gen AI is deployed for first-level responses and initial query classification, which has helped us save approximately 7% of our care agents' processing time. All follow-up communications and escalations are handled personally by our support team, maintaining the human touch that merchants value. As a result of the Gen AI integration, the efficiency gain allows our team to focus their expertise on more complex merchant issues that benefit from human attention.

Toqan is positioned as a productivity booster for employees. How do you measure its real impact? Are there any concerns around over-reliance on AI or potential job displacement?

Toqan, our internal Gen AI platform developed alongside Prosus, has been deployed to enhance employee productivity and decision-making capabilities. Apart from the success we are already observing, we have devised a few internal goals that we aim to achieve with Toqan’s deployment this year. We want to further automate our support processes. We are looking to integrate advanced Gen AI technology updates to automate 60–70% of responses and expand its capabilities across all channels, including emails, chatbots, and self-help tools. We are also exploring Large Language Models (LLMs) for code generation and review, aiming for 65% accuracy and a 50% reduction in Level 1 code review time. Keeping these goals as our north star, we will know how much progress we are making.

Our proactive approach to deployment of GenAI technology is clear. We want to augment our team’s capabilities by automating repetitive tasks while enhancing creative and analytical work. We are running internal campaigns and training programs to equip our workforce with such advanced tools to enhance their productivity. We will continue to invest resources and time in building relevant skill sets and teams to drive PayU’s business success. 

As you look ahead to 2025 and beyond, what are the key hurdles/opportunities in scaling Gen AI further? Are there ethical, regulatory, or technological barriers that could slow down innovation?

We  believe that Gen AI presents both a transformative opportunity and a learning playground for financial services industry players. At PayU, we are focused on making the best and most responsible use of this advanced technology.

With regard to the opportunities, we are particularly excited about expanding our developer tools with LLMs to achieve 65% accuracy for code suggestions and reduce Level 1 code review time by 50%. Additionally, we are enhancing our merchant lifecycle management capabilities through AI-powered revenue forecasting and churn prediction. Although Gen AI implementation requires significant investments in talent, technology, and governance frameworks, we view these as strategic opportunities to differentiate PayU in the market. By addressing evolving regulatory landscapes and computational demands head-on, we aim to build a preferred end-to-end financial services platform for our merchants and partners.