University of Nairobi, Kenya
* Corresponding author
Nexus International University, Uganda

Article Main Content

Reaching out to the financially excluded populations in Kenya with responsive financial services has been elusive for a long time. This is however changing through digital lending. Powered by Financial innovations in IT that include Artificial Intelligence, Machine Learning and Deep Learning among others, credit scoring is now possible for populations devoid of the traditional loan appraisal requirements or any form of credit history. This paper provides a review of the digital lending solutions on offer, the challenges and opportunities and a pathway for improved and more inclusive digital lending solutions for increased uptake and use.

Introduction

Kenya has seen a tremendous growth in trans-formative mobile money products among them digital lending. Leveraging on emerging technologies in IT among them Artificial Intelligence (AI), Machine Learning (ML) and Deep Learning, loan borrowers initiate applications through their digital devices to loan providers who process the loans digitally and disburse approved amounts to the borrowers’ virtual wallets (Franciset al., 2017). Loan borrowers download loan apps, register, apply for loans, get loans on their virtual wallets and repay digitally. In person engagement is minimal and credit scoring is depended on data and information from digital sources that include mobile money transactions, online purchases, social groups and credit reference bureaus among others for credit scoring and risk assessment (AFI, 2023). Since the introduction of digital lending in Kenya through m-swhwari in 2012, statistics show that 77% of borrowers in Kenya have taken only digital loans reflecting on the wider accessibility and reach of digital lending borrowers (Financial Sector Deepening (FSD Kenya, 2019a)) as well as the possibility that these borrowers do not meet traditional lenders thresholds. The design and delivery of digital loans is mobile banking, mobile money and unstructured supplementary service date (USSD) and mobile apps giving these models the advantage of reaching out to customers in remote places and without the necessity of volume. Each individual joining the network from their locality constitutes the volume required for sustainability.

Consequently, the ability of lending solutions to leverage on technology and perform due diligence minus the traditional credit history considerations as well as the need for collateral elevates the model above the traditional lending models. Various models on offer have sought to fill in the gap of financial exclusion and it is necessary that the extent to which mobile money drives financial inclusion and how the product execution should be modeled to enhance it is explored. Below is a diagram showing the various steps in digital borrowing (Fig. 1).

Fig. 1. Digital Borrowers Journey. Source: Study.

While great strides have been made around financial inclusion, there’s still a long way to go. According to global Findex data (The Banker, 2022), over 1.7 billion in the world lack basic financial services. Additionally, over 50% of deserving populations in Africa are un-banked, translating to over 350 billion people. Despite the huge population excluded from formal financial services in Africa, digital lending solutions promise to fill this gap (World Bank Group, 2018) Banking on the characteristics of ICTs (Castells, 2004): Their self-expanding processing and communicating capacity in terms of volume, complexity and speed; The ability to recombine on the basis of digitization and recurrent communication; and Their distributing flexibility through interactive, digitized networking, digital lending has the potential to overcome barriers of geography increasing reach and reducing costs (Banna & Alam, 2021) through the use of big data and cloud computing.

Since its entry into the market in Kenya through M-pesa in 2007, the uptake and use of digital financial services have been on the increase with the COVID-19 pandemic accelerating its use as a result of the measures put in place to mitigate its spread (Wangariet al., 2021). Among the measures were lock downs and social distancing meaning that face to face transactions were discouraged as well as movement to and from financial institutions. Conditioned to a virtual culture during the period, digital lending has persisted post COVID and into the period of economic recovery (World Bank, 2021). The greatest beneficiaries are the Medium and Small Enterprises (MSME’S) as well as households enabling them to manage their financial affairs with ease and unleashing their potential. Interesting is how digital financial use has seamlessly integrated into the lives of even the lesser off in society requiring little or no additional skills. Although digital lending has demonstrated inclusivity for underserved or un-served households and businesses enterprises, its execution is riddled with challenges (Central Bank of Kenya, 2021) that call for regulation and policy for responsible and sustainable execution. Early signals of concern include the inability to repay these loans on time by some of the borrowers affecting their welfare as well as the profitability of providers and in effect sustainability of the loan products. Devoid of clear regulation, new players with little experience are entering the market, without clear supervisory obligations and operational guidelines opening space for severe vulnerabilities.

