How Finsure’s 4,000+ brokers match loans against more than 500 credit policies in seconds

At a glance

Finding the right loan for a client means checking their circumstances against the eligibility rules of hundreds of lenders. Finsure’s 4,000+ brokers and advisers now do that through WikiBrokerAI, an agentic loan processing assistant that went into production in August 2026. It covers more than 500 credit policies from more than 190 lenders across Australia and New Zealand.

Key outcomes
  • Evidence-based matching in seconds across more than 500 lender credit policies
  • Live in production across Australia and New Zealand, August 2026
  • Available standalone and built into Finsure’s Metanoia lending platform
  • Governance and guardrails in place for production use at scale

The challenge

Finsure gives its brokers access to more than 190 lenders. Each one sets its own serviceability rules, acceptable income types, loan-to-value ratio (LVR) limits, location restrictions, and documentation standards. Those rules change constantly, and they sit scattered across hundreds of pages of policy PDFs.

Even experienced brokers struggle to keep up. Policy matching is manual work, and complex cases take real time to work through. It’s also an early step in the mortgage application lifecycle, so its effects carry through everything downstream. A broker who can’t confirm eligibility may turn a client away when an automated evidence-based check would have shown the loan was serviceable.

The solution

A strategic partnership with Mantel

WikiBrokerAI does the policy matching. A broker describes a client’s circumstances in a chat interface, and the assistant returns evidence-based matches drawn from more than 500 credit policies across 190 lenders, with the supporting policy text attached.

Brokers in Australia and New Zealand can use it as a standalone product or inside Metanoia, Finsure’s new end-to-end loan services platform, built with Mantel. The embedded version sits on a design system built to carry AI features generally, not just this one, so the policy search is there while a broker is with a customer rather than in a separate tool they have to break off to open. The same system supports whatever Finsure builds next.

It runs in both Australian and New Zealand regions as a production-grade solution, with foundations, governance, and guardrails in place to operate safely at scale.

How we delivered it

Finsure and Mantel worked through five phases:

  • AI use case roadmap: rapid ideation, prioritisation, and scoring of candidate use cases
  • Proof of concept: delivered against both technical and functional success criteria
  • Pilot: an uplift to the user interface, production-grade foundations established, and production user testing with a subset of brokers
  • WikiBrokerAI to production: feature enhancement, testing, and rollout across the Australian and New Zealand broker network
  • Metanoia to production: WikiBrokerAI’s capability embedded into Metanoia’s mortgage application workflow and interface

“Artificial intelligence gives our customers the opportunity to rethink key business functions and reinvent customer experiences. For Finsure, AI is a prime opportunity to let brokers and advisers refocus the time they'd have spent on manual loan processing into growing and nurturing customer relationships. Brokers and advisers value what Finsure brings them, and we're delighted to partner with Finsure to build on that.”

Nick DenisonClient Engagement Principal, Mantel

How we built it

The solution is built on AWS:

  • Strands agent deployed to Bedrock AgentCore Runtime
  • Anthropic’s Claude Sonnet large language models accessed through Amazon Bedrock
  • Retrieval-augmented generation (RAG) supported by a Bedrock Knowledge Base on top of Aurora PostgreSQL as the vector store
  • Bedrock Guardrails used to block harmful content and support response grounding
  • Amazon Elastic Kubernetes Service (EKS) hosting microservices for the overall solution, with Aurora PostgreSQL as the backend database

Third-party applications and services:

  • LaunchDarkly for feature access
  • Auth0 for user authentication and access management

“Finsure is committed to leading the market in providing brokers an enriched and differentiated experience to better service their customers. Through investment in the evolution of Finsure we are providing our brokers tools to deliver efficiencies, and to better service their customers. WikiBrokerAI is the first of many releases on our Product and Technology roadmap with several more releases planned in 2026 and early 2027.”

Simon BednarCEO, Finsure

See how we’re helping businesses scale with AI-first solutions