AWS · Embedded Finance

Lending engine scaled on AWS to fund SMBs across 3 continents.

We rebuilt the lending decisioning engine as an event-driven service mesh on AWS, unlocking 6× loan throughput and cutting unit cost per decision by 33%.

[Loans / day]
[Unit cost]
-33%
[Continents live]
3
[ The Problem ]

Where the business was stuck.

Underwriting engine couldn't scale to meet SMB lending demand across multiple continents.

[ Key Challenges ]
  • Batch underwriting pipeline capped at ~2K loans/day
  • Regional latency slowed decisions for LATAM and EMEA SMBs
  • Model retraining took weeks
Our approach

How we engineered
the outcome.

STEP 01

Event-driven decisioning on EKS with Kafka Streams

STEP 02

Multi-region deployment with edge-local decisioning

STEP 03

MLOps pipeline for weekly model refresh with shadow eval

[ Solution highlights ]

Delivered — measured — in production.

  • Loan throughput grew from 2K to 12K decisions/day
  • Regional decision latency dropped to <300ms P95
  • Unit cost per decision cut 33%
[ Tech stack ]

Built on AWS.

AWS EKS
Kafka Streams
SageMaker
Terraform
"
We went from turning SMBs away to funding them in real time — same day, three continents.
Chief Product Officer, Embedded Finance

Your case study starts here.

Book a discovery →