finACTION AI LATAM

The 2021 Global Findex of the World Bank found that Latin American women are 12% less likely than men to use formal credit. Gender-based usage gaps in the region obstruct women’s economic security and personal empowerment. Specifically in Chile, researchers suggest that the prevalence of gender bias exhibited by executives who make credit assessment decisions reduces the probability of approval of a loan requested by women by 18.3%. While the ongoing growth of fintech companies and the use of technological solutions like Automated Decision Making systems for credit assignation in the region has the potential to address prevailing financial inclusion barriers, it is necessary to study ADM systems with gender perspective from the Global South, in order to prevent biases presented by individuals to be embedded in the data.

This project aims to improve the assignment of credit scores with a use case to reduce gender bias in ADM systems, including technical and social interventions. We propose the development and implementation of a use case based on the analysis of variables involved in ADM systems used by Latin American fintech startups and companies, which are authorized by the competent authorities and are committed to gender equality, or have experience with or proven sustainability indicators (ESG). 

In order to provide a complete socio-technical approach, the project includes the co-construction, co-design and socialization of a financial education curriculum and respective materials, based on UNDP Mexico’s financial education model, to be implemented with potential women users of financial products and services of the participating fintech companies. Aiming to set precedence on public policy agendas, the project will include a policy brief.

Our target population are women of 20 - 45 years of age, who live in urban areas of Mexico and Latin America, specifically in capital cities and economic centers like Mexico City, Bogotá, and Santiago; who have high-school, higher education or professional qualifications, and middle-class income levels. Previous implementation of PNUD Mexico’s Gender-focused Financial Education materials has been effective in fostering better financial habits of women participants from such demographic groups. As part of the monitoring and evaluation results, 75% of participants had balanced budgets, 80% were constantly saving, and 90% were creating financial goals.

It is expected that the development and implementation of the technological product that we propose will strengthen women's agency in terms of financial inclusion, and contribute to amplify gender mainstreaming in the use of ADM systems in the fintech industry, especially in the Latin American Region. The financial education element, which has had significant results in previous implementations, is expected to have a positive influence on women’s access to credit, responsible use and debt reduction, and prevention of financial vulnerability in the long term. Finally, the policy brief and socialization activities can be used as input and spaces to guide the AI policy and digital transformation agendas in the region.

PIT Policy Lab's network of consultants enables the convergence of Public Interest Technology with different fields of action, professional experiences and academic research. The proposed work takes into account the authors' expertise in international relations, systems engineering, public policy, technology policy, gender studies, and data science. 

Each team member contributes an area of expertise and skill set that intersects with the main objectives of this proposal.  The team in charge of the development of this initiative includes Lucía Tróchez Ardila, Luz Elena González and Alejandra Sánchez, General Manager, and Project Leaders at PIT Policy Lab, respectively; Mariana Villasuso, member of PIT Policy Lab's network of consultants and Financial Inclusion Coordinator at the United Nations Development Program (UNDP) in Mexico; and Edgar Valdés, Data Science specialist and member of PIT Policy Lab's network of consultants.

The most significant partnerships for our project would be the Fintech Association Mexico, which collaborates with 32 members; and FinTech companies such as: Stori (with 1.2 million users in Mexico), Nu Mexico (with 3.2 million users in Mexico.  and Nu Bank (with 70 million users in LATAM), which could benefit from this project by exploring and addressing inefficiencies in the use of ADM systems to assign credit for women clients, and which could become recipients of high-quality curriculum and materials to implement Mexico’s National Policy for Financial Inclusion, which compels financial entities to provide financial education to their users. The team also aims to connect with traditional banking institutions, and non-traditional institutions that provide financial services in the Latin America region, such as MercadoPago and FEMSA Group. Additionally, the Policy Lab is an active member of the Feminist AI Research network, which leads the capacity-building agenda on equitable and feminist AI, and which is currently working around ADM systems, is an academic allied organization for this project.

Our solution has the potential to serve women fintech users from the LAC region, which according to the Inter American Development Bank compose 46% of the final users on average. However, our pilot would start with a goal of impacting 800 LAC women, 200 women per country of implementation.

To contextualize the challenges associated with digital transformation from a public interest and social impact lens in Latin America and the Caribbean. We develop solutions (tailor-made projects and programs) to complex social challenges through a multi-sector and multidisciplinary collaboration model.

Proposals for public policy and regulatory interventions have been developed mainly from the Global North. Our proposed project mainstreams gender perspective in technical and social interventions, in addition to assessing gender bias in ADM systems. In this sense, the financial education element is reinforced by a people centered approach, based on the “Nudge” theory by Sunstein, Kahneman and Thaler. These elements allow participants to change behaviors instead of only grasping theoretically the concepts and impact on their daily lives of both financial education and artificial intelligence. 

