Table of Contents

Introduction..............................................................................................................................1
Product - Munafa:.............................................................................................................................1
Project Scope:....................................................................................................................................1
Small Retailers and Credit Scoring:................................................................................................1
Objectives and deliverables.......................................................................................................2
Objectives:..........................................................................................................................................2
Primary Research......................................................................................................................3
Interviews with Industry Experts....................................................................................................3
Meeting with Finja............................................................................................................................3
Meeting with Jazz Cash....................................................................................................................3
IDG (Inclusive Development Group) findings:...............................................................................4
Kal Pay...............................................................................................................................................4
Field Survey Findings........................................................................................................................4
Findings from Primary Research.....................................................................................................5
Secondary Research:................................................................................................................5
Findings from Secondary Research.................................................................................................6
Variable Analysis:.....................................................................................................................7
Selection:............................................................................................................................................7
Hypothesis Development:.................................................................................................................7
Modelling System-5 Tier Model.............................................................................................12
Tier 1-Deal Breakers.......................................................................................................................12
Tier 2-Data acquisition, process, and pipelines............................................................................13
Tier 3-Verification...........................................................................................................................16
Tier 4- Model Scoring, weightages and interpretation................................................................16
Tier 5-Capacity................................................................................................................................18
Process Flow...........................................................................................................................19
Customer process flow....................................................................................................................19
Muawin’s process flow....................................................................................................................20
Limit Equation........................................................................................................................21
Insights and conclusions........................................................................................................22
Recommendations...................................................................................................................22
Status of deliverables..............................................................................................................24
Introduction
Cashew Financial Services Limited (Muawin), a licensed NBFC by SECP, was founded in
2020, under the leadership of three LUMS graduates, namely Syed Hashim Mahmood Ali, a
BSc Accounting and Finance graduate of 2012; Moeed Qamar, a graduate of BSc Accounting
and Finance 2016 and Naweed Sharif, an MBA 1988 graduate, focuses on empowering small
business to have convenient access to small credit, with their ‘Buy now, Pay Later’ strategy,
aiming to stimulate a faster road toward financial inclusion/empowerment.

Hailing as the first BNPL B2B startup, Muawin is currently offering three digitally operated
products for small retailer, and our project is geared towards the product “Munafa.”

Product - Munafa:

This service offers small retailers easy access to credit for their working capital, in the form
of inventory to help them free up cash and balance their working expenses. Munafa is shariah
compliant, has multiple options (online and offline channels) to pay back loans, has quick
processing time. Good payback behaviour by the small retailers can lead to higher limits and
greater profits.

For the partners, this model serves as higher SKU penetration, customer retention and a
higher sales volume.

The unique selling point that Muawin is leveraging is their shariah compliancy, whereby
retailers would know that the financing they are getting is Islamic. Similarly, Muawin also
aims to seamlessly integrate within the retailers ongoing existing routine, with just a simple
digital loan application, rather than disrupting their business practices, just like the
competitors.

Project Scope:

Our consultancy project is centered around the product ‘Munafa’; to aid the company in fine
tuning their credit scoring model, to better predict the borrowing behaviour of the small
retailer, with a keen eye towards managerial decisions that have to be made to feed the
correct information into the AI/ML models for credit scoring.

In Pakistan, the small retail sector financing is non existent, where most of the transactions
are carried out via cash. This custom further gets trickled down to exploitative lending, as the
usual route of financing for the small retailers. Although it is harsh in nature, this way of
lending is the quickest for the retailers who can not go through the cumbersome paperwork or
financial literacy required to apply for loans through banks. To top if off, the banks take a
month or two in approving the loans with extensive diligence protocols. Under such
circumstances, fintech start ups like Muawin, with their provision of stock helps retailers in
quick financing and also aiding them in rotating the stock to gain profit.

Small Retailers and Credit Scoring:

To minimize the extensive data needed to assess the credit worthiness, Muawin partnered
with major distributors, to gain access to customer insights. With the aim of reducing
documentation or dependence on banks, Muawin was on the right path to make quick

1
decisions on the credit worthiness of retailers, only to face several hurdles from this end of
the supply chain as well. The order bookers from the distributors end, along with Muaein’s
sales representatives had colluded for financial gains and reported KYC’s at times
unethically.

Effective credit scoring hinges on an insight into the financial indicators, but because these
retailers are not very diligent in maintaining financial recording, other variables to look at
creditworthiness had to be evaluated. The consultancy project is therefore focused on
developing a comprehensive approach to evaluating retailer creditworthiness that accounts
for these challenges.

Objectives and deliverables

The core focus of this project at inception was to gauge the credit riskiness of new to credit
customers, whereby, the team had been tasked to just understand the borrower’s willingness
to pay, rather than his/her ability to pay. Upon further discussion, the scope of the project was
expanded to study both the ability as well as willingness to pay, in order to come up with a
holistic credit scoring model.

Objectives:

The project has the following objective:

To devise a credit scoring system that would take into account the right combination of
variables, lowering the risk of default for small retailers.

Initially, the project aimed at developing a Credit Scorecard that would incorporate variables
related to both ability and willingness to pay. The deliverables revolved around extracting
relevant research within the context of Pakistan’s small retail landscape, regarding credit
scoring models. Furthermore, interviews with industry experts as well as companies that have
formulated their own credit scoring models, within the ambit of the fintech landscape were
carried out to better assess the gap Muawin was facing whilst evaluating the credit worthiness
of these retailers. Later on, the scope of the project was altered, with the focus shifting
towards developing a three tiered modeling system that involves identifying specific deal
breakers, data acquisition process, as well as proper verification of the gathered data.

