By Nitin Sareen, Senior Vice President, Consumer and Small Business Banking – Consumer Analytics & Marketing, Wells Fargo India & Philippines
An algorithm that protects you from bank fraud by spotting an anomaly in the angle at which your phone is tilted, or the speed at which you type, is not science fiction
As millions of consumers interact with banks via multiple channels, the amount of data available has grown exponentially. The scale and size of data and the modern technology stack, along with advancements in algorithms has created many new use cases for AI/ ML (Artificial Intelligence/Machine Learning). Let’s explore a few, such as customer segmentation which helps with personalization, simplifying KYC, effective risk decisioning, and fraud prevention.
How AI/ML tools create a single view of the customer
The first challenge is the scale of the data that is now available, then making sense of that data, and finally rendering it usable. That is where AI/ML helps. What was “Big Data” a decade ago is just regular data now. Data lakes, where data from diverse sources flows in at scale and speed, are the first place where graph and semantic AI algorithms start to create data lineage and metadata. Semantic AI uses a combination of knowledge graphs (that connect different data bases and make them searchable) and NLP (Natural Language Processing) to effectively manage data quality and governance before it starts to flow downstream.
The data then gets stored in a data management or customer data platform where graph-based meta-heuristic ML algorithms are leveraged to connect, harmonize, curate, and enhance the data across sources to build semantic knowledge graphs. This data is then serviced via a data fabric or data mesh for downstream users and creates a single view of prospects or customers. This single view can be generated even across channels and interactions that are not linked. The insights thus derived have applications across channels and products and help create a seamless customer journey experience.
Customer segmentation is a key tenet of any customer-focused business, including banks. With the number of dimensions or attributes across which the data is available and stored increasing manifold, customer segmentation involves running advanced high dimensional unsupervised and supervised algorithms which help cluster and/or segment data at scale. Dimensionality reduction algorithms can then embed this data into lower dimensions and enable businesses to easily visualize these multi-dimensional clusters/segments of customers. These segments are used to create look-alike models (prospects/customers who look like or act like your best customers), for a variety of customer targeting and personalization strategies.
Making KYC simpler and managing credit risk
Computer vision based deep neural net models can be leveraged for easier verification of KYC norms. These models integrate with digital products to allow quick, seamless, and even remote onboarding of customers. Additionally, it is particularly important for banks to manage a wide variety of credit and market risks. Sophisticated AI algorithms are used to predict the riskiness of each lending decision and help underwriters make decisions in line with the bank’s risk appetite. Machine learning models help perform regular and rigorous assessment of the existing portfolio for forecasting the extent, timing, and impact of credit defaults and market or investment risks. Risk portfolio managers are constantly demanding these models become faster, more accurate, and perceptive of the micro and macro trends.
Detecting fraud through pattern tracking
Each customer has a particular usage pattern when it comes to their financial instruments, including when, where, and how they use it. A significant deviation from this pattern can indicate a potential fraud. Sufficiently advanced machine learning algorithms are trained to develop this nuanced understanding and knowledge at scale – in some cases even to detect the angle at which the phone is tilted and speed of typing. These models run in real time to prevent frauds from happening, while allowing authentic transactions to go through.
Social listening and customer feedback
Every time a customer interacts with the bank directly or indirectly, it is an opportunity for the bank to reinforce trust. Advance NLP and NLU (Natural Language Understanding) based AI algorithms are used in social listening and customer feedback to allow the banks to understand what the apparent and latent customer pain points are. Additionally, NLG (Natural Language Generation) powered bots interact with the customers to pick up customer feedback. When this information is shared with product managers, they can work on enhancing customer experience.
With the ever-increasing amount of data being generated, banks are leveraging modern technology to store it. Constant innovation and research are making algorithms more advanced and faster at finding insights at speed and scale. The space is only getting more exciting by the day.