Enterprise
LoanPro cuts call times 15% with multi-agent AI on Bedrock
The lending platform built specialized agents that synthesize borrower data instantly, reducing manual navigation and support escalations.
Key takeaways
- LoanPro reduced customer call times by up to 15 percent and internal support tickets by 8 percent using a multi-agent AI assistant built on Amazon Bedrock.
- The system synthesizes around 1,400 account fields into narrative summaries in under 15 seconds, replacing three to five minutes of manual navigation.
- Specialized agents handle discrete tasks—data retrieval, action selection, and routing—making each component easier to evaluate and improve independently.
- The assistant is embedded in LoanPro's existing web portal and logs every interaction for compliance, with Amazon Bedrock Guardrails enforcing content filtering.
- LoanPro serves over 450 lenders with loans reaching approximately 31 percent of U.S. households, making efficiency gains at this scale structurally significant.
When a borrower calls their lender, the servicing agent on the other end faces a familiar scramble: pulling account data from dozens of fields, scanning transaction histories, and checking workflow queues before they can actually help. That manual assembly process burns minutes on every call and creates inconsistent service quality.
LoanPro, an API-first lending platform that serves as the operating system for more than 450 lenders, set out to eliminate that friction. Working with AWS Partner AllCloud and the AWS Generative AI Innovation Center, the company built a multi-agent AI assistant that gives servicing agents immediate access to synthesized borrower context and actionable recommendations. The result: customer call times dropped by up to 15 percent, according to AllCloud's published case study.
Why it matters
Loan servicing operates at scale—LoanPro reports its platform reaches roughly 31 percent of U.S. households. Across the 800-plus servicing professionals at early launch partners, agents previously spent 18 to 20 percent of every call just orienting to accounts. That's 1 to 2.5 minutes across five to seven clicks. The AI assistant consolidates that context into a single click, reclaiming up to 9 minutes of agent capacity per operating hour per agent. Multiply that across hundreds of agents, and the efficiency gain becomes structural.
The architecture
LoanPro built the assistant on Amazon Bedrock, using Anthropic's Claude Sonnet, Haiku, and Opus models. The system is orchestrated with Amazon Bedrock AgentCore, a managed platform for building and scaling agents, and uses the Strands Agents SDK, an open-source framework for building AI agents with tool use.
Rather than deploying a single monolithic model, the team designed a multi-agent system. Specialized agents handle discrete tasks: one retrieves data and generates account summaries, another selects workflow actions and recommendations, and a composer agent routes requests to the right specialist. This separation makes each agent easier to evaluate, tune, and extend independently.
The agents retrieve borrower data through tool use, a capability in Amazon Bedrock for calling external functions. Two primary tools power the system: a data retrieval tool that fetches account details, payment histories, and loan terms from LoanPro's data layer—synthesizing around 1,400 fields per account—and an invoke API tool that triggers workflow actions within the platform.
Amazon Bedrock AgentCore Gateway serves as the secure access layer between agents and tools. It exposes LoanPro's APIs over the Model Context Protocol, handles authentication, and manages connections. AWS Lambda provides the serverless integration layer, scaling automatically with request volume.
Production results
Beyond the 15 percent reduction in call times, lenders on the platform reported an 8 percent decrease in internal support ticket volumes as agents resolved more cases during the initial call. Account summaries now generate in under 15 seconds, replacing what previously took agents three to five minutes of manual navigation.
The assistant is embedded directly into LoanPro's existing web portal. When an agent opens a borrower account, the system automatically generates a narrative summary of account health, recent activity, and outstanding issues. Agents can also ask follow-up questions in natural language through a conversational chat interface.
Every interaction is logged, maintaining an auditable record. Amazon Bedrock Guardrails apply content filtering and grounding validation alongside LoanPro's compliance guardrails and role-based access controls.
The team validated agent accuracy against a benchmark of 150 accounts labeled by subject-matter experts, holding the system to a 99 percent accuracy target, which it met.
What's next
LoanPro and AllCloud are expanding the collaboration to develop additional agents for collections and origination support. The modular architecture makes this practical—new agents can be built and deployed independently using the same infrastructure.
Details of the implementation were first reported by AWS in a blog post co-authored by Mariah Miller from AWS and Colin Terry from LoanPro.