Revolutionizing Credit Assessment: Can Customer Reviews Predict Business Lifespan? A Study of Insolvent Australian Food & Beverage Services

Despite accounting for two-thirds of the country’s employment, Micro, Small and Medium Enterprises (MSME), have endured the obstacle to access external funding due to the lack of financial reporting and reliable evidence proving their repayment ability. Alternative data, therefore, has been widely studied and adopted into lending decision models to address the asymmetric information gap. This initiative has been encouraged by G20 Global Partnership for Financial Inclusion (GPFI). However, there were few studies in Australia concerning the alternative data in the business solvency context.This study examines whether the customer review data from Google Map reviews can predict business lifespan. It also aims to advance the understanding of the reviews on the prediction. The research scope focuses on the insolvent Australian Food and Beverage Service businesses, leveraging explainable machine learning models and Aspect-based Sentiment Analysis, extracted by the Gemini 2.5 Pro model.The research finds that review features and macroeconomic features (average %change of State Consumer Price Index (CPI)) have predictive effects on both business lifespan prediction and classification models. The predictive features include reviews’ informativeness, reviewers’ credibility, changes of review features over the lifespan, and owners’ responsiveness. Directional relationships are also discovered. The Total Number of Reviews and the Number of Pictures posted on reviews have significant positive associations with business lifespan.Additionally, Value for Money Sentiment is the top qualitative textual feature considered by the prediction models. However, it is found in this study that including qualitative textual features (% positive sentiment words, factual words, aspect weight) does not significantly improve prediction accuracy.As the model highly considers reviewers’ credibility and the informativeness of reviews, it suggests that extracting these features could address concerns about fraudulent reviews. The highly predictive features found in this study provide actionable insights into business owners and lenders, guiding their focus toward factors contributing to business viability.