Using AI to Bring Social Capital Into Credit Decisions

Accurately assessing borrowers’ creditworthiness is the perennial challenge for microfinance institutions (MFIs). Clients who live on low incomes often lack credit history and other data that financial institutions rely on for due diligence. As a result, those who are truly creditworthy may be rejected, and those who represent a greater risk may receive loans they shouldn’t, leading to defaults and over-indebtedness.

However, there is a growing realization that there are other types of data that can help inform credit evaluations for “thin file” customers. It has long been recognized that social capital – the benefits and resources derived from social networks and relationships - can have a positive influence on borrowing behavior. Solidarity group lending was born from this understanding. But how can we quantify social capital as part of the due diligence process for individual clients?  

Alternative credit scoring powered by social capital data

The emergence of big data has enabled financial institutions to digitize and refine credit scoring processes with alternative data. Until recently, alternative credit scoring models have focused on digital finance transactions and purchases on mobile phones, such as airtime top-up, data and SMS bundles, as their primary data sources. 

However, as the cost of smartphones falls and GSM/internet coverage increases, we must expect richer data that can capture information about borrowers’ social capital. For MFIs partnering with mobile network operators, data can now be drawn from an ever-expanding array of sources pertaining to social capital, such as:

• Call logs or records (calling or receiving calls on a regular basis, to/from a diverse set of people).
• Social media presence and sign-in frequency (using services of a social network provider such as Facebook or Twitter).
• Memberships held in a digital format (being registered in online communities, knowledge exchange platforms or peer-to-peer networks leveraging internet or SMS-based technology). 

Upon informed consent, AI models could use this mobile data to assess to what extent a borrower actively participates in community life.

MFIs could also take advantage of their agent networks to enrich datasets. Since agents perform transactions in a local environment where they leverage social relationships to grow business, they are in a position to collect community-based information on borrower’s social capital. For example, they could share information on whether a client is a member of an association acknowledged by the community, is the head of their district or a village elder or carries some traditional authority.

Benefits for MFIs and for clients

By incorporating data on social capital in their credit evaluation processes, MFIs can experience multiple benefits, including:

• Increased revenues from credit provision, by reaching new borrowers with high social capital who had previously been excluded due to lack of data or insufficient conventional collateral.
• Reduced credit risk, as the social capital data provides greater assurance that the borrower has access to business opportunities within their community and will be able to use the loan as intended.
• Improved agent network performance. Insights from alternative credit scoring can help MFIs identify and incentivize agents who collect quality social capital data and have a positive impact on loan repayments. 

In turn, clients can benefit by gaining greater access to credit for those who would otherwise have been rejected because of lack of traditional data or collateral. Moreover, if a customized AI model assesses an individual's creditworthiness unfavorably due to social capital considerations, it could still help to advance financial inclusion by identifying multiple loan applicants who lack social capital and bringing them together for group lending.

Proceed with caution

With all these potential benefits, it can be easy to get caught up in enthusiasm for using alternative data on social capital. However, there are also important risks to consider, which require proceeding with caution. If MFIs wish to leverage low-income clients’ social data to assess creditworthiness, they must design a global strategy to overcome the following key challenges.

Algorithmic bias and gender considerations

Algorithms need to be designed to take into account the diversity of people’s social experiences and ensure they are not biased towards certain types of social capital over others. For example, people who don’t have smartphones or even feature phones can of course still actively partake in community life. For these potential clients, only information collected by agents should be considered in datasets and weighted accordingly into algorithms.

Furthermore, men’s social capital in most countries tends to be more expansive, diverse and influential, giving them better access to economic resources, political power and public services. On the other hand, women’s social capital is often more restricted to private and community-oriented networks that provide social support in times of need. Algorithms should include gender-intelligent design to help balance these disparities, giving more weight to data input related to family-oriented or close-knit community networks when performing credit scoring for women.

Customer protection and privacy concerns

Customers’ informed consent is the critical foundation for gathering data on people’s social lives and analyzing it with AI-based models. Personal data must be used fairly and responsibly, and should be gathered solely for the purpose of evaluating a borrower’s social capital for credit applications. 

Technology risk

Technology risk is the most common factor delaying AI-driven project implementation. Limited budgets and regulatory constraints are the primary problems, leading to poor uptake on inflexible and costly platforms. To properly manage IT risk, MFIs must develop a consolidated work plan that combines technology, organizational, training and marketing work streams to expedite a smooth and efficient implementation process.

Corporate culture change

Organizational culture change is essential to the success of AI-based deployments, especially when it comes to processing multifaceted data on borrowers’ social lives. A culture of innovation allows staff to leverage AI potential for credit scoring without fear of failure. Lean decision-making processes and multi-functional teams, which support rapid adjustments of the algorithmic model, will enable an environment where the MFI can refine credit scoring and increase its reliability.

The future of credit evaluations

The integration of social capital data into credit scoring models has the potential to enhance the accuracy and inclusivity of MFIs’ lending practices . However, this innovation must be approached with caution. If MFIs can learn to manage the risks successfully, both MFIs and their clients will reap the benefits as access to credit and the associated opportunities expand to ever more people.