Beyond the Bureau: How Prime is leveraging $3B in historical performance to build the next generation of underwriting alpha.

Investors often ask: “How is your credit model actually different?”

In SMB lending, many fintechs build their risk models on limited, point-in-time snapshots - approaches that often break down during periods of volatility. At Prime, we’ve moved beyond snapshots. We’ve built an ensemble of machine learning models that integrate five independent risk dimensions, enabling a holistic view of each customer and eliminating many of the blind spots common in SMB underwriting.

01. The Foundation: Longitudinal Intelligence

We used eight exclusive Equifax archive datasets totaling 150K+ records, purposefully spanning four distinct economic snapshots (2018, 2019, 2022, 2023).

• The COVID Filter: We deliberately excluded 2020–2021 originations to remove the "noise" of government aid. As our Chief Risk Officer put it: "Model training is about capturing true behavior. If your data includes a period where every balance sheet was artificially buoyed by historic stimulus, you aren't measuring creditworthiness - you're measuring a subsidy."

• Cycle-Resilience: Rigorously validated on two "out-of-time" samples, our model is built to perform when the market shifts.

02. The Architecture: The Ensemble of Five

Traditional commercial bureau scores are only available for ~50% of SMBs. For many lenders, the other 50% is a "decline." Prime sees what they miss. Our modular structure leverages five independent models to generate a unified risk score:

1. Commercial Bureau Model: Raw and summarized commercial delinquency attributes.

2. Consumer Bureau Model: Linked personal credit of principals - enabling a holistic view of both the business and its owners.

3. Commerce Intelligence Model: We ingest 100+ specialized attributes - including 12-month revenue stability, merchant transaction growth, and card-revenue signals - to see real-world health that bank statements alone often miss.

4. Cash Flow Model: Real-time Plaid telemetry for current liquidity and debt-service capacity.

5. The Partner Model (The Moat): We ingest first-party data directly from our SaaS partners (e.g., time on platform, transaction patterns). This "embedded" signal is a leading indicator of risk that no external bureau can access.

03. Performance: 21% Lift at a 5% Default Threshold

This ensemble approach doesn't just provide "more data" - it provides superior risk separation.

• Expanded Market: We approve 21% more applicants at the same ~5% risk threshold.

• Unmatched Precision: In the lowest-risk deciles, our model’s predictive power is 2X more accurate than standard bureau scores. In the lowest-risk deciles, our model is twice as predictive as traditional bureau scores.

• Blind Spot Elimination: Even when a commercial bureau score is missing, Prime uses the other four models to underwrite the applicant safely.

The Strategic Takeaway

Prime’s Credit IP represents a multi-year head start in data science and risk modeling. By combining longitudinal archives with an ensemble that includes proprietary partner signals (varies by partner), we’ve created a "360-degree" view of SMB risk that is nearly impossible to replicate.

For our partners, this means higher conversion and lower losses. For our investors, it means a credit engine built for sustained, scalable alpha.

Want to learn more? Click here to schedule a demo.