G.App (The Global App Monitor)

The solution team consists of researchers who began the solution as an academic project while at the University of Zurich. We envision continuing work on the solution at this stage while still being embedded at the university as a host organization. However, we would consider starting a separate organization if the project moves from a prototype to pilot or growth stage.

Over the past decade, there has been a sharp increase in predatory and fraudulent practices in digital finance and financial technology, including via mobile applications. This has been particularly problematic in many lower- and middle-income countries (LMICs) given the increased ubiquity of mobile devices and internet connection, low financial and digital literacy in some population groups, and a proliferation in the methods “scam” finance app providers are using to exploit vulnerable households and businesses. In addition to the direct harm caused to consumers, this can lead to mistrust of digital finance, which can delay financial inclusion efforts and undermine the benefits of financial technologies.

While the major app platforms do have in-house screening methods and some financial regulators have become more proactive to curb “scam” finance apps, we continue to observe such apps proliferate. As of mid-2021, there were roughly 135,000 “published” finance category apps available on Google’s app platform with over 6.9 billion combined historical downloads. The “published” apps are evolving constantly and exhibit a high turnover rate, with tens of thousands of new finance app packages being added and removed each month. While a share of this turnover is perfectly legitimate, another portion has been linked to a range of problematic behaviors including pure fraud (e.g., no delivery of any actual services), predatory pricing, harassment and abusive debt collection practices, and misuse of sensitive client data, among others. The high volume and turnover has impeded existing consumer protection efforts (e.g., some scam apps have been removed after being flagged by enough victimized consumers, but can be quickly reproduced and replaced by the scam providers). This suggests that current vetting and monitoring efforts are still largely insufficient. 

The app platforms current vetting methods are primarily built around preventing malware, ransomware, or phishing applications that can skim users' data. In general, the existing methods are generically applied towards all apps regardless of category and focus on identifying malicious code. However, many “scam” finance apps do not utilize code-based techniques and indulge in predatory behavior or fraud in other ways. As such, they can evade the app store’s vetting methods and instead get taken down only after many users have been victimized. Other stakeholders have attempted social media monitoring to identify problematic apps or providers. While useful initiatives, they are also reliant on users to first become exposed, victimized, and raise complaints.

Our proposed solution is a market monitoring center to improve transparency on the fintech/finance app market. We plan to set up processes for scraping high-frequency and granular app data, conducting varied quantitative and qualitative analyses to systematize key information concerning providers, products/services, and users, and providing summary reports. This would leverage approaches we have developed in our prior work. The envisioned outputs from these analyses could be varied, adjusted to both global and country-level audiences, and have a range of applied, policy, and academic research implications. This would provide several potential options for setting up B2G, B2B, and/or B2C business models to make the solution financially sustainable in the longer run.

Core project activities could be broadly described as falling into the following categories:

1. Data collection: set up methods for scraping key app data (e.g., observable and hidden meta data, review data, etc.) for the universe of finance apps available on major platforms at regular intervals

2. Analysis and reporting: conduct quantitative and textual/natural language processing analyses to produce various summary data, trend reports, and early warning systems. This could include but is not limited to:

   1. General market monitoring: global overview of the finance app market both as a snapshot for the current month and trend reports – e.g., overall market size in terms of apps and users, what are the types of providers/developers active, what are the types of products or services being providing, what is the quality / terms of their services, what types of consumer protection issues exist, etc. We could similarly also analyze and report on the types of apps that are being “unpublished” each month (as this may help inform signals of where suspect activity lies).

   2. Country reports:similar data and reporting could also be produced at the individual country level at regular intervals, which could be tailored towards local regulators/policymakers.  Moreover, if of interest, reports could also include drill-down to the individual apps and providers that are active at a given point in time to let regulators/policymakers better understand the active landscape in their jurisdiction. For relevant metrics (e.g., consumer protection indicators), these reports could potentially benchmark a given country against global averages or relevant peers and also provide trend forecasting. 

   3. “Early warning” reports: if there are sudden spikes in suspect activity that begin to occur in a given country, we could provide advance notice to relevant stakeholders (e.g., country financial regulators), along with supporting data.

