Transforming home ownership with Amazon Transcribe Call Analytics, Amazon Comprehend, and Amazon Bedrock: Rocket Mortgage’s journey with AWS

This post is co-written with Josh Zook and Alex Hamilton from Rocket Mortgage.

Rocket Mortgage, America’s largest retail mortgage lender, revolutionizes homeownership with Rocket Logic – Synopsis, an AI tool built on AWS.  This innovation has transformed client interactions and operational efficiency through the use of Amazon Transcribe Call Analytics, Amazon Comprehend, and Amazon Bedrock. Through Rocket Logic – Synopsis, Rocket achieved remarkable results: automating post call interaction wrap-up resulting in a projected 40,000 team hours saved annually, and a 10% increase in first-call resolutions saved 20,000 hours annually. In addition to Rocket Logic – Synopsis, 70% of servicing clients choose to self-serve over Gen AI powered mediums such as IVR. Rocket’s “start small, launch and learn, scale fast” approach paired with AWS enablement proved effective, deploying 30,000 servicing calls in 10 days, then scaling four times greater for operations and six times greater for banking.

This post offers insights for businesses aiming to use artificial intelligence (AI) and cloud technologies to enhance customer service and streamline operations. We share how Rocket Mortgage’s use of AWS services set a new industry standard and demonstrate how to apply these principles to transform your client interactions and processes with speed and scalability.

Opportunities for innovation

Rocket services over 2.6 million clients, with 65 million voice interactions and 147 million voice minutes inclusive of banking, operations, and servicing, and generates and processes over 10 PB of data. By focusing on three key personas—clients, client advocates, and business leaders or senior leadership—Rocket aims to create a solution that enhances experiences across the board.

At the heart of this transformation is the recognition that clients value their time, but also benefit from hyper-personalized support in ultra complex moments. With call volumes on the rise, solving this problem at scale was essential. Rocket tapped into a crucial insight: 81% of consumers prefer self-service options. This preference opens exciting possibilities for swift, efficient problem-solving. Imagine a world where answers are available at your fingertips, 24/7, without the need to wait in a queue. By implementing enhanced self-service tools, Rocket is poised to offer faster resolution times, greater client autonomy, and a more satisfying overall experience.

Client advocates, the face of the company, stand to benefit significantly from this transformation. Currently, client advocates spend about 30% of their time on administrative tasks. By streamlining processes, client advocates can focus on what they do best: providing exceptional customer service and nurturing client relationships. This shift promises more engaging work, increased job satisfaction, and opportunities for skill development. Rocket envisions their client advocates evolving into trusted advisors, handling complex inquiries that truly take advantage of their expertise and interpersonal skills.

For business leaders, this wealth of data on trends, sentiment, and performance opens up a treasure trove of opportunities. Decision-makers can now drive significant improvements across the board, employing data-driven strategies to enhance customer satisfaction, optimize operations, and boost overall business performance. Business leaders can look forward to leading more efficient teams, and senior leadership can anticipate improved client loyalty and a stronger bottom line.

Strategic requirements

To further elevate their client interactions, Rocket identified key requirements for their solution. These requirements were essential to make sure the solution could handle the demands of their extensive client base and provide actionable insights to enhance client experiences:

• Sentiment analysis – Tracking client sentiment and preferences was necessary to offer personalized experiences. The solution needed to accurately gauge client emotions and preferences to tailor responses and services effectively.
• Automation – Automating routine tasks, such as call summaries, was essential to free up team members for more meaningful client interactions. This automation would help reduce the manual workload, allowing the team to focus on building stronger client relationships.
• AI integration – Using generative AI to analyze calls was crucial for providing actionable insights and enhancing client interactions. The AI integration needed to be robust enough to process vast amounts of data and deliver precise, meaningful results.
• Data security – Protecting sensitive client information throughout the process was a non-negotiable requirement. Rocket needed to uphold the highest standards of data security, maintaining regulatory compliance, data privacy, and the integrity of client information.
• Compliance and data privacy – Rocket required a solution that met strict compliance and data privacy standards. Given the sensitive nature of the information handled, the solution needed to provide complete data protection and adhere to industry regulations.
• Scalability – Rocket needed a solution capable of handling millions of calls annually and scaling efficiently with growing demand. This requirement was vital to make sure the system could support their expansive and continuously increasing volume of voice interactions.

Solution overview

To meet these requirements, Rocket partnered with the AWS team to deploy the AWS Contact Center Intelligence (CCI) solution Post-Call Analytics, branded internally as Rocket Logic – Synopsis. This solution seamlessly integrates into Rocket’s existing operations, using AI technologies to transcribe and analyze client calls. By utilizing services like Amazon Transcribe Call Analytics, Amazon Comprehend, and Amazon Bedrock, the solution extracts valuable insights such as sentiment, call drivers, and client preferences, enhancing client interactions and providing actionable data for continuous improvement.

At the heart of Rocket are their philosophies, known as their -ISMs, which guide their growth and innovation.  One of these guiding principles is “launch and learn.”

Embracing the mantra of “think big but start small,” Rocket adopted a rapid, iterative approach to achieve a remarkable time to market of just 10 days, compared to the months it would have traditionally taken. This agile methodology allowed them to create space for exploration and innovation. The team initially focused on a few key use cases, starting simple and rapidly iterating based on feedback and results.

To accelerate development and make sure data was quickly put into the hands of the business, they utilized mechanisms such as a hackathon with targeted goals. By using existing solutions and AWS technical teams, Rocket significantly reduced the time to market, allowing for swift deployment. Additionally, they looked to industry tactics to find solutions to common problems, so their approach was both innovative and practical.