The deliverables for the project were as follows:

1. Development of a simple and easy to understand questionnaire (for our target

customers) and data acquisition methods that would extract extensive information
regarding multiple variables with a limited set of questions.
2. Come up with relevant deal breakers, after shortlisting the variables, to sift out
potential high default retailers in the start
3. To build a limit exposure model that would test the dynamic health of the business
impacting the limit of inventory offered to the retailers. This would make it easy for
Muawin to assess when to offer increased credit limits to retailers.
4. To assign a hypothesized credit score to retailers for each variable that assesses the
probability of default by producing a cumulative score based on managerially
furnished weightages.

2
Primary Research
As per the scope of the project, key insights were sought from conducting interviews with the
industry experts, during the selection of relevant variables. Later on, field surveys further
refined the shortlisted variables by analyzing on ground insights for our target customers.

Interviews with Industry Experts

Extensive interviews were conducted with companies such as Finja, JazzCash, IDG, and
Kalpay, where they shared their insight on the credit lending engines they had designed and
how their operations were working.

Meeting with Finja

A meeting was conducted with Zain Bhatti (Business Head IT-DFI) and Rubab Zahra (Data
Lead at Finja). As per their credit model, they were looking at three things, they were mainly
focusing on KYC and variables in line with the ability to pay. Willingness to pay was gauged
indirectly through the sales representatives.

KYC

For KYC, Finja obtained the retailers CNIC, shope ownership status, location and address of
the shop up until union council, which was then cross verified through the utility bill, their
home address, tax return information and their ECIB status

Ability to Pay

Finja’s approach to assessing the merchants creditworthiness, to look at their ability to pay
included anecdotal evidence of weekly sales, the number of distributors providing them
products and their purchase volume and frequency through these distributors, and estimated
disposable income calculated through margin assessments. Additionally, Finja employs a
rule of thumb, whereby if a shopkeeper generates sales of 1 lakh and maintains consistent
weekly sales at the same level, they should be able to repay a loan within two months.
Verification of purchases with the help of the distributors was also important, because if the
shop owner requires credit, he will be willing to provide distributor details and invoice
verification.

Willingness to Pay

To gauge their willingness, Finja looks at their initial intention for getting the loan, their
historical credit data from sources like ECIB, Tasdeeq, and DataCheck, which play a crucial
role in assessing the borrowers creditworthiness. Additionally the integrity of critical data
reflecting the borrowers ability to repay is also looked into. His willingness to provide the
necessary data is also assessed. Furthermore, the form in which the loan is required, whether
in cash or inventory, is taken into account.

Meeting with Jazz Cash

A meeting was conducted with Adeel Ejaz (Loan Product Head - Jazzcash) and Atta Ur
Rehman (Head of Strategic Projects and Partnerships)

3
Previously Jazz cash had outsourced a credit scoring model for both their consumer and
merchant apps, but recently they have developed their own model for the merchant app

The team gave us a general idea of how they went about developing the model. The ticket
size for lending is really small, and they are highly dependent on the data already available
within their ecosystem (they made use of both GSM and Jazz cash data) some of the variables
highlighted were the end of day balance of the merchant in the last 30 days, average
transactions, how long has the merchant been active. These were some of the variables used
to give out loan to the merchants that have never taken one before.

For those who are taking loan again, on top of the variables already mentioned, they look at
the loan repayment history as well. Classification models used to check the eligibility were
logistic regression, random forest and xgboost classifier

IDG (Inclusive Development Group) findings:

An interview was conducted with the regional officer Hammad Rasool, who oversaw their
regional lending to SMEs.

He suggested cash flow lending as the most feasible method for the underbanked population,
who usually do not own assets or cannot substantiate their ownership through documentation.
Therefore, assessing their ability to pay is critical in understanding their risk. He explained
IDG relies on credibility checks done by third parties who scout and report retailer’s market
reputation. While many of the variables included in their propriety model were similar to
those of finjas, a few stood out such as the legal status of the business, succession planning,
and loan exposure. He suggested that quantifiability of variables is imperative, and not much
should be left to subjective analysis.

Kal Pay

A Meeting was conducted with Sher Shah, the CEO for the fintech startup.

Although working in the consumer lending landscape, which is inherently different to SMEs,
over here as well psychometric data is very hard to determine, because the authenticity of
data is very hard to determine. Cost of customer acquisition exponentially increases if the
psychometric variables are incorporated in the questionnaire, and in fact render psychometric
assessments unviable. Therefore willingness to pay can be assessed using demographic data,
and historical trends against them.A great deal of risk assessment for KalPay is reliant upon
their executives on field, who also look after the verification of data.

Field Survey Findings

After visiting 30 shops across all three categories (Pan, Kiryanan and General Stores), the
following insights were gathered:

Utility bills are not part of existing KYC, although they are readily available. This tells us
that when we get the utility bill, we can get the address of the shop and we can cross verify its
ownership. It also helps rationalize the quantity of chillers used against the bills derived. If
they do not match one can conclude ethical issues on retailer’s end. Furthermore payment
history (financial diligence) in terms of bill payments can also be analyzed.

4
Ownership status of the shop can be verified through the utility bills and rent agreement.
People are willing to provide referrals/guarantors if it were a compulsion for loan
application. They suggested their landlords, committee members and business community
members as referrals/guarantors.

One of the major reasons for payment overdues was excessive credit exposure without due
diligence into the sales potential of the shop. The sales representatives were up and cross
selling excessively. Timely repayment of loan was considered a good qualification to extend
credit limit to a merchant all the while ignoring his capacity to sell. This naturally led to
overstocking, inventory build up and inability to pay back credit.

