The era of hyper-personalization has hit financial services in a huge way. Last year, 18% of investment in fintech went into personalized financial management according to Dealroom, while the rise of open banking has unleashed a wave of specialized fintechs that have brought an unprecedented number of personalized recommendation products to the market.
Clearly, the secret is out: In use cases across financial services, deep learning recommender systems deliver a superior, one-to-one experience tailored to every customer’s specific needs and behavior — which ultimately helps financial institutions and fintechs win market share by acquiring customers more effectively while boosting loyalty and retention.
Unlocking the next best action
The boom in personal finance management can be partly attributed to open banking, which has enabled specialized fintechs to enter the market by granting them secure access to transaction data. With a focus on personalized solutions, fintechs are using this data to determine the “next best action,” a customer-centric strategy that prioritizes meeting the unique requirements of each client through personalized offers, services, or deals.
Personetics, a fintech that analyzes transaction data in real time using the AWS AI/ML platform powered by NVIDIA’s accelerated computing, operates as an open banking aggregator. The company analyzes banking data from numerous API endpoints and uses machine learning to categorize it into specific categories, such as restaurant spending or e-commerce. Personetics then shares this categorized data to banks, which use it to offer more personalized suggestions to their customers.
Compared to traditional machine learning systems, deep learning recommendation models can paint a more complete picture of a customer, understanding their preferences and making more accurate predictions. Deep learning models are not limited to explicit data points such as user reviews or how many times the customer logged into the website, but more implicit data points such as the timing and order of customer interactions. They can also analyze relationships over time, not just within a single session.
“The large neural networks used in deep learning are very efficient at uncovering non-trivial and non-linear relationships between users and actions and outcomes,” explains Kevin Levitt, Head of Global Industry Business Development for the Financial Services industry at NVIDIA. “The big capability it brings to the table is the ability to analyze sequences over time and determine the best opportunity based on other conversions that successfully delivered the right message at the right time. This way, banks can facilitate a better experience for the next customer in line.”
For example, call centers can use AI and deep learning models to pull in unstructured data and power client service recommendations. When a customer reaches the call center, an automatic speech recognition model transcribes the conversation, which gets funneled into natural language processing that structures the call data. These inputs, along with historical data from a CRM platform, are then fed into the recommendation system in real time to guide representatives in providing the best service to achieve a desired outcome, such as improved customer sentiment, showing empathy, and offering appropriate solutions.
The “next best action” can be determined across all customer touch points, including the call center, mobile app, website, and even in-person interactions with bankers. Beyond cross-selling, the system can also recommend relevant content, such as market news based on a customer’s investments, educational content based on customers’ financial journey and life stage, and personalized robo-advisory services for managing stock portfolios.
Bringing real personalization to personal finance
Traditionally, if a customer wanted to increase their credit limit, they would call their FI and go through a simple script with a call center representative. The representative would ask how much money the customer wants and the reason for it, then run an approval request to determine if the increase can be granted.
Now, fintechs and FIs can instead use deep learning recommendation systems to understand the customer’s broader life experience and identify alternative opportunities for achieving the same outcome. For instance, they might determine that the customer is better off qualifying for a personal loan or moving money to a different account with a better interest rate, rather than raising their credit limit.
The UK-based fintech Cleo, for example, uses open banking transaction data and deep learning powered by NVIDIA and AWS to drive personalized recommendations to clients in the form of a chatbot. Users can get answers to questions such as “How much am I spending on food shopping this month?” or “How can I save 300 pounds by the end of this month?” The app can also suggest particular steps that users should take to attain a specific life goal, such as building a savings plan for an upcoming holiday.
“Cleo utilizes the AWS platform powered by NVIDIA to give us the flexibility, scale, and cost effectiveness that we need to deliver products at speed, so we can focus on providing the most value to our users,” says Sam Taylor, VP Technology at Cleo. “We have reduced the amount of time users spend with our customer service team by 30%, giving users the ability to self-resolve frequently asked questions and improving user satisfaction rates.”
The same principles apply to companies seeking their own financial services tools, in addition to ones that serve their customers. Attunely, a Seattle-based fintech, uses deep learning to build omnichannel collection models for companies to recover revenue from delinquent accounts. By analyzing alternative data in addition to anonymous debt placements, Attunely’s customers can see anywhere from 5% to 20% more top-line revenue without increasing collection activity.
“The magic is in being able to leverage data from hundreds of millions of anonymous debt placements, and billions of phone calls, letters, text messages, and e-mails,” says Ryan Kosai, CTO at Attunely. “We reprocess all this data nightly and AWS services permit us to process this spiky, high-throughput load at low cost.”
The benefits of deep learning and generative AI
Once fintechs and FIs have a better understanding of who the customer is and where they are in their financial journey, they can leverage generative AI to enhance various communication channels. For example, they can develop better call center scripts, craft more compelling email subject lines, create newsletters with more relevant articles, and select more fitting images for their marketing campaigns. By marrying deep learning with generative AI, they can quickly identify what will most effectively resonate with potential or existing customers —which leads to improved conversion rates and higher customer satisfaction and loyalty.
The path to unlocking that competitive advantage begins with data ingestion and storage. For instance, AWS architecture enables fintechs to have a 360-degree view of data from various sources and effectively store it using AWS architecture. A managed ML service like Amazon Personalize also allows developers to build applications with the same technology used by Amazon.com for real-time personalized recommendations.
“We’re seeing more and more appetite for personalized experiences in the financial services industry built up by services like Netflix and Amazon.com,” says Sam Edge, Global Head of Fintech for Startups at AWS.
This type of personalization also allows fintechs and FIs to surface relevant products for their customers. NerdWallet, a personal finance company, provides tools and advice that make it easy for customers to pay off debt, choose the best financial products and services, and tackle major life changes like buying a house or saving for retirement. By leveraging Amazon SageMaker and NVIDIA accelerated computing, NerdWallet is training their models to more effectively connect customers with personalized financial products.
Of course, the speed of delivery is also key. Choosing the right AI/ML and accelerated computing partner is necessary for two reasons: fintechs and FIs need to reduce time spent training models so they can get to market quickly, and they also must continually retrain those models with new data and inputs to ensure up-to-date accuracy. This is essential to avoid slowing down the customer journey and to ensure an effective, personalized interaction for every customer.
Capital One uses the Merlin application framework powered by NVIDIA GPUs in AWS for building high-performing recommender systems at scale, reducing the time-to-market for data scientists and achieving better online advertising performance. Upon switching from machine learning to deep learning with NVIDIA Merlin, Capital One saw a 60% improvement in click-to-conversion rates for existing customers.
“From the big banks to the fintechs, there’s so much investment in building AI capabilities because they know it’s a point of competitive differentiation,” says Levitt. “The companies that invest in deep learning capabilities for recommendation systems are likely to outperform their competitors in acquiring customers and delivering a superior customer experience, gaining market share in the process.”
Download the State of AI in Financial Services report to discover more about recommender systems and hyper-personalized finance.
