Work · Scienaptic AI

iCUE: putting LLMs into production for US lenders, starting where it was safe

iCUE reached 80% client adoption within three months of launch, with 15% converting to paid. Paid conversion reached 30% within nine months of launch.

Context

Scienaptic sold AI credit decisioning to more than 100 lenders, mostly US credit unions with small credit teams. Understanding the portfolio meant static reports and manual digging, and early warning, growth and portfolio analysis sat in separate tools or spreadsheets.

The hard call

Start at monitoring, not at the application. US lenders must give a specific, explainable reason for every declined application. An LLM in that decision raised explainability questions we couldn’t yet answer, so iCUE’s LLM features worked on loans already booked, where the value was high and the regulatory risk low.

One bundle, not three tools. I packaged the early warning system, the growth product and the LLM features into one module. Separate products would have been simpler to price. One module was easier for a small credit team to adopt and gave every client a reason to upgrade.

What I did

• Defined iCUE as one module on top of decisioning: early warnings, growth and LLM-powered portfolio intelligence.
• Grounded the LLM features in each lender’s own loan, bureau and performance data.
• Shipped plain-English portfolio questions, emerging-risk flags and auto-written monitoring reports.
• Took it to market as a paid add-on for the existing base.

My role

As Director of Product Management, I conceived and launched iCUE with engineering, data science, implementation and sales, reporting to the COO.

Results

What I take from it

Put a new technology where its mistakes are cheap and its value is obvious, then earn the right to move closer to the decision.

Published 19 Oct 2026

Next case: Adding small business lending to a consumer credit platform