Another important factor causing delays in payments was the pricing issues. Prices offered by
Muawin for the major key selling items were suddenly higher than competitors.This had
happened post their strategic shift from various distributors to last mile delivery.

Social prestige was critical for merchants and they were very conscious of their image in the
community. This can be leveraged by the recovery team to enable timely repayments. Like all
of its competitors, Muawin has had a strong focus on new customer acquisition, making it a
major KPI for their sales force which inadvertently led to onboarding of less creditworthy
customers which has now taken shape of stuck capital. Authority of the person running the
shop was an often overseen factor in KYC, as sales team would run loan applications for
people not directly involved in running the shop.

Most of the data requirements within our model were easily met, however FRC was a
document most merchants did not have, providing a sales number/turnover was a complete
no. Question placement and phrasing was critical when being posed to the retailer because
people were comfortable in sharing their personal details after they answered questions
regarding their shop, therefore question placement must be done in this order.The invoiced
documentation was not done systematically, and merchants require time to provide us with
ample invoices.

Most of the Merchants were not paying back because of the recent supply denial of cigarettes
from muawin to which most of the merchants had faced significant business disruptions.
They also directed the money meant to be paid back to Muawin to their cash suppliers i.e the
distributors to keep their daily operations running. All the merchants are not financially
diligent and did not plan their credit cycles. They required constant follow ups for efficient
recoveries.

Findings from Primary Research

Replicating on-ground data to a digital data base is imperative, which points at quantifiable
data points and effective data acquisition mechanisms for further data processing.

Verifiability of data is critical, no matter how essential a data point is in evaluating the
riskiness of a merchant. If it cannot be verified, it is only speculative data. For a model to be
effective robust verification protocols must be developed for each data point.

It was highlighted that psychometric variables are the biggest challenge because of data
inaccuracy and credibility and for willingness, the single most important data point comes out
to be if the merchant has taken a loan and repaid it. However variables not directly linked to
willingness could be used to expose a merchant's willingness to pay back the loan.

5
Secondary Research:
The primary aim of secondary research was to investigate the variables widely analyzed
whilst forming credit scoring model for small retailers, for both the willingness and ability
pay. This research was based on comparable countries to Pakistan, that have similar
socioeconomic conditions.

When analyzing variables for ability to pay, as per the paper “two-stage dynamic credit
scoring model”, specific variables such as age, education, marital status, monthly income,
and tenure in job were the basics that were found in all the research papers. These variables
were tackled differently when it came to modelling them for credit scoring. Banking activity
and possession of credit cards did not relate to the financial behaviors prevalent in Pakistan’s
retail landscape.

In contrast the second paper focused on “an analysis of credit ratings for Small and
Medium sized enterprises (SMEs) in Asia”, where the study investigated a myriad of
financial ratios are useful in predicting a firm’s likelihood of default. The financial ratios
analyzed included leverage, liquidity, profitability and coverage. The methodology
employed for these studies was mostly logistic regression, principal component analysis and
cluster analysis.

The research paper from CGAP, a research firm working towards making financial services
to meet the needs of the world’s most vulnerable countries, addressed the operational
intricacies involved in credit scoring project, where the deployment processes usually used
analytical techniques. Data driven methodologies, where data analysis was used to create new
and improved credit scoring was studied. The key variables in studying the ability to pay
encompassed parameters such as average savings balance, number of deposits, psychometric
data and mobile money use data. Predive modeling was done through logistic regression.

Extensive research was carried out to study the use of psychological variables as a
determinant of credit risk assessment. Paper on the “Effect of Psychological Factors on
Credit Risk”, which was conducted in Mongolia, and “Credit Scoring for Small and
Medium Enterprises: A Behavioral Approach, investigated psychological constructs on the
loan repayment behavior. Variables such as effective decision making, self-control,
entrepreneurial self-efficacy, social stigma, locus of control, attitude towards money, were
gauged through questionnaires designed to be answered by the small retailers.
The results were then incorporated into predictive models using logistic regression.

Findings from Secondary Research

The findings from these research papers underscore the importance of approaching the idea
of credit assessment in a multifaced way, by incorporating a diverse range of variables to
gauge the financial stability of small retailers and well as their personality and behaviors. The
review provided us with a foundational understanding in the development of a comprehensive
credit scoring model, by carefully analyzing the specific needs of the target population.

The incorporation of variables such as owning a property and banking activity highlight the
importance of ownership of asset and the banking behavior, however in the Pakistani retail
context, these factors can vary based on the financial illiteracy, and other disparities such as

6
the population being underbanked or unbanked. When it comes to factoring in the
psychometric variables into credit scoring model, it becomes difficult to gauge this segment
of the model because in Pakistan, informal lending practices are prevalent, and their
propensity to lie and get away with it leads to assessing these variables by asking simple
question not plausible.

Variable Analysis:
Selection:

To develop a credit scoring model, factors that are verifiable and that influence the risk level
of applicants were considered. After relevant primary and secondary research, 48 variables
were shortlisted, which were further vetted, during on field interviews, to be reduced to 22.
The 22 independent variables were to be evaluated collectively through questions designed to
reduce the time taken for merchants to fill the questionnaire. Muawin wanted a seamless
application process, as the goal was to reduce the need of a sales representative
compromising the experience with his own self interest in mind.

These were shortlisted because of constraints in acquisition and verifiability. A model with
variables that cannot be substantiated with quantitative and verifiable data is less practical
and more academic in nature.

Hypothesis Development:

The rationale behind shortlisting each variable has been given below:

1. Gender:

Hypothesis: Male borrowers are at a lesser risk of default as compared to female borrowers.

