Wells Fargo, a major financial services company, has its headquarters in San Francisco, United States.
The company operates in 35 countries and has approximately 70 million customers worldwide. Financially, Wells Fargo reported strong performance in 2023 and generated $19.1 billion in net income, an 11 percent increase from the previous year. The company also reported strong financial performance for the fourth quarter of 2024. Its net income reached $5.1 billion, representing a significant 47% increase compared to the same quarter in the previous year.
The bank has made significant strides in integrating AI into its operations and customer-facing products. A notable example is Fargo™, a virtual assistant in the Wells Fargo Mobile app that uses Google’s conversational AI to help customers with everyday banking questions.
Per its annual report of 2023, consumers interacted over 21 million times last year with the bank via this virtual assistant. The bank has also developed the Enterprise Open Source Data Science Platform, which enables data scientists to create innovative AI solutions more quickly.
In 2019, Emerj’s research covered two developing use cases at Wells Fargo focused on conversational AI. This article explores the latest findings related to two fully mature AI use cases far closer to the core of Wells Fargo’s business objectives:
- Advanced analytics to personalize customer engagement at scale: Leveraging predictive and adaptive analytics for real-time decision-making and generative AI for dynamic content creation to deliver hyper-personalized interactions that drive engagement and revenue.
- Leveraging modern machine learning for fairer loan decisions: Adopting modern machine learning models for more accurate and fair decision-making by uncovering complex patterns traditional models miss.
Using AI to Personalize Customer Engagement at Scale
Institutions in the banking, financial services, and insurance (BFSI) sector face significant challenges in enhancing customer engagement. A 2019 study from the Spanish Journal of Marketing focusing on the retail banking industry identified key factors influencing customer engagement, including market orientation, customer satisfaction, emotional connection, and self-brand connection.
Additionally, the sector grapples with fragmented customer data, as information is often siloed across various channels and departments, hindering the creation of a unified customer view. A white paper published by the World Economic Forum explains that this fragmentation complicates efforts to deliver personalized services and can lead to inconsistent customer experiences.
The paper also details how willing consumers are to share their data if financial institutions provide sufficient personalization in exchange. While 84% of consumers surveyed contended that they have insufficient control over how organizations leverage their data, 75% said they are more willing to share that data with trusted brands.
According to the case study documentation published by Pega, to strengthen customer relationships, the bank sought to enhance personalization at scale — delivering empathetic, proactive, and one-to-one engagement tailored to individual needs.
However, achieving this required overcoming infrastructure, process optimization, and scalability challenges while meeting rising customer expectations for a seamless, frictionless experience.
With over 70 million customers, Wells Fargo faced difficulties personalizing everyday interactions, as it needed to process and analyze billions of digital touchpoints in real time.
The case study claims that the bank partnered with Pega to leverage its Pega Customer Decision Hub as the core technology for its AI-driven customer engagement strategy.
Pega is a software company that provides a platform for workflow automation and AI-powered decisioning, with its Customer Decision Hub serving as what promotional materials describe as an “always-on brain” that uses AI to predict customer needs in real time and deliver personalized next-best actions across channels.
The workflow for Pega’s Customer Decision Hub works as follows, according to the same documentation:
- Data Integration: The system automatically collects and unifies customer data from every possible source, such as website visits, purchase history, customer service interactions, and more, creating a comprehensive 360-degree customer profile in real time.
- Action Definition: Marketers define specific business goals and create targeted offers, messages, and treatments that the platform will use to engage customers strategically.
- AI-Powered Decisioning: The platform uses advanced AI and predictive analytics to instantly evaluate customer data and determine the most optimal next-best action for each customer.
- Engagement Policy Setting: Marketers can establish clear business rules that guide the system, determining customer eligibility for specific actions and prioritizing them based on strategic business objectives.
- Real-Time Personalization: CDH automatically delivers personalized interactions across all channels, instantly adapting to each customer’s context and behavior in milliseconds.
- Continuous Learning: The system continuously records and analyzes customer interactions, automatically updating its predictive models and refining decision-making algorithms.
Screenshot of Customer Decision Hub. (Source: Pega)
Screenshot of Customer Decision Hub. (Source: Pega)
The Pega documentation of the use case mentions the following specific and quantifiable business results of the partnership:
- Analyzed 4 billion digital interactions to identify the “next best conversation” for each individual customer.
- 3-10x increase in customer engagement rate, depending on the channel and use case.
- Successfully personalized messages and experiences for 70 million customers across multiple channels.
Wells Fargo’s financial performance also shows a potential correlation with its AI and personalization investments. The bank reported a robust 11% revenue increase in 2023, with net income reaching $19.1 billion. The implementation of Pega’s Customer Decision Hub could have contributed to these gains by enabling more targeted customer interactions, improving conversion rates, and enhancing operational efficiency.