In addition, the fact that these solutions are downloadable freely from online sources, their convenience and simplicity for customers, provide avenues for exploitation of unsuspecting customers. Without online gate keeping, unscrupulous lending apps find their way on online stores ready to reap from unsuspecting new to credit customers. Devoid of options as a result of a perennial exclusion, undeserved households and SME’s with limited financial literacy have been lured into taking up unsuitable and un affordable loans, leading to over-indebtedness, escalating poverty and bankruptcy (Leong & Sung, 2017). Short loan repayment periods for substantial loan amounts also pose a problem of repayment. This is because business enterprises need a longer period of time for return on investment as this amount are invested into different factors of production like labour, goods, services and even equipment. Shorter periods of repayment mean that the borrower returns what they haven’t invested eating into their capital. Besides, financial inclusion as measured through the 2021 Fin access surveys in Kenya (Finaccess Survey, 2021), does not only measure access but go beyond access to capture usage, quality, impact as well as welfare. In this effect then, over indebtedness, financial distress and costly credit all contribute to unhealthy undertaking defeating the financial inclusion sought.

Benchmarking availed products with the prerequisite qualities of an inclusive financial inclusion product, digital lending products in Kenya fall short of these qualities. A Fin Access study (2021) describes an inclusive financial system as one that has strong consumer protections, targeting bad actors and activities that can cause harm to people, small businesses, and the financial system overall. This includes behavior that is abusive, deceptive, or criminal. The unfortunate scenario however is that these tenets have been flouted necessitating the strict compliance through the Central bank of Kenya digital Lending law 2021. That notwithstanding predator apps freely downloadable from play store have continued to perpetuate the malpractices singled out by the central bank of Kenya. These laws aim at protecting the users from abuse. This is exampled in the fine of Ksh 2,975,000 meted on Mulla Pride Ltd, a company operating two lending apps; KeCredit and Faircash in Kenya found culpable of using names of next of kin provided by borrowers to threaten the borrowers. (Citizen Digital, 2023)

Keen on ensuring compliance and protection of users, the Central Bank of Kenya through the Digital Credit Providers Regulation has provided clear guidelines on the dos and don’ts in digital lending law of 2021 (Central Bank of Kenya, 2021). With the increasing innovations in financial inclusion, there is need to put in place regulation that ensures that data provided by borrowers is not shared with third parties as well as its responsible use considering that most of the new to credit customers may not be clear about their rights and hence need to be protected since they are vulnerable. It is further prudent to balance these regulations in order to allow for the benefits of the new innovations as well as growth on both the demand and supply sides. While the drivers of financial inclusion are access and inclusivity, the products availed must go beyond to ensure that there is value in use by the customers with indicators like actual use of financial products like having accounts, payment of bills, access to credit and insurance and ability overcome financial shocks. With such outcomes, users will experience better incomes, increased wealth and better livelihoods (Fin Access, 2021).

With the foregoing description of an ideal financial inclusion product, it is also logical to understand that digital lending is new and innovative in nature with both the lenders and borrowers experimenting as the product evolves. The execution model driven by the need to devolve services and reach out to the previously excluded segments comes with novelty and we are seeing use of agents as opposed to establishment of bank branches; solutions developed by non-private and non-financial players the Fin-techs and the Big-techs; with carefully crafted regulatory to allow the benefits of these products to be realized as opposed to the stringent regulation governing traditional providers; costing of products based on the new credit scoring parameters where interest may vary depending on the rating of the individual borrower; increased requirement for customer data and in digital form hence prone misuse and therefore need for data protection guidelines; An emergent culture of quick to act tendency that makes users disregard any terms and conditions before proceeding to request for the loan: and Independence and privacy in action where users interact with the system and make decisions singly whether beneficial or risky to their financial welfare. Giving room for innovation and adopting the popular trajectory of ‘innovate then regulate’, digital financial inclusion adoption has been riddled with challenges for the same vulnerable financially excluded and underserved customers that benefit from the opportunities, these new challenges include; novelty risks for customers due to their lack of familiarity with the products, services, and providers and their resulting vulnerability to exploitation and abuse mostly because they lack financial literacy. Noted also in the Microsave (2019) study are agent-related risks due to the new providers offering services that are not yet fully understood and regulation yet to find footing, digital technology-related risks that may disrupt services and loss of data, as well as the risk of privacy or security breach resulting from digital transmittal and storage of data. The aim of this paper therefore is to examine the models on offer, potential technologies to improve on the models and recommend a suitable model going forward

