AIToday
Large Language ModelsAI Business & IndustryAmazon AI BlogPublished: Nov 18, 2026, 04:00 JST

LoanPro AI cuts call times up to 15 percent on Amazon Bedrock AgentCore

Send on LINE
LoanPro AI cuts call times up to 15 percent on Amazon Bedrock AgentCore

3 Key Points

  1. What happened

    LoanPro built a multi-agent AI loan-servicing assistant on Amazon Bedrock and AgentCore. Early results show customer call times reduced by up to 15 percent, per AllCloud's published case study.

  2. Why it matters

    Servicing agents now get synthesized borrower context and next-step recommendations in one click instead of assembling it manually, which LoanPro says has changed how they start every call.

WHO IT HITSLoan servicing agents at lenders on LoanPro's platform benefit most — they open an account and see a full summary with recommended next actions instead of navigating screens — and compliance teams gain an auditable record of each AI-assisted interaction.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

LoanPro is not a lender but an operating system that lenders run on, serving over 450 lenders according to the company. That platform position shaped its AI strategy: the assistant is designed to benefit the institutions across the platform, not a single lending operation. The build ran through two parallel workstreams, one delivering the servicing assistant now in production and another advising on an existing LLM proof of concept that converted natural language into rule automation code for LoanPro's Clojure rules engine. AllCloud reviewed that POC's architecture and recommended a path to production, including a documented architecture and risk assessment.

The technical stack relies on Amazon Bedrock for model access and AgentCore for a managed agent runtime, with a composer agent routing requests to specialized agents built on the Strands Agents SDK. Tools reach borrowers' data through an MCP server on AgentCore Gateway, and a Lambda function validates each caller's signature against LoanPro's single sign-on service. Because lending is regulated, role-based access controls and compliance guardrails apply to every request, and AI actions are written to LoanPro's audit trail and attributed to a named person. The team chose managed services over building the runtime in-house, removing operational overhead and allowing new models to be adopted without separate endpoints.

Beyond the headline call-time reduction, the assistant decreased internal ticket volumes by nearly 8 percent, according to LoanPro, with lenders reporting fewer support escalations as agents resolve more cases on the first call. The same context and recommendations reach agents regardless of experience level, reducing variability tied to individual tenure. The modular architecture is what makes the planned expansion into collections and origination support practical — new agents can be built and deployed independently on the same AgentCore infrastructure without redesigning the core system.

FAQ
How accurate is the LoanPro AI assistant?
It met a 99 percent accuracy target against a benchmark of 150 SME-labeled accounts, measuring data retrieval, summarization and action selection.
How fast does the assistant generate account summaries?
Account summaries are generated in under 15 seconds, according to LoanPro internal benchmarks, replacing what previously took agents 3–5 minutes of manual navigation.
Is the assistant available in LoanPro's existing interface?
Yes, it is embedded directly into LoanPro's existing web portal, with an account summary view and a conversational chat interface that are both logged for audit.
Amazon AI BlogRead Original Article
Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleMicrosoft, Salesforce launch Copilotforce with 38+ sales skills