Rocket Companies, a Detroit-based FinTech company, developed Rocket AI Agent to address the overwhelming complexity of the home buying process by providing 24/7 personalized guidance and support. Built on Amazon Bedrock Agents, the AI assistant combines domain knowledge, personalized guidance, and actionable capabilities to transform client engagement across Rocket's digital properties. The implementation resulted in a threefold increase in conversion rates from web traffic to closed loans, 85% reduction in transfers to customer care, and 68% customer satisfaction scores, while enabling seamless transitions between AI assistance and human support when needed.
Rocket Companies is a Detroit-based FinTech company operating in the mortgage lending and home ownership space with a mission to “Help Everyone Home.” The company has established itself as more than just a mortgage lender, extending its services across the entire home ownership journey including home search, purchasing, financing, and equity management. With 11 petabytes of data and a focus on technology-driven solutions, Rocket recognized an opportunity to leverage generative AI to address the persistent challenge of making home buying less overwhelming for clients.
The core problem Rocket sought to solve was providing trusted guidance to clients at any hour and on any channel, recognizing that the home buying process remains complex and intimidating despite technological advances. This led to the development of Rocket AI Agent, a conversational AI assistant designed to transform client engagement across Rocket’s digital properties.
The Rocket AI Agent represents a sophisticated implementation of Amazon Bedrock Agents, utilizing a modular architecture that supports scalability and job-specific precision. At its core, the system operates through a centralized capability deployed across Rocket’s suite of digital properties, currently comprising eight domain-specific agents that focus on distinct functions such as loan origination, servicing, and broker support.
The architecture follows a structured workflow beginning with client initiation through chat functionality within Rocket’s mobile app or web pages. The Rocket AI Agent API provides a unified interface that routes requests to the appropriate Amazon Bedrock agent based on static criteria such as the originating web or mobile property, or through LLM-based intent identification. Once routed, the agent processes the request by breaking tasks into subtasks, determining the appropriate sequence, and executing actions while leveraging Rocket’s proprietary data through knowledge bases.
Amazon Bedrock Agents serves as the foundation, providing a fully managed capability that extends foundation models using the Reasoning and Acting (ReAct) framework. This allows the agents to interpret user intent, plan and execute tasks, and integrate seamlessly with enterprise data and APIs. The system leverages several key capabilities including agent instructions that define the agent’s role and objectives, Amazon Bedrock Knowledge Bases for fast information retrieval from Rocket’s Learning Center and proprietary documents, action groups for secure operations like submitting leads or scheduling payments, agent memory for contextual awareness across multiple conversation turns, and Amazon Bedrock Guardrails to ensure responsible AI practices.
A critical component of the implementation involves careful data curation and management. Rocket built their enterprise knowledge base using Amazon Bedrock Knowledge Bases, which internally utilizes Amazon Kendra for retrieval across Rocket’s content libraries including FAQs, compliance documents, and servicing workflows. The quality of responses generated by the AI system is directly tied to the quality and structure of this source data, making data curation a foundational element of the LLMOps strategy.
The system maintains contextual awareness by knowing what page a client was viewing and tailoring responses based on this context. This level of personalization is achieved through integration with Rocket’s proprietary data and user context, enabling the AI agent to provide real-time answers about mortgage options, rates, documents, and processes while offering guided self-service actions such as filling out preapproval forms or scheduling payments.
One of the key lessons learned during implementation was the importance of limiting each agent’s scope to 3-5 actions, which led to more maintainable, testable, and high-performing agents. For example, the payment agent focuses exclusively on tasks like scheduling payments and providing due dates, while the refinance agent handles rate simulations and lead capture. Each agent’s capabilities are implemented using Amazon Bedrock action groups with well-documented interfaces and monitored task resolution rates.
This modular approach allows for better testing, maintenance, and optimization of individual components while maintaining overall system coherence. The domain-specific nature of each agent ensures that responses are accurate and relevant to the specific context of the client’s needs within the home buying journey.
Rocket implemented cross-Region inference from the start of development to provide scalable, resilient model performance. This architectural decision allows inference requests to be routed to the optimal AWS Region within the supported geography, improving latency and model availability by automatically distributing load based on capacity. During peak traffic periods such as product launches or interest rate shifts, this architecture has enabled Rocket to avoid Regional service quota bottlenecks, maintain responsiveness, and increase throughput by leveraging compute capacity across multiple AWS Regions.
The system is designed to handle bursty, unpredictable load conditions while maintaining a smooth and consistent user experience. This is particularly important in the mortgage industry where demand can fluctuate significantly based on market conditions and external factors.
Amazon Bedrock Guardrails play a crucial role in supporting Rocket’s responsible AI goals by ensuring that agents stay within appropriate topic boundaries. The system includes uncertainty thresholds using confidence scores and specific keyword triggers to detect when an interaction might require human assistance. In such cases, the Rocket AI Agent proactively transitions the session to a live support agent or provides the user with escalation options.
This approach to graceful escalation is positioned not as a failure of the AI system, but as a critical component of building user trust. The system avoids frustrating conversational loops and ensures that complex or sensitive interactions receive appropriate human care, which is essential in the high-stakes environment of mortgage lending and home buying.