Problem Statement

Despite the envisioned potential in digital lending, the execution model is attracting concerns from regulators as well as an outcry from users themselves (FSD Kenya, 2019b). Among concerns raised are low credit limits, high interest rates, short repayment periods, blacklisting, denial of loan requests and stressful recovery techniques by the lenders (Leong & Sung, 2017, Sommer, 2021). The CBK also has raised concerns in regard to multiple borrowing by customers, uncouth debt recovery tactics where lenders call third parties to pressure borrowers to repay, unkind and abusive messages as well as the deceptive targeting of new to credit borrowers all leading to poor financial health and threatening the sustainability of the models (CBK DL 2021). Besides new to credit customers lack financial literary and end up defaulting on small amounts of money. Considering that these platforms are fully automated, these customers end up getting blacklisted and unable to access any other loan from any other provider. In addition, most of the unbanked populations find it difficult to access loans that can transform their livelihoods substantially mainly because the new models rely on digital data. Considering that majority of the new to credit customers lack significant digital foot print to base decisions on, data models provided are yet to come up with methods of capturing alternative data and providing the right scores outside of the discriminative models that consider lack of credit history as an automatic red flag and therefore risky. It is because of these emergent outcomes that this paper seeks to underscore the methods used to determine the creditworthiness of the new to credit customers, address concerns raised in the fin access 2021 report, examine pitfalls in the models and possible ways of improving these models to be more inclusive and responsive to the financially excluded segments mitigating discrimination and continued exclusion

General Objective

To gain an in-depth understanding of the technologies used in digital lending with the aim of recommending a sustainable trajectory for financial inclusion

Specific Objectives

1. Determine credit scoring mechanisms in digital loans

2. Evaluate technologies that support increased uptake and use

3. Discuss challenges of financial inclusion in digital lending models

4. Propose interventions for a sustainable digital lending execution model

Literature Review

Digital Lending Credit Scoring Mechanisms

Unlike traditional methods that rely on a customer’s credit history and collateral to determine the amount of loan a customer is worth, digital lending relies on Credit data to inform risk-based decisions. While the new lending model depends on digital data for credit scoring, the greatest challenge lies in the fact that the unbanked populations rarely participate in digital activities. Characteristically excluded because of logistical issues like lack of electricity, internet connection, digital devices or their day-to-day activities with majority in rural areas spending most of their day in their farms, these segments lack substantial digital footprint to aid in that data driven risk based decision making. The absence of a previous credit file, or likely to have one that is influenced by unfair or discriminatory treatment as a result of their logistical exclusion, leads to credit that is either more expensive or altogether unavailable. Historical discrimination occasioned by the possible reasons identified and or invisibility in credit exacerbates the discrimination and invisibility (Sahayet al., 2015; Barajaset al., 2020).

Assessing a Customer’s Creditworthiness in Digital Lending

Fintech.co.ke (2023), writing for the Kenya Fin-tech Association gives an overview of critical consideration in use by various fin-techs in Kenya. Unlike traditional banks that rely heavily on credit bureau scores, fin-tech companies in Kenya have embraced alternative methods to assess a borrower’s creditworthiness, revolutionizing the lending landscape. While traditional banks have a standard way of granting credit facilities based on financial documents provided by the borrower, fin-tech entities try to assess the ability to pay and the willingness or intention to pay through an alternative method. Generally, this alternative method is understanding the customer profile based on specific pre-set parameters like understanding their personal information, M-pesa transactions, social information through digital footprints, use of phone, actions on various social media platforms, understanding the call behavior in terms of calls made, calls received, calls missed, duration of the call among others.

Digital lenders analyze borrowers’ M-Pesa transaction history to understand their cash flow, spending habits, and repayment capacity. Regular and timely transactions might indicate financial stability, positively influencing the credit decision. Fintech companies also explore borrowers’ social media activities to understand their behavior, interests, and social connections. Engaging with reputable brands and active participation in professional networks can be considered positive indicators for creditworthiness. Borrowers’ call behavior, such as frequency, duration, and missed calls, can reveal patterns related to their reliability and financial stability. For instance, consistent and responsible communication might reflect positively on a borrower’s character. Analyzing a borrower’s digital footprint, including online interactions and behavior, can provide additional insights into their financial habits and personality traits. Algorithms are then developed incorporating M-Pesa transaction history, social media activities, and mobile phone usage data to assess a borrower’s creditworthiness. The algorithm assigns different weights to each parameter based on historical performance data and market trends, generating a customized scorecard for each borrower. Consequently, rather than relying solely on standardized credit bureau scores, using machine learning and AI to enable continuous refinement and enhancement based on real-time data, Kenyan Digital Credit Providers (DCPs) adopt customized scorecards to cater to various borrower segments and products. For example, a scorecard of a lender that offers loans to SME will be tailored to evaluate business owners based on their m-pesa transaction history, business revenue patterns, and social media presence. Depending on the performance of the initial loan, the scorecard gets better or is downgraded ensuring precise risk assessment.

