As per the latest world bank report (2022), 26% of the adult population doesn’t have a bank account yet while only 24% owns a credit card. The stats are worse for Women and population outside the developed economies. Access to Credit has been identified as one of the major force multiplier for reduction in poverty. Few very interesting stat is 86% of the world population (Age 15+) is having access to mobile and 63% is having access to Internet. 64% of the population have either made or received payments digitally. This indicate a huge eligible market for Credit products including Digital lending, BNPL & Credit Cards.
The major barrier to access to credit includes lack of credible scores, avoidance of debt including for religious reasons, absence or difficulty to reach out to financial institutions among others. Among this, credible credit score including ineligibility in bureau has been marked as the single most important reason for financial under exposure.
What is a Credit Score and how one develops it??
The Credit Score is an insights on the reported financial history of any individual. To put it simply, it defines your credit worthiness and risk profile from a transactional perspective. It covers your exposure to financial world including number of accounts opened, lines availed (Read number of loans/ Card / BNPL taken) and performance on these products.
There are multiple companies including Experian, Equifax, CRIF Highmark, Transunion who specializes in creating scores for customers. While there are many scoring models, the one created by Fair Isaac Co is the most famous (FICO Score). In India, TU has tied up with several banks to create CIBIL again based on FICO model. The scores vary between 300- 850/900 depending on bureau. Generally, to create the score, customer’s size and count of exposure, imputed income, DPDs, Past hard & soft delinquencies, type of exposure (Secured / Unsecured), utilizations, FOIR is considered among other factors. Vintage of lines, addresses, no of enquiries are few other factors which impact it.
One of the most important requirement for creating the bureau score is reporting of exposures & lines by the Banks /NBFCs. All customers who don’t have bureau scores are generally termed as NTC (New to Credit) or NTB (New to Bureau) and are assigned either a ‘0’ or ‘-1’ score. This is a catch-22 situation for such customers as to get the bureau score, most institutions look a previous history which is absent for these cases. Few of the customers based on their profile, income and company may get it but a larger chunk remain excluded from mainstream modes of credit. There are 1.7B unbanked population in World with 19% customers in India still not having bank account. Less than 15% of the population have access to formal credit. Situation is much worse in LDC (Lowest development countries).
The answer lies with Alternate Scoring Model
So, as realized above, the major reason for not having access to formal credit is absence of sufficient financial tradelines through which a bureau can create the scoring for the customers.
Here comes Alternate scoring models. While these customers have limited exposure to financial world, almost 5 out of 6 people are using mobile and 75% of these people are having access to digital landscape either through smartphones or internet and have done some kind of digital payments. In India, 81% of the customers have access of banks and good 64% have used UPI to make payments (World Bank, 2022). A larger chunk of customers also have huge social media footprints and have been active users for some time. Many of these customers while not having bureau scores are working full time, have sufficient income sources or are well capable of paying back. Add to this, large parallel & informal economies in many countries which employs a large part of these customer set and it is a very potent customer base.
Your legacy scores / credit history can’t be the only judgement for future approval!!
Alternate scoring models are created for these customers with the help these alternate mediums of information including customer’s financial transactions, mobile imprints, social footprints, emails, demographics, Geographical indicators, personas among others. There are various agencies who excel in creating alternate scoring models. Some claim to use as many as 10000+ different data points to create such scores.
Let’s see these in details:
Non-Transactional:
- Demographics: Age, DOB, Gender, Education, Marital status, Employment type, Type of accommodation, and so on
- Geographic: Locality, Pincode, Address, frequency of visits, map based tracking, travel distance, time, duration, frequency etc
- Mobile Imprint: For customers who use mobiles, it provide a wealth of information about them:
- Soft Mobile: Number of apps, no of contacts, Demographics from customer data, Type of mobile, software, type of apps, type of connection, vintage of connection,
- Deep Mobile: Type of mobile usage, transactional data, type of promotional messages, typos & pronunciations, Mobile data used, accessibility given to app, app usages and sites visited, Browser history, accessibility provided, SMS/Call pattern analysis among many others
- Social Imprint:
- Basic: Type & number of social network affiliation, mode of social media access, type of content seen or posted, No of contacts on social, No of active hours, duration of login, login time stamp
- Deep Social: Analysis of posts seen & posted, No of groups, Broadcasts, type of interactions, emojis used, IP address & type of browser used, likes & dislikes, followers & followings and their history among others
- Psychometrics: For customers filling out certain feedback forms & reviews, or filling in a basic questionnaire for understanding psychographic understanding about the customer
- App usages: A huge way to understand the customer propensity to credit is through their interactions at online & app companies based on order type as well as RFM (Recency, Frequency & Monetary)
- Food delivery
- Retail marketplaces
- Grocery
- OTT platforms
- Real estate
- Travel portals
- Medical & Fitness
- Music apps
- others
Transactional:
Transactional data for customers are accessed in real time through a mix of SMS/ Email based transactions, other payment apps and open banking access
- Cash Data: Numbers of bank accounts, recency & frequency of transactions, UPI usage and digital footprints
- Mobile payments: Payment of different entities through Digital mode
- Utility payments: Electricity, Phone, Gas, Broadband bills among other
- Rental & Government payments: Rent, Municipality bills, Taxes etc
These datapoints are gathered through cookies (2nd / 3rd party), consented tracking, approval for open banking access, email trackers among others. One of the major issue with this data gathering is privacy and with increased GDPR laws in Europe and similar laws elsewhere, collecting these datapoints are becoming difficult and borderline illegal.