Primary research highlighted that male borrower exhibited better financial diligence,
compared to females. This could be attributed to factors such as males calling the shots in the
Pakistani context.

Variables Options Scores Weightages

Male 1
Gender 2
Female 0

2. Age

Hypothesis: There exists a negative relationship between increasing age and credit default.

Although the experience in financial management increases with age It was also found out,
that the age bracket of 25-50 year old retailers, are the most financially diligent and are
deeply involved in the business with a better understanding of financial management so
there’s a better chance for them to service their debt . However, people above 50 losing the
will to work with changing priorities, but later on, during on-field visits, albeit limited, gave
us an opposing outlook.

Variables Options Scores Weightages

Age <18 0 4

7
 18-25 1
25-50 3
>50 2

3. Marital status

Hypothesis: The risk of default decreases if a retailer is married.

Apart from navigating the process of managing the business, being married adds another
level of responsibility on the retailer. If a client is married, it means that they have stakes
involved and are less likely to do something which might cause them trouble.

Variables Options Scores Weightages

Married 1
Marital Status 4
Unmarried 0

4. Location of Home and Shop address

Hypothesis: A merchant’s home being of the same locality as his shop provides an added
security in terms of trackability, which lowers the risk of default.

If a merchant lives in the same city as his shop, he is more easily trackable if he goes
overdue. Furthermore, he will also remain more vigilant in making timely payments as he
does not have a home far way to flee as opposed to a person who may default and run away.

Variables Options Scores Weightages

Same city 1
Location of Home and Shop address 4
Different city 0

5. No of dependents at home

Hypothesis: More dependents would increase the expenses for the retailer, resulting in a
higher risk of default.

As the number of dependents for the retailer increases, their financial obligations increases.
When it comes to managing their education, health care and daily living costs, inadvertently
straining their budget and resources. This could reduce the retailer’s ability to meet their loan
obligations promptly unless there are multiple sources of income streams for the retailer.

8
 6. Ownership of the shop

Hypothesis: An owned shop would identify greater stake in the business, resulting in lower
risk of default

If the retailer hasn’t rented the property, rather owns it, shows greater commitment to the
business’s long term viability. Ownership also provides control over the premise, giving the
retailer the autonomy to make decision regarding the shop. This greater control over the
operating environment would reduce the likelihood of default.

Variables Options Scores Weightages

Kiryana 1
Type of shop Pan Shop 0 11
General Store 2

7. Tenure in business

Hypothesis: A longer tenure would amount to lower risk of default.

With increased experience, comes an established customer base providing a consistent source
of revenue. With a steady stream of income, the retailers are better equipped to meet their
financial obligations.

Variables Options Scores Weightages

< 1 year 0
1-5 years 1
Tenure in business 6
5-10 years 2
>10 years 3

8. Ownership status of godown:

Hypothesis: Ownership would amount to cost stability leading to lower risk of default

Godown ownership
shows Variables Options Scores Weightages capital
No dependents 1
1-3 3
No of dependents at home 6
4-6 2
>6 0

9
investment, that indicates the retailers commitment to the business. This also allows for better
inventory management, optimizing operations for the retailer.

Variables Options Scores Weightages

Owned 1
Ownership of the shop 4
Rented 0

9. Usage of POS

Hypothesis: Utilization of POS results in a more streamlined payment process, mitigating

operational risk and lowering the risk of default.

Existence of point of sale also indicates technology acceptance, transparency, and flexibility
on part of the merchant. This indicates better financial management because of the
automation.

Variables Options Scores Weightages

Uses POS 1
Usage of POS 5
Does not use POS 0

10. Usage of Inventory Management System

Hypothesis: Businesses that utilize Inventory Management Systems (IMS) exhibit higher
control over their operations, leading to a lower risk of default.

IMS indicates a higher control over the business improved inventory accuracy, reduced
stockouts, and enhanced cost-efficiency

Variables Options Scores Weightages

Uses IMS 1
Usage of Inventory
Does not use 2
Management System 0
IMS

11. Bank Account Ownership

Hypothesis: Bank account ownership shows financial stability leading to lower risk of default

10
Businesses or individuals with bank account ownership are likely to have established
financial relationships with banking institutions, which may indicate financial stability and
responsibility. Ownership may be taken as a positive indicator of creditworthiness, as it
demonstrates the ability to access financial services and manage financial transactions
effectively.

Variables Options Scores Weightages

Yes 1
Bank Account Ownership 7
No
0

12. Source of Income

Hypothesis: Multiple sources of income would amount to consistent income streams lowering
the risk of default.

Diversified sources of income may mitigate the risk of default by reducing dependence on a
single revenue stream. Businesses or individuals with multiple income sources are better
equipped to withstand fluctuations in one source of income, enhancing financial resilience.

Variables Options Scores Weightages

only 1 0
Source of Income 5
>1 1

13. Type of Shop:

Hypothesis: Shops within our category definition have inherent risks. Owing to their business
size and practices, each category of merchant should have a different risk profile.

We believe each category of shops shall be ranked differently in terms of their score because
their business size and dynamics cause them to develop different attitudes to business.
Financial diligence for example is the lowest among pan shops and highest in general store
merchants.

Variables Options Scores Weightages

Kiryana 1
Type of shop Pan Shop 0 11
General Store 2

14. ECIB Status:

Hypothesis: Increased delay would increase the risk of default

11
A positive ECIB status, indicating timely payments would lower the risk of default, also
providing reassurance to lenders. Retailers who have not delayed on their payment would be
given the highest score. Those retailers, that have deferred on their payments would be given
a lower score as per the increase in the number of days.