During this “launch and learn” process, Rocket anticipated and managed challenges such as scaling issues and burst volume management using Drip Hopper and serverless technologies through AWS. They also fine-tuned the Anthropic’s Claude 3 Haiku large language model (LLM) on Amazon Bedrock for call classification and data extraction.

The following diagram illustrates the solution architecture.

Post-Call Analytics provides an entire architecture around ingesting audio files in a fully automated workflow with AWS Step Functions, which is initiated when an audio file is delivered to a configured Amazon Simple Storage Service (Amazon S3) bucket. After a few minutes, a transcript is produced with Amazon Transcribe Call Analytics and saved to another S3 bucket for processing by other business intelligence (BI) tools. These transcripts are saved for further processing by BI tools, with stringent security measures making sure personally identifiable information (PII) is redacted and data is encrypted.  The PII is redacted throughout, but client ID and interaction ID are used to correlate and trace across the data sets.  Downstream applications use those ids to pull from client data services in the UI presentation layer.

Enhancing the analysis, Amazon Comprehend is used for sentiment analysis and entity extraction, providing deeper insights into client interactions. Generative AI is integrated to generate concise call summaries and actionable insights, significantly reducing the manual workload and allowing team members to focus on building stronger client relationships. This generative AI capability, powered by Amazon Bedrock, Anthropic’s Claude Sonnet 3, and customizable prompts, enables Rocket to deliver real-time, contextually relevant information. Data is securely stored and managed within AWS, using Amazon S3 and Amazon DynamoDB, with robust encryption and access controls provided by AWS Key Management Service (AWS KMS) and AWS Identity and Access Management (IAM) policies. This comprehensive setup enables Rocket to efficiently manage, analyze, and act on client interaction data, thereby enhancing both client experience and operational efficiency.

Achieving excellence

The implementation of Rocket Logic – Synopsis has yielded remarkable results for Rocket:

• Efficiency gains – Automating call transcription and sentiment analysis is projected to save the servicing team nearly 40,000 hours annually
• Enhanced client experience – Approximately 70% of servicing clients fully self-serve over Gen AI powered mediums such as IVR; allowing clients to resolve inquiries without needing team member intervention
• Increased first-call resolutions – There has been a nearly 10% increase in first-call resolutions, saving approximately 20,000 team member hours annually
• Proactive client solutions – The tool’s predictive capabilities have improved, allowing Rocket to proactively address client needs before they even make a call
• Start small, launch and learn, scale fast – Rocket started with 30,000 servicing calls with a 10-day time to market, and then scaled four times greater for operations, followed by six times greater for banking

Roadmap

Looking ahead, Rocket plans to continue enhancing Rocket Logic – Synopsis by using the vast amount of data gathered from call transcripts. Future developments will include:

• Advanced predictive analytics – Further improving the tool’s ability to anticipate client needs and offer solutions proactively
• Omnichannel integration – Expanding the AI capabilities to other communication channels such as emails and chats
• Client preference tracking – Refining the technology to better understand and adapt to individual client preferences, providing more personalized interactions
• Enhanced personalization – Utilizing data to create even more tailored client experiences, including understanding preferences for communication channels and timing

Conclusion

The collaboration between Rocket Mortgage and AWS has revolutionized the homeownership process by integrating advanced AI solutions into client interactions. Rocket Logic – Synopsis enhances operational efficiency significantly and improves the client experience. As Rocket continues to innovate and expand its AI capabilities, they remain committed to providing personalized, efficient, and seamless homeownership experiences for their clients. The success of Rocket Logic – Synopsis demonstrates the transformative power of technology in creating more efficient, responsive, and personalized client experiences. To learn more, visit Amazon Transcribe Call Analytics, Amazon Comprehend, and Amazon Bedrock.

---

About the authors

Josh Zook is the Chief Technology Officer of Rocket Mortgage, working alongside the teams that are shipping the products that clients and partners are using every day to make home ownership a reality. He started in Technology in 1984 by writing a program in BASIC to calculate his weight on the moon using an Apple IIe. Since then, he has been on a relentless pursuit in using technology to make life easier by solving slightly more complex problems. Josh believes the key to success is curiosity combined with the grit and grind to make ideas reality. This has led to a steady paycheck since he was 10 years old, with jobs in landscaping, sandwich artistry, sporting goods sales, satellite installation, firefighter, and bookstore aficionado… just to name a few.

Alex Hamilton is a Director of Engineering at Rocket Mortgage, spearheading the AI driven digital strategy to help everyone home. He’s been shaping the tech scene at Rocket for over 11 years, including launching one of the company’s first models to boost trading revenue and bring modern event streaming and containerization to Rocket. Alex is passionate about solving novel engineering problems and bringing magical client experiences to life. Outside of work Alex enjoys traveling, weekend brunch, and firing up the grill!

Ritesh Shah is a Senior Worldwide GenAI Specialist at AWS. He partners with customers like Rocket to drive AI adoption, resulting in millions of dollars in top and bottom line impact for these customers. Outside work, Ritesh tries to be a dad to his AWSome daughter.  Connect with him on LinkedIn.

Venkata Santosh Sajjan Alla is a Senior Solutions Architect at AWS Financial Services, where he partners with North American FinTech companies like Rocket to drive cloud strategy and accelerate AI adoption. His expertise in AI & ML, and cloud native architecture has helped organizations unlock new revenue streams, enhance operational efficiency, and achieve substantial business transformation. By modernizing financial institutions with secure, scalable infrastructures, Sajjan enables them to stay competitive in a rapidly evolving, data-driven landscape. Outside of work, he enjoys spending time with his family and is a proud father to his daughter.

from AWS Machine Learning Blog https://ift.tt/odp3NBS