CashAI v5.5 Drives Dave's ExtraCash Growth

Dave Inc. deployed its proprietary cash-flow underwriting engine, `CashAI v5.5`, to expand credit access via the ExtraCash product while tightening credit performance. The release correlates with a 50% year-over-year lift in ExtraCash originations in Q4 2025, a 12% sequential improvement in the 28-day past due rate, and a 62% top-line increase that quarter. Adjusted net income rose 92% year over year, and management guided $690-$710 million revenue and $290-$305 million adjusted EBITDA for 2026.

CashAI v5.5 is described as a real-time, automated cash-flow underwriting engine that analyzes members' primary bank account activity to assess affordability and default risk. Key practical characteristics practitioners should note are:

• •Real-time bank-transaction ingestion and feature extraction from primary account feeds, supporting liquidity and volatility signals
• •Automated behavioural and cash-flow scoring that augments or replaces traditional thin-file credit signals
• •Decisioning thresholds tuned to balance origination growth with short-term delinquency control

The public disclosures do not enumerate model architecture, training data volumes, or validation suites. However, the outcomes imply better predictive separation between approve/decline cohorts and improved calibration for short-window delinquencies. For ML teams, the likely components include engineered temporal features, stabilized risk scoring, population-level cohort monitoring, and rapid online retraining or parameter updates to keep pace with consumer behavior shifts.

This is a clear commercial example of ML-driven underwriting materially improving both growth and unit economics. Many fintechs experiment with cash-flow models; Dave's outcome is notable because it drove simultaneous increases in originations and improvements in short-form delinquency metrics, which typically trade off. For practitioners, the case highlights the operational necessities of production risk ML:

• •robust real-time data pipelines and fraud/poisoning defenses
• •drift detection and governance to avoid calibration degradation as volumes expand
• •tight integration between credit ops and ML model control to tune loss-taking vs. growth

Compared with legacy credit models that rely on bureau scores, cash-flow-first underwriting better captures liquidity volatility for near-term installments. Dave's guidance implies the model scaled beyond pilots into a strategic lever for market share and profitability.

Monitor disclosures around model governance, feature drift monitoring, and default cohort behavior over 12 months as originations scale. Watch for regulatory scrutiny or explainability challenges if CashAI v5.5 becomes central to underwriting decisions and member outcomes.