Variables Options Scores Weightages

No Default 3
No Data 2
ECIB Delayed 30-60 days 2 33
Delayed 61-90 days 1
Delayed 90+ days 0

Modelling System-5 Tier Model

Figure 1: 5 Tier Model

The modelling system we propose is a 5 tier process which allows Muawin to

comprehensively review a potential customer. The process begins with deciding if the
customer is credit worthy or not. This decision will be based on several critical factors that
we believe make a customer fall above a risk threshold that Muawin should operate on.

12
Tier 1-Deal Breakers

It is also important to identify that at the point of data acquisition, some findings will be
categorized as deal breakers because of which Muawin will not service the retail customer.
These will be regulatory requirements that must be followed. Deal breakers will include:

If CNIC is not submitted

If the new customer has defaulted on credit before, reflected in his ECIB report.

If electricity bill is not paid, not submitted.

Any unethical practice or malpractice caught which questions behavior and morality

Is the loan applicant incharge of making and implementing financial and

administrative decisions?

Tier 2-Data acquisition, process, and pipelines

If an applicant clears the tier 1 of the modelling system, the next step is to collect data points
for each of the variables mentioned before. The data acquisition strategy to collect this is such
that it shall have minimal reliance on manual acquisition avoiding any risks associated with
the representative’s gathering data. For this data we propose the following pipelines and
operating methods.

Data pipelines

Figure 2: Data acquisition pipeline

13
Data acquisition through pictures

In order to minimize the role of representatives for data gathering we propose that merchants
take pictures of the following themselves and upload on Muawin’s platform. An OCR
developed will aid in extracting the data required from the pictures sent by the merchant.

Fridge: To rationalize the utility bill and visually estimate the merchant’s capacity for the
required inventory on credit.

Aisles: For product assortment and aid to internal verification team to cross check provided
information.

Shop Legal Documents: For ownership status which is reflected in the scoring model. Also,
to verify that the person in authority is also the owner or has rental agreements in his name.

CNIC: Primarily for KYCs.

Utility Bills: To check for ethical consumption of utility, consumer ID and cross check
ownership information provided.

14
 Figure 3: LESCO Bill

15
 Figure 4: Ownership verification from LESCO website

Figure 5: Payment history verified from LESCO Bill

Data acquisition through questionnaire

The formulation of the questionnaire covering all important variables, the process of
shortlisting them and assigning weightages is comprehensively laid out earlier. This
questionnaire is filled out via Muawin’s app to reduce dependency on the sales team and
avoid possible manual discrepancies. Also, Muawin aims to shift focus from khokas and
small mom-and-pop stores to LMTs and marts in future. This allows Muawin to operationally
make data acquisition relatively seamless by expecting greater digital exposure and financial
diligence from the retailers. This financial diligence is key for Muawin and retailers to
mutually benefit and also allows Muawin to assess not just the ability of the retailer to pay
back but guage their willingness as well.

16
Based on our market research, retailers are willing to provide all the information covered in
the final streamlined questionnaire.

Figure 6:Questionnaire

Tier 3-Verification

Key role of the Verification team is to cross check the acquired data. Any intentional
misinformation or information gaps are highlighted in this process which also provide
insights into the retailer’s borrowing behaviours and provide a closer understanding of their
willingness to payback. Monthly cross checks with and without the retailer’s knowledge will
be conducted.

The Verification process we propose involve 5 following sources:

NADRA for cross checking information such as home address and family certificate to cross
check no. of dependents at home. Shop Agreement/Ownership documents to check if the
shop is rented or owned, check if the person in authority is the owner or has rental
agreements of the shop in his name. LESCO (consumer ID) to cross check utility bills.
Godown Agreements and Ownership documents and visits to check inventory
management systems. This ensures check on operational efficiency, visual examination of
piled inventory and assessments on capacity.

Verification done through govt. provided portal to licensed firms to check relevant bank
details such as ownership of an account.

Tier 4- Model Scoring, weightages and interpretation

Modelling

17
A credit scoring model was to be developed to evaluate risk for small retailers, a category
rich with various types of retailers. For the scope of this model, we narrowed the retailers to
only pan shops, general stores and kiryana stores as these were customers that are the most
feasible target market.

To model risk, we researched various new emerging market players in Bangladesh, Mongolia
and Italy to better understand which key variables and data points were being used and how
they could be implemented in our context. A sheer lack of structured business processes and
data points has become a critical problem in implementing successful models in our context.

Primary research and expert interviews proved to be more beneficial when it came to
understanding risk for small retailers. Various forms of KYC and variables to expose
riskiness were gathered. However, each had a basic undertone where KYC was intended to
simply transfer customer data to a virtual realm and riskiness being evaluated as the ability to
payback loans (financial). Willingness was a constantly overseen aspect of risk but with good
reason. For willingness to pay back loans, experts mentioned several reservations that
included issues with data integrity and credibility along with high acquisition costs. Any
suggestions aimed at implementing gamified surveys or psychometric analysis led to issues
with financial viability and verifiability.

It was only inevitable to develop a model with greater managerial judgment. We went onto
consider variables that could expose willingness to attain, and payback loans which were
scarcely seen in models exercised around. For instance, the ability to furnish documents was
also a clear signal to the merchant showing great willingness to attain a loan, become part of
the credit ecosystem and invariably pay back loans to maintain their credit status. The
variables are further looked upon in detail in the following section which talks about each
variable’s rationale and how it exposes risk to a merchant.

Scoring:

A scoring system for each variable where scoring created a ranking among each possible
answer for a variable. The ranking created a distinction between each answer therefore
ranking the riskiness for each. Following exhibit shows scoring.