In addition to the credit bureau and KYC verification, lending apps incorporate analyses for underwriting and fraud detection in order to protect borrowers and lenders from potential risks. These mechanisms leverage advanced technologies to identify fraudulent activities and patterns in real-time. These include AI-driven algorithms to monitor borrower behavior and identify suspicious patterns that might indicate fraudulent activities. The system flags accounts showing multiple loan applications within a short time-frame, attempts to borrow from various lenders simultaneously, or engages in other abnormal behaviors, prompting further investigation.

Another useful source of customer data is the Credit Reference Bureau (CRB) factored in at the up of a Digital lending platform. There are three credit Bureau agencies in Kenya, namely: TransUnion, Creditinfo Kenya Ltd and Metropol Corporation Limited Fintech.co.ke (2023). These CRBs provide API integration with lending apps to automatically pick up specific information from their reports and run tools to analyze the data provided by them. Depending on the CRB chosen and the datasets subscribed to, a lender can create an alternate scorecard for the customer to arrive at a final scorecard. Examples of information fetched from these bureaus are; the Customer’s credit score, total loan outstanding Loans overdue within 30 days or 60 days, or 90 days, amounts written off, number and total outstanding balances of running loans, etc. CRBs are particularly important in assessing a client’s ability and willingness to pay. For example, if a customer has written off loans, this may be a cause for concern. Similarly, too many inquiries in a month reflect the customer’s desperation. A lending fin-tech would build its rule engine based on these parameters.

Considered are also other qualitative factors as they encompass various non-financial parameters that provide insights into a borrower’s character, reputation, and social behavior. These factors can be equally vital in assessing a borrower’s creditworthiness, complementing the traditional credit score derived from financial data. Lenders can gain a more holistic view of their creditworthiness by analyzing a borrower’s behavior beyond just financial transactions. It’s important to note that qualitative factors do not replace traditional credit scores but rather supplement them. Traditional credit scores based on financial data provide valuable insights into borrowers’ credit history, while qualitative factors offer a glimpse into their character and social interactions. For example, a lending platform may assign higher weight to a borrower’s active participation in professional networks on LinkedIn, viewing it as a positive indicator of stability and repayment potential. On the other hand, frequent and excessive interactions on gambling websites might raise concerns and warrant a lower score. Another example, a DCP might use Natural Language Processing (NLP) algorithms to analyze a borrower’s posts, comments, and interactions on social media. The tone of their language, the topics they discuss, and the people they associate with can all provide valuable insights into their personality and reliability.

Bigtech and Fintech Companies

The emergent models in digital lending targeting the financially excluded populations are driven by Fintech and bigtech companies. Digital technologies in use particularly machine learning, AI and DL technologies rely on easily available data that they can obtain from their clients’ digital footprint or by scraping the Web (Frostet al., 2019). The difference between big tech and fintech is that while fintech are startup companies that develop digital lending apps and rely on third party sources to access customer data, bigtech have the advantage of being custodians of huge chuck of that digital data particularly on mobile phone use. An example of a big tech is a teleco company like safaricom that controls a customer’s calling data, browsing, mobile money use including online purchases as well as customer networks like Whats App, Linked-in and other social groups and using this alternative data to rate customers is easier than a fin-techs that have to procure these data. Artificial intelligence also allows big tech companies to convert soft information collected through social media or other means into hard information. Consequently, the network advantage allows them to access more data, improve their models, and ultimately increase their outreach further. As established in the Microsave (2019) and Mutheu (2023) studies, Big-tech companies were better rated that fin-tech companies with most fin-tech being accused of high interest rates, calling defaulters and their next of kin as debt recovery mechanisms and also very law loan limits. This is an indicator of poor investment in technology that is near accurate in defaulter assessment, credit scoring as well as debt collection. That notwithstanding, several studies show that the credit scoring mechanisms of platform data used by both fin tech and big-tech companies perform better in predicting default compared to banks relying on credit registry data.

