Deepak Sharma,
President and Chief Digital Officer,
gives insights about the progress of AI at Kotak Mahindra Bank:
Ravi Lalwani: Briefly describe the key AI project during the lockdown.
During the pandemic, Kotak Mahindra Bank has continued with its digital-first organic growth strategy, which is driven by its ABCD charter that focuses on – AI enriched app, biometric-enabled branch, context enhanced customer experience and data empowered design.
During the pandemic, Kotak Mahindra Bank has continued with its digital-first organic growth strategy, which is driven by its ABCD charter that focuses on – AI enriched app, biometric-enabled branch, context enhanced customer experience and data empowered design.
Deepak Sharma: We have launched and upgraded 3 major AI-led projects in the last 2 quarters. These are around an AI-based nudge engine in our new internet banking platform, upgrading the chat skills of Keya, our conversational chatbot, and automating fund transfers by corporate customers. The bank has also recently upgraded the net banking platform of Kotak Mahindra Bank. Among other enhancements, one of the new features displays contextual messages and recommendations based on customer profile and other factors at multiple placeholders across the site. During the pandemic, we also upgraded our Keya chatbot use cases from 80,000 to over 200,000. It has resulted in higher conversational accuracy. Lastly, the eGBO (electronic general banking operations) project assists in automating fund transfer requests sent by corporate customers. Since launch, we have processed more than 23,000 requests with a cumulative value of over `800 billion.
What were the business objectives of those projects?
Net Banking 2.0 provides better recommendations to the user based on various factors such as login time/date, profile, product holdings, and user behavior by enhancing customer experience, increasing contextual cross-sell and upsell. The platform also educates the customer about various products and the new features of net banking. Similarly, Keya with better Natural Language Processing (NLP) accuracy can handle seamless conversations, thereby providing end-to-end information and request handling with higher accuracy of over 93%. The eGBO project improves accuracy and TAT for corporate client transactions (high-value RTGS, NEFT & IFT – Immediate Fund Transfers). It transforms a manual process into an automated system. It delivers E2E reconciliations and alerts to the clients using automated bots, cognitive services and API integrations. It improves scalability and business efficiency.
Briefly describe the technologies underlying the project.
The technologies underlying the project were NLP, supervised learning model and propensity-based recommendations. For eGBO, we used an innovative fractal science methodology ie, Cognitive Machine Reading (CMR). It is a ‘full stack intelligent data hyper automation platform’ which consists of data ingestion, Robotic Process Automation (RPA), machine learning, and artificial intelligence.
Name the technology companies providing products or services for the project.
The technologies have been developed by the bank internal IT and BIU teams using the existing technology stack. We work with our partner ecosystem companies such as Antworks and Active.ai in select areas like NLP and conversation UI.
What are the main internal and external data sources for the projects?
For our internet banking platform, transaction details, product holdings, and customer profile data are the major internal data sources used to show recommendations. The propensity of recommendations is calculated using these data points. For Keya, it is in form of unstructured data around intent, issues, conversations, voice, and text inputs across various customer interfaces.
Briefly describe the team structure, size, and members that developed tested, and rolled out the project.
Projects are done with cross-functional internal teams. that bring expertise in customer insights, design, data & ML, solution design, technology, and products.
Briefly describe the challenges faced during the project and the lessons learned.
For net banking, one of the areas is around accessing data that is stored in different product databases and showing recommendations with minimum latency. Unstructured data poses a certain set of challenges for automation and standardizing it to the maximum extent possible is important. Also, process re-engineering is important before automation. Storing huge volumes of data and managing a surge in volumes is also important and hence, our new net banking platform creates the additional capacity to accommodate higher volumes.