Year 1: 

Months 1–6: Planning and alignment with allied companies, NDA signatures, retrieval of datasets and training on data lifecycle management within the companies.

Months 7-12: Initial focus groups with stakeholders. Analysis of results and survey information to define priorities of interventions.  Data exploration and establishment of explainability techniques. Development of curriculum for social intervention. Communication of social intervention with potential participants.

Year 2:

Month 1-6: Technical development of ADM analysis and variable identification, communication of initial findings with allied companies. Development of policy report. Start of implementation of curriculum. 

Months 7-12: Implementation of AI models for gender-based equitable credit assignment. Continued implementation of curriculum. Development of policy report.

Year 3: 

• Planning of sustainability strategy. Exploration of partnerships to scale the intervention. Policy recommendations and report conclusion. Final result metrics survey.

Year 4:

• Planning and implementation of multi-sector roundtables and focus groups. Documentation of interventions’ final results. Report design, publication and dissemination. First stage conclusion. Brand build-up. Establishment of partnerships to scale the intervention.

Year 5:

• Scale-up of the solution: Development of timeline of a second stage. Implementation of a second stage of the project. Brand build-up. Gathering of additional information and promotion towards a second set of women in the crowd-working space. 

The project aligns with SDGs 1.4, 5.5, 8.10, 9.3. Two metrics will be used to measure progress in the social intervention: (1) Access, in terms of use of financial services, saving behavior and financial planning; and (2) Financial education, in terms of understanding of concepts such as interest rates, capital, commissions, default rates, among others. The technical intervention’s metrics will be the obtaining of better credit opportunities for the women participants within the allied fintech companies.

The initiative has the potential to direct the use of AI in the Latin American fintech ecosystem towards equitable design and incorporation of gender perspective. It is expected that the development and implementation of the technological product that we propose will strengthen women's agency in terms of financial inclusion, and contribute to amplify gender mainstreaming in the use of ADM systems in the fintech industry, especially in the Latin American Region. 

The financial education element, which has had significant results in previous implementations, is expected to have a positive influence on women’s access to credit, responsible use and debt reduction, and prevention of financial vulnerability in the long term. 

Finally, the policy brief and socialization activities can be used as input and spaces to guide the AI policy and digital transformation agendas in the region.

The findings and learnings from the use case and the policy brief have the potential to inform the design of fair and ethical AI systems in fintechs in the Latin American region. 

The data consumed would be credit scores and other variables of interest in accordance with availability from the fintech companies. As a product, gender-informed credit scores will be produced.

It is also possible to generate a recommendation system with explainability features for credit assignment uses and client support. We will follow the Responsible Artificial Intelligence patterns that start with a heuristic understanding of the operations. The main deliverable of this process is a matrix of explainability through the indicators that are used, where the operation must give the reasons for the use of each one of them. 

At this point we start by testing simple models that are highly understandable, starting with linear regression, decision trees, random trees, and then combinations of weak models with XGbost for the predictability side of the indicators. The second point of explainability is in clustering algorithms (mostly unsupervised) starting with vector machines, k-means clustering. At this point, if the problem requires it, artificial neural networks are generated to search on the associated indicators. When deciding which types of algorithms to use, recommendation systems can be built with collaborative filters or context filters.

The success metrics for our AI system will be Heuristic Metrics: mean or median of the labels, distribution of the labels in the training and test data, indicators of the most probable labels, validation of predictions by experts; Precision and Recall: Used in binary classification problems to evaluate the model's ability to correctly identify positive and negative instances; F1 Score: Combination of precision and recall that provides a single value to measure the model's overall performance; Mean Squared Error (MSE): Used in regression problems and measures the average squared difference between the predicted and actual values; Root Mean Squared Error (RMSE): This is the square root of the MSE and is commonly used in regression problems. Receiver Operating Characteristic (ROC) Curve: Graphical plot that shows the trade-off between the true positive rate and false positive rate for different threshold values. It is commonly used in binary classification problems. 

The project includes the implementation of two to four design thinking focus groups that would incorporate diverse perspectives in the initial and final stages of the project. The first set of focus groups will be centered on the definition of the fintech user’s priorities, needs and contexts, which will influence the design of the curriculum and materials. The second set of focus groups would aim to gain information on the effectiveness, appropriation, opportunities and best practices of our intervention. 

We propose to implement an open to the public round table event for the use case, centered on mainstreaming gender perspectives in fintech companies with a socio-technical approach. This event would be held at open collaboration spaces such as Fintech México; through associations, like the ASOFOM (Mexican Association of Multiple Purpose Financial Companies); and independent commissions such as the CNVB (National Banking and Securities Commission).

Additionally, we would create a landing page within our website (www.policylab.tech) where the general public could review the information about the project, download the policy brief and have access to resources and materials regarding the use case.