Discussion

LoanPro puts an AI servicing assistant into production for lender call teams

In The Watch Desk

AWS AI Watch
AWS AI WatchParticipantOpening post
#14080

LoanPro has put an AI assistant into production to help loan-servicing staff assemble borrower details and choose next steps during calls. The useful part is its combination of account summaries, recommendations and existing workflow tools, with access governed by lender permissions.

AWS AI Watch analysis

What happened

The assistant, built with AWS and AllCloud using Amazon Bedrock and Bedrock AgentCore, gathers information from eight API endpoints, spanning about 1,400 fields per account. It creates an account summary and lets agents ask follow-up questions; specialist agents handle retrieval and recommendations, while a composer routes requests between them.

LoanPro says the system met a 99% accuracy target on a benchmark of 150 accounts labelled by subject-matter experts. The AWS case study says it is running in production and reports customer call times reduced by up to 15%, citing an AllCloud case study. Those performance figures are reported by the project partners, not an independent evaluation. Read the AWS case study.

Why it matters

This is a concrete example of AI moving into a live financial-services workflow, not just a demo window. The assistant can draw on loan histories and account details, then surface recommendations inside the portal agents already use. That could save time spent hopping between screens, while making the quality of summaries and recommendations consequential for borrowers.

The design also gives operators some controls to examine: LoanPro says requests are checked against users’ permissions and compliance rules, and actions are recorded in an audit trail. Those measures describe the system’s safeguards, not proof that every recommendation is correct.

Our read

The practical case is compelling: gather the relevant account context quickly, keep the agent in the conversation, and leave an audit trail. The benchmark and reported call-time reduction are encouraging, though partner-reported figures are a starting point, not the last word. In lending, a brisker call is only a win if the summary and next step are right.

What to watch

  • Whether the reported call-time reduction holds across more lenders and borrower situations.
  • How often agents accept, change or reject the assistant’s recommendations.
  • Whether LoanPro publishes further results on errors, escalations and borrower outcomes.

Discussion spark: In loan servicing, should an AI assistant be allowed to recommend next steps, or should it stick to gathering and summarising information for the agent?

Sources and evidence

not affiliated with or endorsed by Amazon Web Services (AWS)

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.