Weightages:

Hypothesised weightages have been assigned to each variable as per their importance in
terms of exposing the riskiness of the loan applicant. Variables like ECIB delay status comes
out to be one of the most critical variables in exposing both the willingness and ability to pay
for a loan applicant therefore has the highest weightage.

Following diagrams details each variable, its total score and associated weightages within the
model:

S/No Weightage Variables Total Score Saleem’s attained score

1 0.02 Gender 1 1
2 0.04 Age 3 3
3 0.04 Marital Status 1 1
4 0.05 Utility Bill Payments 1 1

18
 5 0.04 Location of Home and Shop Address 1 1
6 0.06 No. of dependents 3 1
7 0.04 Ownership of the Shop 1 0
8 0.06 Tenure 3 1
9 0.02 Ownership status of godown 2 0
10 0.05 POS 1 0
11 0.02 IMS 1 0
12 0.07 Bank Account 1 0
13 0.05 Sources of Income 1 0
14 0.11 Type of Shop 2 1
15 0.33 ECIB 3 2
Total = 1 Attained Score 0.505
Function for calculating the score = (Attained Score/Total Score) *weightage
Table 1: Model- weightages and score

Tier 5-Capacity

Capacity Model

When a new customer comes in to work with Muawin, the a key component is to figure out
the capacity of the retailer to sell the inventory they intend to purchase on credit. This allows
Muawin to reduce the risk of default and get their payments back in due time. First step in the
capacity model is to translate the credit scores into capacity limiter percentages.

From the credit scoring model, if a customer scores below 0.55 Muawin will not lend to this
customer. For scores above 0.55 respective capacity limiter percentage will be added to the
capacity model.

Lending status Conditions Capacity Limiter

Do not Lend Score<0.55 0%
Lend Score 0.56-0.60 30%
Lend Score 0.61-0.70 40%
Lend Score 0.71-0.75 50%
Lend Score 0.76-0.85 70%
Lend Score 0.86+ 80%

Limit = Moving Average x Capacity Limiter

Table 2: Capacity Limiter

Lets assume a customer’s credit score is 0.58, the corresponding percentage limit is 30%.
From the initial credit of PKR 5000, month 0, the retailer makes sales of PKR 50,000. For the
first month therefore, the credit limit that Muawin provides to the retail customer is 30% of
PKR 50,000, which is PKR15,000.

19
 Months Historical / actual sales Moving average Credit Limit for the month %
Dec 50,000 5000* -
Jan 60,000 55,000 15,000 30
Feb 70,000 65,000 19,250 35
Mar 90,000 80,000 26,000 40
Apr 105,000 97,500 36,000 45
May 115,000 110,000 53,625 55

Table 3: Capacity Model- moving average

If the customers continue to make timely payments, they will be incentivized by a 5%

increase in their capacity limiter percentage. That is, in the second month the customer will
get a 35% of sales.

Credit limit for the second month will be determined using the moving average model.
Moving average will be used account for seasonality. That is, for the month of Feb, referring
to figure x, the customer will be given 35% (5% increment assuming payment was on time)
of PKR 55,000 instead of PKR 60,000. This provides a cushion to the retail customer in
months that his sales are lower than the preceding month. A gradual increase of percentage
offered incentivizes the customer to continue to do better and allows Muawin to reduce their
risk. This allows Muawin for behavioral assessment of following procedures over time to
determine the merchants willingness to pay back as well.

With this model, merchants are to be incentivised to keep track of their invoices of the
required stock for credit limit enhancement. This will help in creating financial diligence in
retailers where data acquisition will become easier overtime.

Incase they donot make timely in any month, as discussed earlier, their incentives will freeze.

Payment Delays

In case of delay in payment by the retailer to Muawin, the retailer forgoes all his assigned
credit limit. This limit is built by a simple increment of 5% of limit after every timely
payment. This “built” limit is forgone and he starts over post repayment, at he limits he was
assigned at his loan application.

Another important caveat to this mechanism is a possible reduction in his score for ECIB if
his payment delay exceeds his last delay reported on ECIB. Now his score will be reassigned
as per his latest ECIB delay status, following a new limit assignment as per the new total
credit score.

20
Process Flow
Customer process flow

The merchant’s process flow is kept as simple and systemic as possible keeping in view
customer psychology and the nature of their day-to-day work. Figure z shows the process
Muawin requires the customer to follow:

Figure 7: Customer’s process flow

The merchant will first approach Muawin and apply for a loan on the Muawin’s app. After
applying, he will be required to upload 4 images as part of Tier 1 clearance. Merchant is
required to place CNIC close to his face and take a picture for information on KYCs. Picture
of Rental or ownership agreement, Lesco bill and chillers in the shop are required. These 4
images are used my Muawin to extract relevant data mentioned in Muawin’s work flow
ahead.

For Tier 2 clearance, the merchant is required to fill the questionnaire prepared on the app.
This questionnaire covers all the variables for which data is required. This is essentially the
main data acquisition stage of the customer

There will be no information required by the customer at Tier 3 of the process flow. At this
stage, Muawin’s internal team will be working on the acquired data, details of which will be
covered later under Muawin’s process flow.

Lastly, as part of Tier 4, Retailers are required to document and submit 15 day sales invoices
of the inventory bought on credit from Muawin. These invoices will be used by Muawin for
capacity assessment explained further.

Tier 5 also doesnot require any submissions from the customers.

Muawin’s process flow

While the customer’s submit their required data, Muawin’s internal team extracts information
at each tier and processes it per their model. Figure a the process flow of Muawin spanned
over 5 tiers.