Technologies in Credit Scoring

In digital lending, a variety of scoring techniques are employed, ranging from conventional statistical methods to cutting-edge approaches. Artificial intelligence and machine learning are two of the newest, cutting-edge technologies used in credit scoring. The use of computational tools to accomplish tasks that have historically required human sophistication is known as artificial intelligence (AI) (World Bank Group, 2019a). AI gives machines the ability to adapt to new inputs, learn from experience, and carry out tasks that humans would perform (FSB (Financial Stability Board), 2017). Deep learning: a technology that recreates the brains power and natural language processing are major components of most AI. These methods make use of computers’ capacity to learn from experience and carry out tasks that enable augmentation of digital data and information of an individual and create a credit scoring parameter out of the various data obtained. This is as a result of the rapid advancements in fundamental technologies like computing power, data, and creative algorithms today. These technologies allow computers to be trained to process and identify patterns in data, even though the data may come from various sources and be of different types, in order to perform specific tasks. This makes it possible to match an individual’s data sourced from mobile wallet transactions, e-commerce activities, credit reference bureau’s, social media engagements including a borrowers online groups and associations, calling habits among other data used in credits scoring, analyzed and make relevant conclusions that facilitate assigning of a weight that is used to determine the credit score that informs a customer’s credit limit and terms and conditions of the digital loan to be advanced. All these are determined without human intervention (World Bank Group, 2019a). Noticeably is the fact that these methods are progressively being applied to identify intricate patterns in vast datasets from progressively inventive and varied sources that continue to evolve (SAS 2019) enabling populations that would otherwise live with financial exclusion get financial services. Amazingly too is the fact that, deep learning algorithms provide the structure and operation of the human brain discerning underlying human tendencies that provide certain indicators useful in understanding the behaviors’ of borrowers (SAS 2019). Deep learning algorithms have found relevance in reinforcement learning, unsupervised learning, or supervision used in predicting future occurrences.

Machine Learning (ML) is considered one of the most innovative technologies in the 20th century with the potential of resolving the credit scoring discrimination evident in the early digital lending models. ML has the capacity to access raw data, combine, join and aggregate input data; engineer features either manually or sing an expert, input the features and select the useful features then apply ML algorithm to the training data set and interpret and assess the results. To debunk these possibilities, from a data set obtained from a lender, ML can be used to fetch these data, identify the various combination of features for example interest on loan, repayment period, category of risk, other features like gender, income, credit history among others and return scores for the categories. From these features ML has the capacity to combine useful features as identified in the recommendations (Mutheu, 2023) and train the data and interpret and assess the results. By comparing returned results for various combinations, it is possible to identify the most suitable model for increased uptake and use of digital lending solutions.

Supervised learning is an additional method used in credit scoring. In this case, independent features or variables and data with labels, such as dependent variables or events, are used to develop the algorithm. This could be gender, association e.g., being on LinkedIn or a known online professional group, consistency in bill payment or even volume of money held in virtual wallets in a month. Next, the algorithm uses these features to predict values of the label of interest that are unknown or in the future. The algorithm then picks up a general classification rule that it can apply to forecast the labels of additional data set observations. Regression, decision trees, random forests, gradient boosting, and deep neural networks are some of the supervised techniques that are in use in digital lending. Typically, decision trees are schematic diagrams in the form of trees that are used to display statistical probabilities. One of the most popular supervised learning methods is classification and regression trees (CART), which divide data into subsets by continuously identifying the best feature. The partition improves the isolation of the label with each split. Decision trees can be used in digital lending to identify the best features for digital lending and dropping those that are unfavorable. Random forests are a combination of tree predictors such that each tree depends on a sample or subset of the model development data (or training data) selected at random (Breiman, 2001). Working with multiple different subdata sets can help reduce the risk of over-fitting. To demonstrate, in determining various variables in digital lending, customers can be asked about different attributes for example interest, repayment period, loan amounts etc. Depending on the choices the respondents favor, those not preferred are dropped and the preferred ones are subjected to more options with an aim of arriving at a combination of favorable features as the final output. The combination is then validated through actual testing on users.

Deep neural networks (DNN) train the algorithm to learn on its own by recognizing patterns using multiple layers of processing, as opposed to organizing data to run through predefined equations. Sieving through massive data of a client and processing each set incrementally on the output of the layer before it, each layer of nodes trains on a set of features. As a result, as one moves through each layer, more intricate layers of features are produced, each of which aggregates and recombines knowledge from the one before it. Because of this feature, deep learning networks can recognize extremely intricate nonlinear patterns with massive amounts of data and dimensions (Press 2017) which in effect increases accuracy of decisions taken. Conversely, over-fitting of deep neural networks may also occur though this is mitigated through validating of the new model. DNN can be used to systematically process a customers massive data to incrementally decipher patterns of behaviour that can lead to fair scoring.