   4. Individual suspect app lists: provision of targeted and filtered lists of highly suspect apps (to select end-users -- e.g., regulators-- for additional checks and targeted take downs as required). While it is important to note that further manual verification would be required, this would still greatly improve the speed and efficiency of screening and monitoring the thousands of finance apps being newly-released every month.
      

3. Data provision: serve as a repository to make some of the aggregated data underlying our reports publicly available or available upon request for policy and academic research purposes via a project website or secure data portal.

The target population for our solution consists of individuals at high risk of being victimized by “scam” finance apps. While victims can come from any geographic location or income strata, in practice, we have observed that this problem disproportionately affects poorer individuals in lower and middle income countries, where it has accompanied a large recent spread of digital financial services and internet connectivity (e.g., it has been particularly problematic in South Asia and Southeast Asia). The hardest hit populations tend to have lower levels of financial and digital literacy, and are more likely to seek alternative sources of finance (since they tend to be excluded from formal financial systems).

On one hand, there is a well-document digital divide that persists, where internet and smartphone penetration are predictably lower in such markets. On the other hand, the cost of smartphones has continued to fall dramatically, making them increasingly attainable even for poorer households in LMICs. For example, one can acquire new entry-level smartphones in India for as low as 800 rupees (~10.5 USD) and even cheaper used. The GSMA has documented that of the 4 billion people globally who do not yet use mobile internet, the vast majority – 3.4 billion live in an area already covered by mobile broadband. Given the high coverage and decreasing smartphone prices, they project over 250 million new mobile internet adopters annually, between now and 2025, primarily coming from LMICs.

The expected goal of the project is consequently to improve safety and consumer protection in digital financial services particularly for this target population, during this period of rapid digital transformation. We aim to do so through several means. First, our approach has potential to directly improve screening and monitoring of the fintech app markets. Second, we would aim to systematize data collection and information available on them in  order to improve transparency around these markets. Third, with the information and knowledge gained, we would work for consumer advocacy and try to bring together relevant stakeholders (e.g., policymakers, NGOs, academic researchers, etc.) to improve related consumer protection regulation and policies.

Our core team consists of researchers whose focus areas are at the intersection between fintech, development economics, and data science. While we retain a rigorous academic approach to research, we have also engaged in many collaborations with relevant industry stakeholders, giving us unique perspectives on applying our approaches and methods to real-world scenarios and problems. This network also has provided proximity to many of the target markets our solution aims to assist and allows us to gather feedback on best ways to implement and measure impact of our solution.

We provided early thought-leadership on the growing problem of predatory and fraudulent fintech apps, including identifying several “hot spots” of suspect activity at the very outset of the pandemic and disseminating ideas on strategies for combating it through a variety of mediums. We then put those ideas to practice by securing funding in order to test our approach as a concept. Our preliminary work has received attention and interest from relevant stakeholders to see how well it can scale to a real-time monitoring tool.

Finally, we note that our team consists of several individuals who are either currently residing in or nationals of South Asian countries where the problem has been particularly pronounced.

We completed a "Concept" stage via an academic study that serves as a general “proof of concept” of the approach. Specifically, the concept funding was provided by Innovations for Poverty Action, who remain collaborative in the project. The funding was used to acquire historical high frequency data on mobile applications and to fund time for several research assistants. We developed methods for systematically categorizing a subset of personal loan apps into suspect vs. legitimate classes. We then trained and validated machine learning models on a broader dataset of apps released prior to 2021 and tested their out-of-sample performance in predicting class for apps newly-released in 2021. These initial models were shown to have high ( 80-90%) out-of-sample accuracy in binary class predictions and fair (70-80%) out-of-sample accuracy in multi-class predictions on unseen data (and we have been doing further fine-tuning since to even further increase accuracy). Those that are classified as “suspect” are shown to be much more likely to have been taken down from the Google Play store in subsequent periods. We released a policy brief and webinar to present our preliminary findings. 

Currently, we are working on several aspects of business development that will allow us to translate this prior work into a functioning prototype. This includes setting up in-house capability for scraping the necessary app data, transforming processes for intake of real-time data, and starting development of a user-interface. At the same time, we are in discussions with potential collaborators about testing the prototype of a market monitoring tool in real-world settings to gain useful feedback on best ways to tailor it to end-user needs.