The bank’s ability to personalize experiences for 70 million customers and the platform’s capability to process 4 billion digital interactions suggests a direct link between advanced AI technologies and financial performance. The 47% year-over-year net income growth in Q4 2024 further underscores the potential strategic value of their AI-driven customer engagement approach.
Leveraging Modern Machine Learning for Fairer Loan Decisions
Traditional linear regression models can present challenges for banks automating the loan approval process due to their reliance on simplified assumptions that may not capture the complexity of a borrower’s financial situation. A research study from the University of Melbourne shows that these models often assume a linear relationship between variables and may not account for non-linear interactions, leading to inaccurate predictions and potential rejections of eligible applicants.
To enhance accuracy and efficiency, a research paper from the Dept. of Computer Science at the University of Nigeria highlights that modern machine learning models are necessary, as they can process a broader range of data, identify complex patterns, and provide better risk assessments, enabling banks to make more informed and fair decisions.
According to a blog published by NVIDIA, Wells Fargo faced significant challenges in its loan processing systems and chose to address these challenges by leveraging NVIDIA’s GPU-accelerated machine learning and deep learning models to improve processing speed, accuracy, and the ability to provide transparent, explainable AI (XAI) solutions.
By adopting these advanced models, Wells Fargo aimed to enhance its decision-making process by moving beyond traditional methods like regression and risk models. Hence, the bank adopted machine learning and deep learning techniques, which are more accurate and complex, necessitating the use of explainable AI to ensure that customers and regulators could easily interpret decisions.
Per the blog cited above, NVIDIA’s GPU-accelerated technology enhances Wells Fargo’s machine learning and deep learning models, enabling more accurate and efficient decision-making. Here’s a breakdown of how NVIDIA claims the system works for Wells Fargo:
- GPU Acceleration: NVIDIA’s powerful GPUs speed up the processing of large data sets, making the AI models run faster and more efficiently and enabling Wells Fargo to handle complex computations.
- Advanced AI Models: Wells Fargo uses deep learning models like ReLU Neural Networks, which are more complex than traditional models but significantly improve the accuracy of predictions. These models can understand various variables and nuances in loan applications, such as economic conditions and borrower behavior.
- Explainable AI (XAI): The LIFE (Linear Iterative Feature Embedding)algorithm, developed by Wells Fargo, uses XAI techniques to ensure the decision-making process is transparent and interpretable. XAI helps break down complex machine learning results into more straightforward, understandable outputs.
The NVIDIA blog claims Wells Fargo developed the LIFE algorithm to enhance the interpretability and accuracy of its AI models. It aims to provide both high prediction accuracy and ease of explanation, allowing the bank to make complex decision-making processes understandable. LIFE simplifies the outputs of deep learning models, breaking them down into linear sub-models.
A figure from the Wells Fargo research paper “Linear Iterative Feature Embedding: An Ensemble Framework for Interpretable Model” from Sudjianto, Agus et. al. showing “the general framework of LIFE algorithm and options colored in red in each step are used” in developing the algorithm. (Source: Wells Fargo)
The breakdown of models enables the bank to identify which factors are most significant in determining the outcome of a loan application, such as the borrower’s debt-to-income ratio or credit score.
The model works according to the following process:
- LIFE analyzes up to 80 variables per loan application, ensuring comprehensive risk assessment.
- It provides clear, data-backed explanations for loan outcomes by identifying significant factors such as:
– Debt-to-income ratio
– FICO credit score
– Income stability
– Payment history - When a loan is rejected, the model generates specific reasons linked to the applicant’s financial profile, ensuring transparency.
It was in April 2021 that Wells Fargo published the blog about the LIFE algorithm, utilizing NVIDIA’s GPU acceleration for the AI-driven loan approval processes.
However, by March 2022, reports emerged indicating that Wells Fargo had rejected over half of its Black applicants seeking mortgage refinancing during a period of significant refinancing activity. As reported by Bloomberg, in June 2024, this racial disparity case turned into a class action lawsuit of at least 119,100 people.
The disparity raised concerns about potential biases within the bank’s lending practices, suggesting that systemic issues may persist despite technological advancements like the LIFE algorithm.
Even though Wells Fargo did not respond to Bloomberg’s claim, the juxtaposition of these events highlights a critical challenge in the financial industry: the integration of advanced AI systems does not automatically eliminate existing biases.
While tools like the LIFE algorithm are designed to enhance fairness and transparency, their effectiveness depends on the data they are trained on and the broader institutional practices in place.
