Over the past few months we’ve been experimenting with a small AI pipeline that scores real-world SMBs (automotive, cafes, services, logistics) within 24–48 hours — using a mix of financial documents, operational metrics, and qualitative patterns.
The surprising part for us was how consistent the predictions became even on messy SMB data, where traditional credit models usually fail. We also discovered that for small businesses, “risk factors” look very different from classic corporate lending: seasonality, owner behavior, local demand spikes, etc.
We’re still learning, and I’m curious if anyone here has tried building something similar or has thoughts on the general feasibility/limitations.
Not trying to promote anything — happy to discuss the technical side, training issues, data quality challenges, modeling approaches, etc.
If useful, here’s the project we’re building around it:
https://rivellium.com
0 comments
—
We built a AI system that scores RW SMBs in 24–48h (curious about feedback) · HackerTrans
The surprising part for us was how consistent the predictions became even on messy SMB data, where traditional credit models usually fail. We also discovered that for small businesses, “risk factors” look very different from classic corporate lending: seasonality, owner behavior, local demand spikes, etc.
We’re still learning, and I’m curious if anyone here has tried building something similar or has thoughts on the general feasibility/limitations.
Not trying to promote anything — happy to discuss the technical side, training issues, data quality challenges, modeling approaches, etc.
If useful, here’s the project we’re building around it: https://rivellium.com