Leveraging AI and ML to Understand Customer Behavior in Banking – Decode with the Experts

AI and Machine Learning – Powering the future of business

Artificial intelligence (AI) and machine learning (ML) are no longer futuristic concepts – they drive innovation today. According to a recent Gartner study, AI adoption has surged by 270% over the past four years, with a 37% increase in the number of organizations implementing AI in some form since 2021. These technologies are powerful tools for automating tasks, enhancing decision-making, and personalizing customer experiences. However, their effectiveness hinges on one critical factor: a robust data foundation. 

AI and ML need high-quality, well-managed data to deliver their full potential. Machine learning models, in particular, rely on vast amounts of accurate data to learn, improve, and perform complex tasks. Even the most advanced AI systems can fall short without a strong data management strategy. A modern MDM becomes essential, ensuring data accuracy, consistency, and accessibility across an organization. 

Mastech InfoTrellis recognizes the importance of this foundational element. We collaborated with a leading global bank to establish a robust customer data management system. By leveraging our expertise in Master Data Management (MDM) and AI, we helped the bank gain deeper insights into its customer relationships, enhancing the overall customer experience. Our approach focused on integrating and mastering data from various sources, ensuring it was ready for advanced AI and ML applications. 

Building stronger customer relationships for a leading financial institution 

The bank aimed to understand better its customer relationships to enhance the overall customer experience. By leveraging machine learning, they sought to improve their ability to identify and analyze these relationships, ensuring a more personalized and compelling customer interaction. 

The bank sought a robust MDM solution to manage its customer data. Mastech InfoTrellis was selected to address these challenges. By understanding the data's context and location across modern and legacy systems, we integrated critical data points to ensure accuracy and accessibility for informed decision-making. 

Canonical data model development 

We implemented a sustainable AI/ML strategy focused on high-quality data and data-driven decision-making to ensure responsible AI practices. We established a governance framework, developed technology capabilities, and implemented responsible AI practices. Our data professionals created a canonical data model to structure customer data, enhancing understanding across various business lines. This model was designed to meet the specific needs of each department. 

Mastering data with defined rules 

The next step was to develop deterministic and probabilistic rules to master the data effectively. Data quality was prioritized to ensure accuracy and up-to-date information. By improving data transparency and visibility, decision-makers gained the ability to make informed choices. This foundation facilitated easier access and analysis of data for the data science team. 

Leveraging machine learning for customer relationship identification 

We took a two-step approach to tackle the challenge of identifying authentic customer relationships. 

Establishing a solid MDM foundation 

The first step focused on creating a robust MDM framework. This provided a reliable base for integrating and mastering customer data, ensuring quality and consistency across the bank's operations. The clean and structured data formed the backbone for applying advanced AI and ML techniques. 

Deploying ML solutions 

Leveraging MDM as a foundation, the second approach integrated ML to enhance customer identification. This automated process, surpassing manual methods in speed, cost-effectiveness, and accuracy, offers a more scalable solution. 

Here's a breakdown of the process: 

This bank employed a sophisticated model leveraging ML capabilities to enhance customer relationship identification. Initially, the team analyzed the existing database, representing primary customer connections. Subsequently, they expanded the model's scope by introducing additional datasets and allowing it to assess various potential linkages. Employing a graphical neural network built on the TensorFlow platform, the model assigned relationship strength scores to nodes, identifying primary connections with scores exceeding 50%. As the model progressed, it uncovered previously unassigned or ambiguous customer records, prompting further investigation. The model identified and grouped relationships through this iterative process, presenting a more nuanced understanding than conventional methods. 

The model was refined through a feedback loop with relationship officers. They assessed the identified connections against existing ones, ensuring accuracy and relevance. This iterative process aligned the model with the bank's goals and customer-centric philosophy. By automating the process, the bank reduced manual effort and human error. Beyond transactional data, this holistic understanding of customers deepened trust and improved service delivery, benefiting both the bank and its clients.

Ensuring data integrity and enhancing relationship mapping 

To ensure data security and ethical AI practices, we implemented rigorous data governance policies before granting access to our data science team. This step was crucial, as the model relied on sensitive customer data. 

Leveraging TensorFlow on the Google Cloud Platform (GCP), we developed an accurate model that generated confidence scores for identified relationships. By providing these scores to the client's team, we empowered them to assess the reliability of the model's findings. 

We used Power BI reports to compare identified relationships against existing datasets to validate the model's results. This verification process ensured the accuracy and reliability of the model's output. 

Recognizing the complexity of relationships within customer data, we integrated a graph database into our solution. This approach allowed us to capture and analyze intricate connections efficiently, significantly improving the model's ability to accurately map customer relationships. 

We integrated machine learning models into its production environment to process real-time incoming data. We established new relationships within our graphical data store by leveraging various access service frameworks. This approach enabled us to:  

• Improve decision-making: We refined user interfaces and screens to streamline daily tasks for bankers. 

• Foster collaboration: A feedback loop between technology and business stakeholders ensured ongoing model improvement. 

• Drive continuous improvement: The data science team actively incorporated banker feedback to refine models and maintain their relevance. 

This iterative process allowed us to enhance the efficiency and effectiveness of our data-driven solutions. 

Enterprise-wide impact 

We want to make these insights and relationships available enterprise-wide. This entails integrating feedback into the canonical model, enriching the mastering database, and enabling seamless data flow across the broader ecosystem. By establishing this interconnected infrastructure, we aim to maximize the utility and accessibility of insights derived from the data, empowering stakeholders across the financial institution to make informed decisions and drive value creation. 

What the future holds for our client 

Focusing on future developments, we have outlined a vision for enhancing the platform to serve the bank's needs better. Central to this vision is the creation of a unique relationship ID, which complements the existing unique customer ID. This relationship ID would allow for multiple views of customer relationships, enabling different divisions to tailor the data according to their specific requirements. 

Our client says, "The Relationship ID, that unique relationship ID beyond the unique customer ID, and the way that we link these relationships is an asset to us as a company that can be seen through a marketing view, a risk view, a sales and service view. We anticipate having multiple views of relationships available within our company that each of our division leaders can mold and shape to meet their needs." 

The relationship ID is a valuable asset with potential marketing, risk assessment, sales, and service applications. This enhanced data management approach is expected to provide a solid foundation for AI and ML models, improving their accuracy and efficiency and ultimately driving growth and innovation. "But what we can track at an enterprise level is really critical to moving forward and building amazing AI machine learning models on top of. I just don't see how you do it without having that strong MDM foundation." 

As the collaboration between Mastech InfoTrellis and the bank progresses, the aim is to empower relationship officers to navigate customer relationships more effectively and efficiently. By providing accurate and accessible data through the relationship ID, officers can make more informed decisions, leading to improved customer experiences and more robust, more profitable connections.