21
 Figure 8: Muawin’s process flow

At tier 1, as the customer submit the required document, Muawin will extract the data
required. CNIC will be used for KYCs such as name, gender, age and home adress to guage
riskiness assessed and marked by area. Picture of Rental or Owner ship Agreement is
required which, together with CNIC, is used to identify if the person in authority is either the
owner or has the rental agreement on his name. Lesco Bill picture is required for Consumer
ID and check if the merchant has previously defaulted. Lastly, an image of Chillers at the
shop to guage any ethical malpractices. This is done by cross checking his Utility bill with
the chiller space.

Once the customer has filled the questionnaire, in Tier 2, the internal team will extract data
data and translate on to the model respective to each variable. At this stage data acquisition
for the scoring model is complete. Secondly, the verfification team begins cross checking the
data acquired so far in both questionnaire and the 4 pictures before.

Once the data is organized in the model, tier 3 requires Muawin to assign scores per variable
and generate a credit score for the customer.

If the score is below 0.55 the customer will not clear Tier 3 and Muawin will not service that
particular customer. If the score is above 0.55, customer will clear Tier 3 and and initial loan
of PKR 5000 will be provided in Tier 4 . With this loan, customer will be called and
explained invoice building activity.

Once the customer has submited 15 day sales invoices of relavant products bought on credit
from Muawin in Tier 4, the internal team will estimate 30 day capacity which will be plugged
into the moving averages model as explained in the capacity model above in Tier 5. Score
limiter rules will be applied and loan for the next month will be extended to the customer per
the limit equation explained ahead.

Limit Equation
With the help of initial primary and secondary research coupled with market surveys and
interviews with industry experts, an equation can be derived which shows the relationship
between 3 key components of the model:

Limit (y) of the credit provided by Muawin to a retail customer.

The Credit score (cr) assigned based on findings from the questionnaire and data acquired.

22
Assessment of the Capacity (ca) of a retail customer to sell inventory bought on credit and
consequently pay back to Muawin.

Hence the relationship that takes shape is:

y=f (Cr ,Ca)

That is, The limit of credit provided by Muawin is a function of the credit score assigned to
the retail customer and his capacity to sell inventory bought on credit by Muawin assessed by
the capacity model.

Insights and conclusions

Developing a credit scoring system for this underserved and underbanked business
community especially in terms of credit instruments, has been an arduous task. While we
have aimed to encompass the credit worthiness of the small retail category within this model,
any model can only capture their riskiness to a limit aways leaving room for uncertainty of
default. This is primarily due to the enigmatic behavioural aspects within the broader cultural
construct.

Many FinTech’s have done extensive work to develop credit lending engines, models, and
systems, yet a unanimous agreement remains among them all; there will always remain room
for improvement. After having done detailed research our credit lending system is another
step towards refining the risk evaluation mechanisms at place. Following is a brief rundown
on the salient features of our system:

 An Enhanced KYC that includes a two-dimensional verification of ownership/rental

status of the shop and loan applicant through rental agreement while verifying it with
their utility bills. It further exposes their Authority level at the shop while assuming
rental deeds are made in the name of the person in authority at the shop. Previously,
models would often oversee authority levels at the shop by running KYC of any
person present at the shop claiming ownership. This would bare them of creating the
most financially diligent person a loan applicant, which is the actual person running
the shop.
 Dynamic Business Health Monitoring through accurate capacity of sale surveillance
via invoices submitted. This has a two-fold effect, firstly it limits over exposure of
credit by limiting inventory orders to their actual sales capabilities and prevents over
stocking at the merchant end.
 Credit Limits assigned on the bases of credit score and capacity of sales both.
Rationalizing their credit needs and worthiness both become a integral part of this
system, where each score bracket provides a limit of credit to be provided on their
capacity to sell inventory.
 Reduced OPEX by integrating digital data acquisition as opposed to previously
practised manual data collection done by an active sales force.
 Removing Risks associated with an active salesforce such a collusion among loan
applicants and the sales force and unethical practices.
 Stringent SOP’s (Tier1) right at the start of the loan application to avoid excessively
risky applicants to be screened out without further processing their data.

23
Recommendations
Whatsapp API Integration

Muawin’s target market is not technologically savvy, with most under the bracket not even
owning a POS system for processing their sales transactions. Filling up a questionnaire on an
app can be a daunting activity if the customer is not proficient at navigating through the app,
with many not reaching completion.

By integrating the whole questionnaire into whatsapp will enable a smoother customer
journey. Being used to the whatsapp interface, it would be easier for the retailers to answer
the questions and allow for a simplified interaction, and instant communication with the
customer support team. Important notifications can also be directly sent to the retailer’s
whatsapp, ensuring easy engagement and response time.

To further automate the process, automated responses could also be put into place for
frequently asked queries, and this would reduce dependence on the support team further.

J-Score

It was analysed through research that the incorporation of psychometric variables into the
credit scoring model keeping the Pakistani landscape in mind would be a difficult feat to
carry, however one thing that can be done is to get data from third party service providers
such as jazz, that have a proprietary system in place to check for behavioural tendencies of
the retailers. They do not document intensive data for verification of variables that expose
willingness and ability to pay, in fact use analytics from their existing ecosystem that
includes the sim, jazz cash and jazz world. Data from their activities and financial behaviour
is used to predict risk. An example is their financial decisions on jazz cash loan repayments,
another is their behaviours while using the SIM is a strong predictor of future actions.

These behavioural variables can be given a score and be incorporated into the model.

Remote On-boarding teams

Muawin was trying to decrease the dependence on their sales team for onboarding customers
because of their self interest involved in the process. Apart from the customer support team to
deal with any complaints, a dedicated team should be in place to guide the customers through
the process.

