Overview of biases
| Categories of biases | Sub-bias type | Frequency | Exemplary articles | Core findings |
|---|---|---|---|---|
| Algorithm to user | Algorithmic bias | 11 | Schultz and Fabozzi (2022) | AI and machine learning in banking and financial institutions can perpetuate bias, particularly in lending and credit decisions |
| Evaluation bias | 8 | Amaral et al. (2023) | AI and machine learning can introduce biases in finance, but when evaluated through risk-based pricing, they can improve profits and provide traditionally marginalized customers with access to loans | |
| User interaction bias | 3 | Liu et al. (2024) | AI technology in the banking industry may reduce the positive impacts of Chinese clients' perceptions of Guanxi on their engagement, reflecting attitudinal differences between bankers and clients | |
| Emergent bias | 3 | Tigges et al. (2024) | Adoption of AI and machine learning in banking and financial institutions has the potential to enhance credit accessibility but also raises concerns about bias and discrimination | |
| Data to algorithm | Sampling bias | 6 | Tserng et al. (2011) | An enforced support vector machine-based model (ESVM) outperforms the traditional logistic regression model in predicting construction contractor default risk, by correcting sampling bias from class imbalance |
| Representation bias | 6 | Repetto (2025) | Multicriteria interpretability-driven deep learning technique has been developed to create robust AI models for credit risk prediction, overcoming biases resulting from data scarcity in financial institutions | |
| Measurement bias | 8 | Shi et al. (2023) | Multi-step convex optimization approach based on MSW-LASSO improves parameter estimation in high-dimensional index tracking, reducing measurement bias in AI-driven financial modeling | |
| Linking bias | 1 | Packin and Lev-Aretz (2024) | AI and machine learning in banking and financial institutions can perpetuate bias through faulty inputs, proxy discrimination and surveillance capitalism | |
| Omitted variable bias | 1 | Bockel-Rickermann et al. (2025) | Advanced machine learning methods can improve the estimation of individual bid responses in banking, especially when addressing confounding bias in observational data | |
| User to data | Aggregation bias | 1 | Zhang et al. (2022) | Meta-algorithms can reduce the risk of choosing wrong experts in financial environments, especially when base expert algorithms are sensitive to certain environments or parameters, mitigating the bias in expert selection |
| Social bias | 3 | Tubadji et al. (2021) | Consumers' adoption of robo-advisory AI in banking is influenced by cultural norms, social capital and trust | |
| Behavioral bias | 14 | Athota et al. (2023) | AI and machine learning can help reduce cognitive biases in financial planning, but there is uncertainty about whether financial planners will fully rely on AI-based decision making | |
| Content production bias | 1 | Gurdgiev and O'Loughlin (2020) | AI and machine learning can be used to detect and mitigate biases in investor sentiment data, which is crucial for financial institutions to make informed decisions in cryptocurrency markets | |
| Historical bias | 2 | Bono et al. (2021) | Machine-learning models in credit scoring are more accurate overall and do as well as traditional models on relevant fairness criteria, but they may perpetuate or amplify human biases from the past | |
| Population bias | 1 | Li and Goel (2025) | AI and machine learning systems used in banking and financial institutions can exhibit biases that may lead to discriminatory outcomes |
| Categories of biases | Sub-bias type | Frequency | Exemplary articles | Core findings |
|---|---|---|---|---|
| Algorithm to user | Algorithmic bias | 11 | AI and machine learning in banking and financial institutions can perpetuate bias, particularly in lending and credit decisions | |
| Evaluation bias | 8 | AI and machine learning can introduce biases in finance, but when evaluated through risk-based pricing, they can improve profits and provide traditionally marginalized customers with access to loans | ||
| User interaction bias | 3 | AI technology in the banking industry may reduce the positive impacts of Chinese clients' perceptions of Guanxi on their engagement, reflecting attitudinal differences between bankers and clients | ||
| Emergent bias | 3 | Adoption of AI and machine learning in banking and financial institutions has the potential to enhance credit accessibility but also raises concerns about bias and discrimination | ||
| Data to algorithm | Sampling bias | 6 | An enforced support vector machine-based model (ESVM) outperforms the traditional logistic regression model in predicting construction contractor default risk, by correcting sampling bias from class imbalance | |
| Representation bias | 6 | Multicriteria interpretability-driven deep learning technique has been developed to create robust AI models for credit risk prediction, overcoming biases resulting from data scarcity in financial institutions | ||
| Measurement bias | 8 | Multi-step convex optimization approach based on MSW-LASSO improves parameter estimation in high-dimensional index tracking, reducing measurement bias in AI-driven financial modeling | ||
| Linking bias | 1 | AI and machine learning in banking and financial institutions can perpetuate bias through faulty inputs, proxy discrimination and surveillance capitalism | ||
| Omitted variable bias | 1 | Advanced machine learning methods can improve the estimation of individual bid responses in banking, especially when addressing confounding bias in observational data | ||
| User to data | Aggregation bias | 1 | Meta-algorithms can reduce the risk of choosing wrong experts in financial environments, especially when base expert algorithms are sensitive to certain environments or parameters, mitigating the bias in expert selection | |
| Social bias | 3 | Consumers' adoption of robo-advisory AI in banking is influenced by cultural norms, social capital and trust | ||
| Behavioral bias | 14 | AI and machine learning can help reduce cognitive biases in financial planning, but there is uncertainty about whether financial planners will fully rely on AI-based decision making | ||
| Content production bias | 1 | AI and machine learning can be used to detect and mitigate biases in investor sentiment data, which is crucial for financial institutions to make informed decisions in cryptocurrency markets | ||
| Historical bias | 2 | Machine-learning models in credit scoring are more accurate overall and do as well as traditional models on relevant fairness criteria, but they may perpetuate or amplify human biases from the past | ||
| Population bias | 1 | AI and machine learning systems used in banking and financial institutions can exhibit biases that may lead to discriminatory outcomes |
Note(s): The reported frequencies are based on 65 articles. Total frequency exceeds the total number of articles since certain articles address more than one bias
Sharing content requires targeting cookies to be enabled. Please update your cookie preferences to use this feature.