United StatesFinance and insurance2025 – ongoingLendingTree
Multi-agent mortgage assistant on Amazon Bedrock
LendingTree's AI mortgage assistant resolved over 97% of conversations without a human
Problem
Borrowers face choices on loan type, term and rate, and unfamiliar terms such as discount points and debt-to-income ratio. Basic chatbots answered simple questions. LendingTree wanted an assistant that could handle harder questions and match borrowers with offers, under mortgage rules on content filtering and personal data.
Approach
LendingTree built three agents on Amazon Bedrock. A supervisor agent plans each request and routes it. An education agent answers mortgage questions from indexed documents. A matching agent collects preferences and calls LendingTree's offer, eligibility and rate services. Guardrails screen inputs and outputs for content and personal data. The assistant has run in production since late 2025.
Results
Takeaway
About 75% of early conversations were educational, and over half of recent ones involved rate comparison or lender matching. The figures cover launch through the first quarter of 2026 and come from AWS's write-up with LendingTree.
Categories
- Industry
- Finance and insurance
- Business function
- Customer serviceMarketing and sales
- Innovation type
- Product and service innovation
- AI technology
- AI agentsGenerative AI
- Adoption
- Off-the-shelf tool or platform
- Evidence
- Vendor case study
Sources
Source summaries
An AWS blog post on the design of LendingTree's supervisor, education and matching agents. It reports about 1,960 conversations and 12,100 messages from launch through Q1 2026, with over 97% resolved without human escalation.
Full details are in the original sources.