Vantle

AI-Powered Risk Assessment Platform

Vantle had built a profitable lending business on manual underwriting. Their credit analysts were experienced and thorough — but they were also the bottleneck.

Every loan application sat in queue for 24-48 hours while analysts pulled data from six different sources, cross-referenced regulatory requirements, and made judgment calls that were as much art as science.

The board wanted to 10x loan volume. The analysts couldn't 10x themselves. And the three previous attempts to "add AI" had failed because they treated the problem as a technology challenge rather than a strategy one.

I started by shadowing the credit analysts for two weeks. Not to understand their process — I needed to understand their judgment. What signals did they weight? Where did they override the data? When did they trust their gut?

The insight was counterintuitive: the analysts weren't doing one job, they were doing three. Sixty percent of applications were straightforward approvals or rejections where the data spoke clearly. Thirty percent required one specific piece of additional context. Only ten percent genuinely needed expert judgment.

Instead of building one monolithic AI system to replace analysts (the approach that had failed three times), I designed a three-tier architecture:

Tier 1: Automated Decision Engine — For the 60% of clear-cut cases, an ML pipeline that aggregated data from all six sources, applied regulatory rules, and made instant decisions. No human in the loop.

Tier 2: Augmented Analysis — For the 30% that needed context, the system pre-assembled all relevant data, flagged the specific question that needed answering, and presented it to an analyst who could resolve it in 2-3 minutes instead of 2-3 hours.

Tier 3: Expert Review — For the 10% that genuinely required judgment, the system provided a comprehensive risk profile with confidence intervals, letting analysts focus their expertise where it mattered most.

The platform was built on a event-driven architecture with three core services:

The key architectural decision was making the system self-improving. Every Tier 3 expert decision became training data. Every Tier 2 resolution taught the system what "additional context" actually meant. Over time, the boundaries between tiers shifted automatically.

The real outcome wasn't the numbers — it was what the analysts did with their freed-up time. They shifted from processing applications to designing new lending products. The team that had been a cost center became an innovation engine.

OUTCOMES

CAPABILITIES APPLIED

Let's talk about what you're building.