Unsupervised learning techniques are employed in addition to supervised learning techniques. Unsupervised learning techniques involve feeding an algorithm with data that lacks labels, or events. For example, the algorithm must find clusters of observations that show similar underlying characteristics in order to identify patterns in the data. This is important for us because new to credit customers lack the known parameters for credit scoring, but data obtained though scanty provide the basis for analysis and therefore the algorithms do not forecast new or unknown data; rather, they investigate the characteristics of the novel data that has been analyzed. K-means clustering, hierarchical clustering, and clustering are examples of unsupervised techniques that if improved have the potential to respond to the peculiar characteristics of thin file customers as the resultant are groups rather than classes because, in contrast to classification, clustering analyzes the data to produce the class label rather than the data being labeled. With special training to ensure biases as a result of exclusion are shelved, produced labels may provide the actual rating of new borrowers lacking extensive digital data and information. In order to maximize similarity within a group and minimize similarity between groups, the data are grouped according to this principle. In other words, groups are created so that the objects within them are extremely distinct from those of other groups while also being extremely similar to one another. Rather than being predictive, clustering algorithms are descriptive. A clustering algorithm could be used, for instance, to find a borrower who shares traits with a borrower who is challenging to evaluate. The average default assessment of the cluster—should the algorithm identify a suitable cluster for the borrower—may serve as a rough approximation of the borrower’s default assessment. Unsupervised techniques may be very useful in avoiding over fitting and profiling of customers.

Additional Methods Associated with Credit Scoring Systems Incorporate Automated Feature Engineering (AFE). The effectiveness of feature engineering, which comprises turning the input data set into features in order to better comprehend the underlying structures in the data and increase model accuracy, is crucial to the success of machine learning models. While most of the new to credit customers lack requisite data for credit scoring, there are reasons for the shortfall and AFE may be used to investigate the data for a better understanding and fair scoring. This process may be iterative with different feature sets being developed and assessed at different stages. As a result a large number of feature candidates may be produced by this process, necessitating the careful selection of pertinent candidates in order to prevent over-fitting in later model developments meaning that improvement of models have to be continuous in order to address emerging data characteristics. Another method that lies in the middle of supervised and unsupervised learning is called reinforcement learning. Through reinforcement learning, a machine can pick up behaviors based on input from its surroundings (Champandard & Alex, n.d.). The objective is to train an algorithm that learns the best course of action by taking into account the environment, acting, and receiving feedback in the form of rewards from the environment with the optimal conduct optimizing the benefit without the need for human intervention.

Credit scoring also makes use of a machine’s capacity to comprehend, analyze, and produce human language, including speech through Natural Language Learning. NLP, a subfield of AI, aids machines in comprehending, interpreting, and using human language (World Bank Group, 2019b) and its the process by which a machine attempts to decipher spoken or written language. Beyond comprehending language structure, it includes a system’s capacity to understand intent, resolve context and word ambiguity, and even produce coherent language on its own. Character, associations, and language can be a factor in measuring a borrowers goodwill. Lastly, block-chain plays a crucial role in guaranteeing that customer behavior across various providers is recorded. Data must be verified by several sources in order to be updated on the block chain ensuring that borrowers activities are updated, and it is possible to even quantify their engagement with other providers. Block chain provides traceability of a customers activities online and for this matter in borrowing. Considered secure because of the absence of centralized points of vulnerability which could be entry points for insecurity, lending apps that integrate block chain technology are able to store data with a far higher level of security. Stored data is then easily and securely shared among lender companies to give consumers a more secure way to get their credit scores and share credit history with other providers. This technology can be explored further to mitigate over indebtedness, multiple borrowing and also capture a customer’s borrowing history and hence accurate assessment.

Conclusion

The forgoing discussions provide hope in the search for responsive models for thin file customers as well as new to credit customers. An integration of the various technologies discussed in lending apps with the aim of mitigating the shortfalls of each other may be the best trajectory in addressing the gaps in the existing digital lending models availed in Kenya. Besides the discussed technologies, there is need for stakeholder discussions on the best execution approach; prioritizing financial education of new to credit customers, regulation to deter opportunist merchants from taking advantage of vulnerable first time borrowers or those excluded because they are poor as well as coming up with humane and realistic guidelines on how to treat customers while protecting businesses.