We have several objectives for applying to MIT Solve.

First, we aim to secure sufficient financial runway to build a more sustainable team to cover us during the prototype stage. We would expand the team by either hiring additional staff and/or scaling up FTE of existing part-time staff to better allow them to have stability during this project period.

Second, we would benefit from legal and business model advice in transitioning the project from a prototype to pilot and growth stage. For example, this could include assistance with fine-tuning and getting feedback on product-market fit and deciding on the best legal structure to set up an eventual spin-off organization.

Third, we could leverage its network to find talent as we scale up our solution team.
 
Finally, once the prototype is ready, we would also greatly benefit from support in product/service distribution and public relations to disseminate the solution and attract potential end-users. We have developed rough estimates of revenue models and believe that we could acquire financial sustainability either through B2G or B2C paths. However, it would be important to scale and an expanded network and PR provided via the Global Economic Prosperity Challenge could be invaluable for this.

While there are an increasing number of stakeholders who are trying to curb “scam” finance apps, these existing solutions appear incomplete and we continue to observe such apps proliferate. First, the app stores vetting methods are primarily built around preventing malware, ransomware, or phishing applications that can skim users' data. Their methods are generically applied towards all apps regardless of category and focus on identifying malicious code. However, many “scam” finance apps do not utilize code-based techniques and indulge in predatory behavior or fraud in other ways. As such, they can evade the app store’s vetting methods and instead get taken down only after many users have been victimized. Secondly, regulators in most markets have taken a mostly reactive approach thus far, where they have waited for significant complaints to arise first before raising issue with the big tech platforms for direct takedowns. Finally, and in a similar vein, other stakeholders have attempted social media monitoring to identify problematic apps or providers. In both the latter cases, the approaches are ultimately reliant on users to first become exposed, victimized, and raise complaints.

By contrast, our solution combines several different approaches to increase transparency on the finance app markets and improve screening and
monitoring efforts. First, we would set up in-house methods for collecting relevant app data that would allow us to set up an independent 3rd-party assessment of the various apps available at any given point in time. Second, we would leverage state-of-the-art NLP techniques to understand specific context of apps and the types of consumer protection issues contained within. Third, we would set up independent methods for labeling and developing predictive models for identifying problematic apps. Finally, we could then apply the models to real-time data to flag scam apps. This would improve efficiency in screening the tens of thousands of nearly released finance apps per year and reduce the likelihood or length of time that scam apps remain on the market.

Our theory of change is that the combined spread of consumer-facing and suptech tools (B2C, B2G, and B2B) can help both with immediate reduction of spread of scam finance apps and longer-term systems that help to limit their re-emergence. It envisions that a multi-pronged approach has the highest chances of success where:

1. A B2C channel (consumer-facing app) would directly assist consumers in avoiding scam apps
2. A B2G channel (regulator-facing web tool) would help financial regulators, supervisors, and LEA in monitoring and more efficiently removing problematic apps
3. A B2B channel (general market monitoring center) would help to spread general knowledge, transparency and scientific research to increase information on the scale and scope of the problem, raise awareness to develop coordinated policy solutions at global or individual country level

There is a wide body of literature on financial consumer protection issues in LMIC settings which document both the array of issues faced as well the impact of different mitigation interventions and policies. This body of evidence suggests that timely consumer information, reduced information asymmetries, and nudges can reduce susceptibility to a range of problematic behaviors from financial service providers and improve consumer welfare by reducing search frictions and likelihood of decision failures. While inchoate, there are also some growing examples (and related studies) of suptech solutions which demonstrate how they can increase efficiency and speed of resolving issues (e.g., for automating handling of customer complaints resolution).

The project aims to improve consumer protection and security in the fintech app markets. While this is pertinent in any context, it has particular relevance in lower-income community and country-settings. On the one hand, rapid digital transformation holds promise to increase financial access to marginalized populations, which can contribute to poverty reduction (SDG 1), improve overall well-being (SDG 3), foster further economic development (SDG 8), and help reduce inequality (gender, social, and otherwise) (SDG 10). On the other hand, there has been a documented rise in predatory and fraudulent practices in such markets, which creates direct risks to consumers and may drive persistent financial exclusion.  We will define success for the project through a variety of metrics.  

