Observational Policy Ranking Guides SMB Financial Decisions.

Small and medium-sized businesses (SMBs) often require timely financial guidance, but historical accounting logs typically record self-selected and co-occurring business changes rather than randomized recommendations. This presents a challenge for deriving actionable insights. Researchers framed this problem as observational policy ranking, where a policy selects one of 34 ledger-derived business-change categories to optimize a target financial Key Performance Indicator (KPI) based on pre-decision financial information. Using 85,078 company-month observations from 7,505 firms, the study introduces Covariate-Adjusted Residual Policy Learning (CAR-PL). This action-wise R-learner operates directly on multi-hot logs and incorporates regularization based on observational support. CAR-PL was compared against several baselines, including an uplift T-Learner, a conservative contextual value model, a zero-shot LLM, and non-personalized references. Results show CAR-PL achieved the highest Gross Profit point estimate (0.084), while the T-Learner led for Revenue (0.085), and the contextual value model for Quick Ratio (0.062). CAR-PL and the T-Learner were not statistically separable on growth KPIs. Notably, CAR-PL selected a broader range of categories (33-34) and produced less concentrated selections, suggesting more diverse guidance. These findings support objective-specific ranking of SMB financial guidance from complex, multi-action accounting logs.