Recommendations

Below is a summary recommendation of some of the technologies and their role in addressing the gaps in availed lending models. While the table below captures some of the discussed technologies in this paper, it does not in any way suggest that this is the only viable trajectory and would further call on other scholars in the field to identify other technologies that could further improve on the models on offer. This paper therefore recommends modeling of a financial inclusion solution guided as given in Table I.

Intervention Purpose
Machine Learning Algorithms that are specific to borrowers’ conditions. For example, while it is easy for traditional banks to schedule repayment of loans for schools in termly basis and particular months of the term, or for farmers after a planting cycle, current lending apps have short repayment periods that have no consideration for this kind of information which is critical. ML algorithms can be trained to recognize specific information provided at the time of application so that loans are customized as this may reduce default and attract more borrowers. ML Techniques that can be used here include automated feature engineering and reinforced learning.
Artificial Intelligence This embedded in lending apps will facilitate recognition of patterns in varied data, according to new customers points based on similarities in actions of other borrowers with sufficient foot prints. AI will also assist in identifying a customer interests and online preferences hence improved scoring accuracy.
Deep Neural Networks DNN can be used to systematically process a customers massive data to incrementally decipher patterns of behaviour that can lead to fair scoring.
Clustering This is a necessary technique when combined with AI to assume characteristics of a population with similar attributes. For example new customers belonging to a social group where existing customers have exhibited good performance.
Block chain This technology will help in keeping a trail of actions so that the system can make the best decision for both customers and lenders. It is particularly useful in flagging multiple borrowing, over-indebtedness and credit history.
Reinforcement Learning It is critical that decisions are based on both the information provided as well as factors that affect the actions. As mentioned earlier farmers will invest during the farming period but once they harvest, they have the capacity to repay loans advanced in full. Loans must not be a one size fits all model but consideration of factors that are specific and constitute a healthy lending environment for all players. In addition, its inc.
Automated Feature Engineering This is useful in ensuring that input data from new to credit customers is investigated for the purpose of understanding underlying structures in the data in order to eliminate bias and increase model accuracy.
Regulation Regulation is important to set the rules of operations, licensing, governance, for identifying the genuine players and ensuring that action is enforced at the times of infringement
Stakeholder forums Borrowing and lending is a human undertaking, offline discussions are necessary to identify aspects that matter. For example, it is inconveniencing for a customer to be blacklisted and denied a loan because of defaulting very small amounts of money.
Table I. Digital Lending Financial Inclusion Guidelines