1. At a high-level, our primary objective is to improve knowledge sharing and awareness on consumer protection issues and mitigation methods surrounding finance app markets. To this end, we could measure success in terms of:
   • The number of B2G (government regulators), B2B (NGOs, academics), and B2C clients who are using our tool.
   • The number of conferences, seminars, workshops, etc. attended in which we present our tool or related research
   • The number of written materials (e.g., academic papers, policy briefs, blogs, etc.)
2. At a more targeted level, the desired impact is to reduce the likelihood that consumers in affected markets are falling victim to scam finance apps. To this end, we could measure success in terms of:
   • The number and relative propensity of problematic apps in a given market.
   • The number of consumers accessing problematic apps in a given market.
   • The length of time that problematic apps remain on a given market before being flagged and taken down.
   • Monitoring of consumer complaints (frequency, type, and semantic content) in affected markets
   • The number or rate of legitimate apps in a given market
   • The number of consumers accessing legitimate apps in a given market
3. Finally, as the ultimate objective is to reduce harms and improve welfare outcomes for consumers, we could also envision directly conducting an academic impact evaluation. For such an approach, we would seek additional academic research funding and consider setting up either an experimental or quasi-experimental approach (e.g., randomly targeted marketing or A/B testing for the B2C app or staggering entry of the B2G tool(s) into different markets) and conducting consumer-level surveys in order establish causal attribution of the effect of the tools on relevant household outcomes (e.g., such as those listed above relevant for the SDGs).

We leverage a variety of web/data automation, machine-learning, and ICT technologies in order to build and deliver our market monitoring tool. Our solution combines them in a way that allows us to increase transparency on the finance app markets and improve both screening and monitoring efforts.

First, in regards to web and data automation, we use Python programming to identify, parse, scrape, and store relevant app meta, download, and review data at regular intervals. Note that this includes information on apps that are publicly available, but often hidden under meta data.

Second, in terms of machine learning technologies, we use recent natural language processing models for analyzing textual content in the meta and review data for developing certain flags. Secondly, we use varied machine-learning algorithms in other predictive models. 

Finally, we would use a web- and/or app-based user interface in order to deliver our tool to and interact with end-users.

We also note that the approach and analyses inherently rely on Big Data and require significant computational capacity.

Given our past collaborations in academia and industry, we are used to creating data documentation, maintaining interpretable ML models, and meeting requests of stakeholders to conduct sensitivity and robustness tests on any developed models. While some underlying details would be proprietary, we could provide model cards and other interpretability aids.   

We currently have 4 individuals on our solution team. All team members are part-time staff, who are contributing between 20-40% FTE to the project. Depending on the amount of funding we can raise, we plan to scale up the team by either supporting more FTE for the current part-time staff or hiring additional staff or contractors to advance the project / tool development.

We have been working (part-time) on our solution for 2-3 years. The team leads, Jonathan Fu and Mrinal Mishra, began work on it as an academic project in Q2 2021 and completed a concept note and policy brief in Q1 2022. Between Q2 2022 and Q3 2023, they started transforming the prior work into an industry-orientated solution. Since Q4 2023, they brought on new team members to advance the project and further re-orientate towards the business development end of the solution.

Our team is committed to ensuring an inclusive and welcoming environment for all team members. While our current team is fairly small, we have a diversity of cultural backgrounds, professional experience, and technical competencies that position us well to solve the problem at hand. For example, this includes data scientists with experience in machine-learning applications (including natural language processing) in order to build and maintain the predictive models, research analysts with strong knowledge of financial markets and the context of local country markets for assistance in manual desk research and labeling tasks, and IT professionals who are specialized in web scraping and user-interface development. 

Moreover, as a key objective of this project phase is to scale and build a sustainable team, we would actively seek to include diverse team members in terms of gender, culture, linguistic, and other important dimensions in fulfilling our staffing requirements. We would follow established guidelines, such as those promoted by the U.S. EOOC or CIPD, to ensure equitable recruitment and promotion, ensure benefits packages that support employees with disabilities, create a safe and inclusive working environment, and gathering employee input. In a similar vein, we would provide support to any applicants or project team members with disabilities to ensure they have equal opportunity in consideration for their roles and/or ability to thrive in them.