Additionally, above datapoints are generally collected in a non-structural datasets and herein comes the expertise of these companies offering alternative scoring models. These companies keep developing algorithms to provide the right insights through the optimal use of AI/ML, and ensure that these datapoints are structured in implementable solutions. For this generally, deep learning programs using complex and hybrid Neural networks are used. (Naik, n.d.)
Process for Alternate Scoring Model:
- System integration and resource availability: Multiple channels on different platforms are connected through intermediary API / calls for real time extraction and actioning on datapoints
- Data collection & analysis: Data is collected from different sources including CDP, DMP, CRM through external & internal medium. This data is generally in unstructured format. Using Natural language processing (NLP) & AI/ML especially deep learning through Neural networks, this data is converted into structural format.
- Business Insights: This structured data is then used for running multiple simulations and persona match. Based on this, a tentative scoring is assigned to the entity.
- Continuous Monitoring: Continuous monitoring is required to ensure avoidance of bias, change in data set and performance of created personas. This helps in avoiding frauds & further refinement in the model. The Alternate scorecard of datapoints (Challenger) are also run against the regular credit models (Champions) to understand the difference and accordingly take corrective actions on datapoints.
Advantages V/s Disadvantages of Alternate Scoring Models:
Advantages:
- Visibility for NTC: Visibility towards otherwise opaque payment potential of New to credit customers
- Recency: Non Historic trends and insights are based on recent activities
- Third Party Validation: Overall Credit worthiness is checked based on multi channel third party information
- Mitigate Bias Possibilities: The information provided is generally from multiple transactional and non transaction data source and hence the opportunity of bias comes down significantly
- Access to underbanked & Unbanked too: This also provide opportunity to tap in unbanked customers as well based on non-bank data points
- Reduces Fraud chances: As customer social and digital footprints across media is difficult to impersonate, it reduces chance for multiple fraudulent activities
Disadvantages:
- Data Privacy, potential implications & future laws: This is one of the most important factor to take care when collecting customer information. This is also significant as most of the 3rd party information is collected without customer explicit consent
- Data Accuracy: As data is collected across multiple channels wherein the same media may be used by multiple persons. It will dilute the accuracy of data. Also many of the variables might remain absent which might have impact on overall scoring model
- Data Size: Data size may be too small for set of customers and mayn’t be sufficient to provide relevant and accurate score
- Scoring Model Biasness: The Scoring model will invariably have certain bias and there are possibilities that for certain pockets of customers, this bias results in abnormal scoring either False positive and True Negative
- Scoring model Flaws: Scoring models may provide for excess weightage to certain segments leading to incorrect scoring
While the above flaws remains, Alternate scoring model can still become the most significant factor in allowing access to credit to a much larger segment, helps in alleviating poverty and improve financial stability.
References:
Naik, P. (n.d.). Alternate data for Credit scoring. Experfy.com. https://talent.experfy.com/hubfs/Pillar%20Page%20Files/Alternative%20Data%20for%20Credit%20Scoring.pdf
World Bank Group. (2022, June 30). The Global Findex Database 2021. World Bank. Retrieved August 8, 2022, from https://www.worldbank.org/en/publication/globalfindex
Shaji, A. M. (2020, July 16). Overview of Alternative Credit Scoring Models. Enterslice. Retrieved August 8, 2022, from https://enterslice.com/learning/alternative-credit-scoring-models/
Alternative Credit Scoring: Banking for the Unbanked. (n.d.). Credolab. Retrieved August 8, 2022, from https://www.credolab.com/blog/alternative-credit-scoring-banking-for-the-unbanked
Biyase, M., & Fisher, B. (2017). Determinants of Access to Formal Credit by the Poor Households. Studia Universitatis Babes-Bolyai Oeconomica, 62(1), 50–60. https://doi.org/10.1515/subboec-2017-0004
Oleksiuk, A. (2022, March 30). Mobile Data + Machine Learning = Better Credit Scoring for the Underbanked. Intellias. Retrieved August 13, 2022, from https://intellias.com/mobile-data-machine-learning-better-credit-scoring-for-the-underbanked/
Agarwal, S., Alok, S., Ghosh, P., & Gupta, S. (2019). Fintech and Credit Scoring for the Millennials: Evidence using Mobile and Social Footprints. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.3507827