References

  1. AFI. (2023). The supervision of fintech in the african region framework 2. (accessed on 13 May 2024). https://www.afiglobal.org/wp-content/uploads/2023/02/The-Supervision-of-Fintech-in-the-African-Region.pdf.
     Google Scholar
  2. Banna H., & Alam M. R. (2021). Impact of digital financial inclusion on ASEAN banking stability: Implications for the post-Covid-19 era. Studies in Economics and Finance, 38(2), 504–523. https://doi.org/10.1108/SEF-09-2020-0388.
     Google Scholar
  3. Barajas, A., Beck, T., Belhaj, M., & Ben Naceur, S. (2020). Financial inclusion: What have we learned so far? What do we have to learn? Washington, DC: International Monetary Fund. https://www.imf.org/en/Publications/WP/Issues/2020/08/07/Financial-Inclusion-What-Have-We-Learned-So-Far-What-Do-We-Have-to.
     Google Scholar
  4. Breiman, L. (2001). Random Forests. In Statistics department, university of California–Berkeley, Berkeley.
     Google Scholar
  5. Castells, M. (2004). The Network Society: A Cross-cultural Perspective. Northampton, ME: Edward Elgar Publishing, Inc.
     Google Scholar
  6. Central Bank of Kenya. (2021). Central bank of Kenya (Digital credit providers) regulations, 2021, Act (Cap 491). https://www.centralbank.go.ke/wp-content/uploads/2022/03/L-.N.-No.-46-Central-Bank-of-Kenya-Digital-Credit-Providers-Regulations-2022.pdf.
     Google Scholar
  7. Champandard, & Alex, J. (n.d.). “Reinforcement learning.” Artificial intelligence depot. http://reinforcementlearning.ai-depot.com/.
     Google Scholar
  8. Citizen Digital. (2023). Digital lenders warned against calling families of loan applicants. https://www.citizen.digital/news/digital-lenders-warned-against-calling-families-friends-of-loan-applicants-n328337.
     Google Scholar
  9. Finaccess Survey. (2021). 2021 Finaaccess house hold survey. https://www.knbs.or.ke/wp-content/uploads/2021/12/2021-FinAccess-Survey-launch.
     Google Scholar
  10. Fintech.co.ke. (2023). Understanding the structure of a successful digital credit provider in Kenya. https://fin-tech.co.ke/2023/07/25/understanding-the-structure-of-a-successful-digital-credit-provider-in-kenya/.
     Google Scholar
  11. Francis, E., Blumenstock, J., & Robinson, J. (2017). Digital Credit: A Snapshot of the Current Landscape and Open Research Questions. CEGA White Paper.
     Google Scholar
  12. Frost, J., Gambacorta, L., Huang, Y., Shin, H., & Zbinden, P. (2019). BigTech and the changing structure of financial intermediation, Economic Policy.
     Google Scholar
  13. FSB (Financial Stability Board). (2017). Artificial intelligence and machine learning in financial services: Market developments and financial stability implications. FSB, Basel, Switzerland. http://www.fsb.org/wpcontent/uploads/P011117.
     Google Scholar
  14. FSD Kenya. (2019a). Digital Credit in Kenya: Facts and figures from FinAccess 2019. Focus Note. Nairobi: FSD Kenya. www.fsdkenya.org.
     Google Scholar
  15. FSD Kenya. (2019b). 2018 Annual report. Nairobi, Kenya: FSD Kenya. Fsdkenya.org.
     Google Scholar
  16. Leong, K., & Sung, A. (2017). FinTech (Financial Technology): What is it and how to use technologies to create business value in fintech way?. https://www.researchgate.net/publication/334364303_FinTech_Financial_Technology_What_is_It_and_How_to_Use_Technologies_to_Create_Business_Value_in_Fintech_Way.
     Google Scholar
  17. Mutheu, M. (2023). Examining Digital Lending Models for Financial Inclusion, A Case of Nairobi County. An Unpublished Research Study Towards the Award of a Master’s Degree in ICT4D at the Nexus. International University Kampala.
     Google Scholar
  18. Microsave. (2019). Making digital credit truly responsible. Digital credit kenya final report. https://www.microsave.net/2019/09/18/11310/.
     Google Scholar
  19. Sahay, R., Cihak, M., N’Diaye, MP., Barajas, MA., Pena, MD., Bi, R., Gao, M., Kyobe, MA., Nguyen, L., Saborowski, C., Svirydzenka, K. (2015). Rethinking financial deepening: Stability and growth in emerging markets. (Washington, DC: International Monetary Fund, May 2015). https://www.imf.org/external/pubs/ft/sdn/2015/sdn1508.pdf.
     Google Scholar
  20. Sommer, C. (2021). Addressing the challenges of digital lending for credit markets and the financial systems in low- and middle-income countries. idos german institute of development and sustainability. https://Www.Idos-Research.De/En/Briefing-Paper/Article/Addressing-The-Challenges-Of-Digital-Lending-For-Credit-Markets-And-The-Financial-System-In-Low-And-Middle-In-come-Countries/.
     Google Scholar
  21. The Banker. (2022). Bridging afircas financial gender gap, africas digital financial journey. https://www.afi-global.org/wp-content/uploads/publications/2017-11/AFI2017_Gender_full_AW_ISBN_digital.pdf.
     Google Scholar
  22. Wangari, E. N., Gichuki, P., Abuor, A. A., Wambui, J., Okeyo, S. O., Oyatsi, H. T. N., Odikara, S., & Kulohoma, B. W. (2021). Kenya’s response to the COVID-19 pandemic: A balance between minimizing morbidity and adverse economic impact. AAS Open Research, 4, 3. https://doi.org/10.12688/aasopenres. Working Paper no. WP/2022/080.
     Google Scholar
  23. World Bank. (2021). The global findex database 2021: Financial inclusion, digital payments, and resilience in the age of COVID-19. https://worldbank.org/en/publication/global.findex.
     Google Scholar
  24. World Bank Group. (2018). The future of financial inclusion in Africa. https://documents1.worldbank.org/curated/ru/719111532533639732/pdf/128850-WP-AFR-Digital-Access-The-Future-of-Financial-Inclusion-in-Africa-PUBLIC.pdf.
     Google Scholar
  25. World Bank Group. (2019a). Credit scoring approaches and guideline. www.worldbank.org.
     Google Scholar
  26. World Bank Group. (2019b). National Financial Inclusion Strategies Resource Center, Policy brief . Washington, DC: World Bank Group. https://www.worldbank.org/en/topic/financialinclusion/brief/financial-inclusion-strategies-resource-center/.
     Google Scholar