The implementation has demonstrated significant measurable impact across multiple dimensions. The most notable result is a threefold increase in conversion rates from web traffic to closed loans, attributed to the AI agent’s ability to capture leads around the clock, even outside traditional business hours. This represents a substantial improvement in lead conversion efficiency and revenue generation.
Operational efficiency gains have been particularly pronounced through chat containment, with an 85% decrease in transfers to customer care and a 45% decrease in transfers to servicing specialists. This reduction in handoffs to human agents has freed up team capacity to focus on more complex, high-impact client needs while maintaining service quality.
Customer satisfaction scores have reached 68% of clients providing high satisfaction ratings across servicing and origination chat interactions. The top drivers of satisfaction include quick response times, clear communication, and accurate information, all contributing to greater client trust and reduced friction in the home buying process.
Rocket is advancing toward multi-agent collaboration as the next phase of their agentic AI strategy. This evolution will enable the orchestration of agents across domains to deliver intelligent, end-to-end experiences that mirror the complexity of real client journeys. The multi-agent approach promises several benefits including end-to-end personalization through context sharing and coordination between domain-specific agents, back-office integration for automating document verification and workflow processes, seamless context switching between servicing and origination within single conversations, and sophisticated orchestration of multistep tasks spanning multiple business units.
This represents a significant evolution from reactive support to proactive workflow automation, positioning Rocket to handle entire client journeys from discovery and qualification to servicing and beyond through coordinated AI assistance.
The development and deployment process revealed several critical insights for organizations building generative AI applications at scale. Data quality emerged as paramount, with the recognition that response quality is directly tied to source data quality and structure. The team learned that assigning tight scopes to individual agents improves maintainability and performance, while graceful escalation mechanisms are essential for maintaining user trust.
User behavior evolution was identified as a constant factor, with clients interacting with the system in unexpected ways and usage patterns changing over time. This necessitated significant investment in observability and user feedback loops to enable rapid adaptation. The early implementation of cross-Region inference proved crucial for handling variable load conditions and maintaining consistent performance.
The AI agent is embedded directly into Rocket’s web and mobile services, delivering support exactly when and where clients need it. This integration extends across the majority of Rocket’s web pages and mobile apps, supporting clients during loan origination, servicing, and within Rocket’s third-party broker system (Rocket Pro). The seamless integration ensures that AI assistance is available wherever clients interact with Rocket digitally.
The system’s ability to perform meaningful actions on behalf of clients, such as submitting leads or scheduling payments, demonstrates sophisticated integration with Rocket’s backend services and APIs. This level of integration transforms the AI from a simple chatbot into a capable digital assistant that can drive real business outcomes.
The implementation includes expanded language support, notably Spanish-language assistance, to better serve a diverse and growing demographic. This multilingual capability reflects Rocket’s commitment to accessibility and inclusion in their mission to “Help Everyone Home.” The 24/7 availability combined with multilingual support significantly expands the reach and utility of the AI assistant.
The focus on accessibility extends beyond language to include intuitive, personalized self-service capabilities that adapt to each client’s stage in the homeownership journey and their preferences. This personalized approach allows clients to engage with the system on their own terms while maintaining the option to escalate to human bankers when needed.
Snorkel developed a specialized benchmark dataset for evaluating AI agents in insurance underwriting, leveraging their expert network of Chartered Property and Casualty Underwriters (CPCUs). The benchmark simulates an AI copilot that assists junior underwriters by reasoning over proprietary knowledge, using multiple tools including databases and underwriting guidelines, and engaging in multi-turn conversations. The evaluation revealed significant performance variations across frontier models (single digits to ~80% accuracy), with notable error modes including tool use failures (36% of conversations) and hallucinations from pretrained domain knowledge, particularly from OpenAI models which hallucinated non-existent insurance products 15-45% of the time.
DoorDash faced challenges in scaling personalization and maintaining product catalogs as they expanded beyond restaurants into new verticals like grocery, retail, and convenience stores, dealing with millions of SKUs and cold-start scenarios for new customers and products. They implemented a layered approach combining traditional machine learning with fine-tuned LLMs, RAG systems, and LLM agents to automate product knowledge graph construction, enable contextual personalization, and provide recommendations even without historical user interaction data. The solution resulted in faster, more cost-effective catalog processing, improved personalization for cold-start scenarios, and the foundation for future agentic shopping experiences that can adapt to real-time contexts like emergency situations.
LinkedIn developed Hiring Assistant, an AI agent designed to transform the recruiting workflow by automating repetitive tasks like candidate sourcing, evaluation, and engagement across 1.2+ billion profiles. The system addresses the challenge of recruiters spending excessive time on pattern-recognition tasks rather than high-value decision-making and relationship building. Using a plan-and-execute agent architecture with specialized sub-agents for intake, sourcing, evaluation, outreach, screening, and learning, Hiring Assistant combines real-time conversational interfaces with large-scale asynchronous execution. The solution leverages LinkedIn's Economic Graph for talent insights, custom fine-tuned LLMs for candidate evaluation, and cognitive memory systems that learn from recruiter behavior over time. The result is a globally available agentic product that enables recruiters to work with greater speed, scale, and intelligence while maintaining human-in-the-loop control for critical decisions.
