AWS for Industries

Broadridge Advances the Proxy Voting Data Process Using Artificial Intelligence on AWS

How does a company build a solution to a complex business challenge when the required technology and skills are new to the organization? Global fintech company Broadridge Financial Solutions (Broadridge) faced this question as it considered how to improve the operational efficiency of its Proxy Policies and Insights (PPI) tool.

Broadridge has used Amazon Web Services (AWS) since 2017. It developed projects using support from AWS Professional Services, which supplements customer teams with specialized skills and experience to achieve results on AWS.

In late 2019, when Broadridge decided to automate part of its PPI tool using artificial intelligence (AI), the company turned to AWS for technical solutions and capacity building. Broadridge’s PPI team used AWS services and consultation to develop an AI architecture that would reduce manual effort and deliver faster data access to customers.

Broadridge built its solution using Amazon Comprehend and Amazon Textract. Amazon Comprehend is a natural-language processing service that uses machine learning to uncover information in unstructured data. Amazon Textract, a machine learning service, automatically extracts text, handwriting, and data from scanned documents, forms, and tables.


Reducing Complexity in Document Processing

 Broadridge specializes in investor communications, technology-driven solutions, and data analytics. It supports proxy voting services for most of the public companies in North America. The company also handles millions of trades per day involving trillions of dollars, manages shareholder voting in 120 countries, and processes over six billion customer communications annually.

Broadridge’s PPI team uses a manual maker-checker system to process U.S. Securities and Exchange Commission filings. It also uses that system to extract 130 proxy data points from 300,000 meetings held over the past 10 years. The proxy information supplements other existing data and provides insights to broker-dealers, institutional investors, mutual funds, retail investors, regulators, and academics. These insights inform voting decisions and help institutional investors execute smarter governance.

 The manual maker-checker system enables a high degree of accuracy, but the process is complex, labor intensive, and difficult to scale. The PPI operations team spends thousands of hours annually on extracting and validating proxy data.

 Broadridge initially tried to automate the process using regular expressions and other approaches that would seek out data points and relevant keywords, but the data points were too complex. The company also considered using open-source AI models but worried about maintaining the models and developing the underlying infrastructure.


Designing a Company-Driven Solution

 AI was a new technology for Broadridge, and the company turned to AWS to learn how to use AI solutions and design a scalable and extensible AI architecture. In early 2020, Broadridge participated in a 3-day workshop facilitated by multiple AWS teams, including AWS Professional Services.

The company used expertise from the participating AWS teams to define the project’s scope and align on how to measure and validate accuracy for an extensible, event-driven serverless architecture. The first phase took approximately 35 data points and aimed for 70 percent accuracy in processing. The second phase focused on putting the models into production.

 The initiative relied on executive and management support, which included leadership from Martin Koopman, bank broker-dealer and president of Investor Communication Solutions at Broadridge. Koopman held a Broadridge AI summit and organized AWS DeepRacer events, which provide hands-on machine learning experience through a cloud-based, 3D racing simulator and fully autonomous 1/18-scale race car.

 The AWS consultation process helped Broadridge understand its skill gap and build a new data science and architecture team to move the project forward. During each phase of the project, Broadridge used support from AWS Professional Services to train the new team and troubleshoot issues. Broadridge conducted an AWS Well-Architected Review to identify focus areas prior to user acceptance testing. The review applies the principles of the AWS Well-Architected Framework, which was developed to help cloud architects build secure, high-performing, resilient, and efficient infrastructure for their applications.

 Broadridge developed a fully managed serverless architecture that uses AWS Lambda for pre- and postprocessing and AWS Step Functions to orchestrate the pipelines. AWS Lambda is a serverless compute service that lets customers run code without provisioning or managing servers. AWS Step Functions is a visual workflow builder used to orchestrate AWS services, automate processes, and build serverless applications. The solution’s key AI building blocks—Amazon Textract and Amazon Comprehend—assemble the results.


Building an AI Solution That Saves Time and Reduces Costs

 Using the event-driven serverless architecture on AWS, Broadridge automated data extraction and achieved the required 70 percent accuracy threshold. In early 2021, the company launched the automated process in production. While Broadridge continues to manually audit the data, it expects the automated process to save thousands of hours annually. This will help Broadridge save costs and achieve scale.

With faster document processing, Broadridge expects to deliver data more quickly to its customers and reduce its manual process delivery time of 1–7 days. As Broadridge continues to automate data processing, it hopes to further increase scalability and expand the number of data points available to customers, delivering deeper insights.

 Automating the maker segment of the maker-checker system also helps Broadridge enhance voter experience by personalizing voting insights according to issues that matter to clients. Regulators, academics, and nonprofit organizations use the PPI tool for free.


Embracing an AI Mindset

Broadridge plans to use AWS Professional Services as it works on two new use cases for data extraction. The first focuses solely on data extraction. The second involves building a reference architecture to process document types in any format using AI services on AWS.

 Broadridge is also building AI engagement across the organization. The enterprise architecture team has led several company lunch-and-learn events to explain AI on AWS. As the company continues to innovate and scale, it recognizes that increasing AI awareness and skill in the organization is both a realized benefit and an area for future growth.

Saumin Patel

Saumin Patel

Saumin Patel is VP, Enterprise Architecture at Broadridge Financial Solutions, based in Greater New York City area. He leads Operational Efficiency team helping various business and operations teams in automating manual, siloed processes leveraging AWS AI and cloud services. Additionally, he helps with Broadridge's digital transformation journey and other enterprise architecture areas. Outside of work, he enjoys spending time with family, playing cricket, and volunteering with various local and religious charities.

Gopi Krishnamurthy

Gopi Krishnamurthy

Gopi Krishnamurthy is a senior Solutions Architect at Amazon Web Services based in New York City. He works with large financial services customers as their trusted advisor to transform their workloads and migrate to the cloud. His core interests include serverless technologies and machine learning. Outside of work, he likes to spend time with his family and explore a wide range of music.

Winnie Tung

Winnie Tung

Winnie Tung is a Machine Learning Engineer at AWS Professional Services. Winnie has over 30 years of experience solving some of the world’s most difficult technical problems in the financial services industry. Since joined AWS, she specialized in developing Al/ML solutions for the real world. She helps customers to operationalize and management Al/ML solutions at scale.