To Replace salesman with telephonic on-boarding teams which provide customer service and
guides with the SOPS Invoice building activity. This helps in minimising any risks associated
to the on field personal.

Collateralising Initial Loan Amount

Cash flow-based lending for this segment will remain a higher risk as opposed to collateral
lending. No matter how effective a KYC or predictive modelling tool maybe the relationship
among the lender and the business is not that of a high dependence as defaulting or delaying
payments will barely stop the enterprise from running. This naturally gives rise to the free
rider problem where the merchants will show great willingness to become a part of the loan
application to attain the initial PKR 5000 loan amount as it requires low efforts. The strategic

24
intent of Muawin is to onboard a merchant on their earliest to begin building their credit
history, which is why a small loan amount is proposed after the loan application is complete.
In the absence of the ability to do due diligence like banks; to avoid initial bad debts a system
to create small collaterals for the initial loan amount can be made. This maybe a digital
agreement to collateralise existing inventory at the shop.

Status of deliverables

STATUS OF DELIVERABLES
Months Weeks Current Changes

Secondary Research Primary and Secondary Research on

Week 3
both ability & willingness to pay
January Interviews (Willingness)
Variables Selection (Willingness) Variable Selection (ability &
Week 4
Questionnaire Development willingness)
Surveying current and new to credit
Week 1 customers Questionnaire Development
Data Collection
Week 2 Testing and Modelling Questionnaire Optimization
Surveying new to credit customers &
February Week 3 Scorecard Development
recalibrating questionnaire
Translate the psychometric variables into
Consolidating findings from survey
quantifiable credit scoring criteria
Week 4 Develop a credit scoring model Created data acquisition pipelines
Test the scorecard for internal consistency Developed procedures for data
and reliability verification
Week 1 Implement score card and modelling Client's focus changed
March Evaluate the model from results and make Developed a 5-tier credit scoring
Week 2
changes system model

Figure 9: Status of Deliverables

Project TORs and expectations have changed twice in the duration of the project. In the first
half, initial TOR covered the willingness of retail customers to pay back. It was a
comprehensive study of psychometric and behavioral study of small retail customers to
provide Muawin a framework/model. This model was intended for Muawin to onboard new
to credit customers and assess their risk of default in a case where past data was unavailable.

However as secondary research at this stage progressed, it was deemed more value adding to
create a comprehensive model which not only includes willingness to pay but also hard data

that can gauge the ability of a retail customer to pay back as well. It was realized that
willingness and ability are related where one can give valuable insights about the other.

So the scope of the project broadened. Till this understanding, the aim was to create a model
which Muawin will compare to their current one to check if components can be picked and
plugged in their current system. Data to run the model was either to be acquired by a team

25
from market surveys or given by Muawin so that the models can be compared using the same
customer data.

Over the span of the project, key discussions of data acquisition, data pipelines and
verifiability came in. I was realized that the operational end of creating the data base in along
with running the model with that data was equaly important if not more.

So the second change to the scope took place in the second half of the project where, after
discussions with CEO and founder, Hashim and head of growth team, Zeeshan, that it is more
value adding to Muawin to better their processes of data acquisition, data pipelines and
verifiability. It was also brought to light that Muawin aims to shift focus from small retailers
at khokhas and small pop and pop stores to LMTs and marts in future. This changed the
behavioral premises of the retailers we were initially working with as now more financial
diligence and digital understanding can be expected. This led to conversations on
streamlining data acquisition from manual sales team to digitizing processes in future.

Secondly it was highlighted that due to differences in the Muawins current model and the
model that we are proposing the same data cannot be used. If we acquire data from Muwain’s
of their current customers some data points will be missing as we propose a few additional
variables in the model. Using a mock data will reduce the exercise to merely an academic
one. Therefore, the second change in the scope is that more focus will be laid on the
operational of new ti credit customer acquisition and how that process can be improved.
Running the model and cross comparison is now a secondary deliverable.

26
 Exhibits

27
 Exhibit 1: Credit Scoring Model

Exhibit 2: Market Surveys

Picture of shelves Picture of CNIC

Picture of Chillers Picture of Exterior

28
 References:

Mavri, M., Angelis, V., Ioannou, G., Gaki, Ε., & Koufodontis, I. (2008). A two-stage

dynamic credit scoring model, based on customers’ profile and time horizon. Journal

of Financial Services Marketing, 13(1), 17–27. https://doi.org/10.1057/fsm.2008.2

Yoshino, N., & Taghizadeh–Hesary, F. (2015). Analysis of Credit Ratings for Small and

Medium-Sized Enterprises: Evidence from Asia. Asian Development Review, 32(2),

18–37. https://doi.org/10.1162/adev_a_00050

Ganbat, M., Batbaatar, E., Bazarragchaa, G., Ider, T., Gantumur, E., Dashkhorol, L.,

Altantsatsralt, K., Nemekh, M., Dashdondog, E., & Namsrai, O. (2021). Effect of

psychological factors on credit risk: a case study of the microlending service in

Mongolia. Behavioral Sciences, 11(4), 47. https://doi.org/10.3390/bs11040047

Goel, A., & Rastogi, S. (2021). Credit scoring of small and medium enterprises: a
behavioural approach. Journal of Entrepreneurship in Emerging Economies.
https://doi.org/10.1108/jeee-03-2021-0093
‌Roy, P. K., & Shaw, K. (2021). A multicriteria credit scoring model for SMEs
using hybrid BWM and TOPSIS. Financial Innovation, 7(1).
https://doi.org/10.1186/s40854-021-00295-5

29