For all team members, we would aim to pass skills, knowledge, and other resources (e.g., network) developed through our past research and activities and help mentor them towards career development in research and/or data science related to digital transformation. For example, we would plan to contribute a portion of funds to organize networking activities with external stakeholders and implementing partners in which our staff would participate. This would help with their career development and would be in support of broader sustainable development initiatives (particularly around financial inclusion and consumer protection).

Our value proposition is to improve consumer protection and general functioning of digital finance and fintech app markets. Potential customers and beneficiaries are varied and include the following:  

1. Financial regulators and supervisors / financial crime and law enforcement agencies (B2G):
   • Products or services provided: we would develop and provide varied reports on country-level finance app markets, including higher-level summaries of product and provider composition, types of consumer protection issues being observed, targeted lists of finance apps exhibiting problematic behaviors or characteristics; dashboards or reports triangulating on high probability scam app developers / providers
   • How do we provide these products? We envision an annual subscription-based approach where registered organizations would be provided access to a website to securely transfer reports and other target lists at regular intervals
   • Why do they need them? Mitigation of consumer protection issues and financial crimes are part of their mandate.
2. Global standard setting bodies, international development organizations, academic, NGOs (e.g., fintech associations) (B2B):        
   • Products or services provided: we will systematize data collection, analysis, and crisis mapping on the finance app market, globally and at the country level. While this approach would overlap to some degree with the B2G approach, it would place greater emphasis on systematic data scraping and provision, where we would serve as a repository to make data underlying our reports publicly available or available upon request for policy and academic research purposes (albeit, with some level of aggregation)
   • How do we provide these products? Via a dedicated end-user portal (website)
   • Why do they need them? Despite large supply and high demand, the finance app market remains notably opaque, with little systematized information on the full range of providers, products and services, and consumer protection issues contained within. We believe a market monitoring center has potential to generate wider interest to increase knowledge sharing and research on the fintech/finance app market. This could help reduce negative externalities from problematic providers on legitimate providers (e.g., reputational risks) and inform consumer protection regulation and policies.
3. B2C: App users / consumers:
   • Products or services provided: we would provide information services that improve transparency on apps that individuals are interested in using and/or automated diagnostics that run prior to downloading or installing app packages; at a minimum, this could include risk ratings for individual apps; however, it could also detail the characteristics that are raising red flags in order to help promote development of longer-term financial and digital capabilities in target markets.
   • How would we provide these products? We envision that the user interface would be either a standalone app or website where registered users would be charged a nominal annual fee for a subscription.
   • Why do they need them? Many country markets have seen a recent proliferation of scam finance apps. We seek to directly reduce consumer risk and harm from use of (predatory and fraudulent) digital financial services.

We have thus far raised money for our project through donations and grants. Specifically, we have received the following amounts to date from the following sources:  

• 2021 Consumer Protection Initiative, Innovations for Poverty Action / Bill & Melinda Gates Foundation - USD 65,000
• 2022 UZH Innovation Grant, University of Zurich - CHF 10,000
• 2022 Project Grant: Improving Sustainability in Lending, Schwyzer Winiker Stiftung - CHF 50,000

Our current objective at this stage is to develop a proof of concept for our solution that would allow us to start providing a subscription-based service to government bodies, other relevant organizations, or directly to consumers in target markets (i.e., via a B2G, B2B, and/or B2C approach). In order to reach a point of financial sustainability, we estimate needing to reach between 250,000 - 300,000 USD in annual revenue. We would adjust pricing based on the specific package of services being provided, however, we roughly envision initially charging subscription fees of between 50,000 - 100,000 USD annually for larger entities (e.g., government regulators, multilateral organizations, NGOs) and a nominal fee of 1 USD annually for individual customers (given the social orientation of the solution).

We note that while the business model approaches at first glance may seem different, the underlying data, analyses, and business development setup remain similar. Therefore, we believe there are significant synergies that would allow us to simultaneously explore them to see which has most potential to achieve financial sustainability, before selecting those to prioritize.