Small Businesses and Digital Financial Services Predictive Modelling and Segmentation for Market Sizing and Product Design By Sinja Buri, Morne van der Westhuizen, Soren Heitmann Table of Contents Executive Summary 4 Introduction 6 Data 7 Methods 8 Supervised Segmentation 9 MSME Clustering using Survey Data 10 Predictive Modelling for MSME Identification & Behavioral Clustering of MSMEs 11 Results 13 Qualitative MSME Segmentation using Survey Data 13 Non-Agent Entrepreneurs 14 Savvy Mediums 15 Basic Smalls 16 Mobile Money Agents 17 Transaction behavior of qualitative MSME segments 18 Behavioral quantitative MSME Clustering using Mobile Money Transaction data 20 Results of the Supervised Segmentation 20 Results from the Clustering of Predicted MSMEs 21 Implications 24 Conclusion 25 References 26 Figures 1. Overview of Research Phases 8 2. Sankey Diagram - Overlap of Segments based on Survey Data (Qualitative Segmentation) and 10 Segments from the Supervised Segmentation 3. Logistic Regression – Receiving Operator Characteristic (ROC) Curve 11 4. Logistic Regression: Probability to be a business based on the Average Number of Trading Days 12 5. Distribution of interviewed MSMEs across Clusters 13 6. Transaction Activity of MSME Segments identified based on Survey Data 18 7. Distribution of identified MSMEs across Clusters 21 8. Clusters of predictive modelling – Average Revenue Per User in USD 22 9. Clusters of predictive modelling - Average Monthly Account Balance in USD 23 2 Tables 1. Business categories based on numbers of employees 7 2. Definition of Segments resulting from Supervised Segmentation 9 3. Mobile Money Transaction Profile of Non-Agent Entrepreneurs 14 4. Mobile Money Transaction Profile of Savvy Mediums 15 5. Mobile Money Transaction Profile of Basic Smalls 16 6. Mobile Money Transaction Profile of Mobile Money Agents 17 7. Transaction characteristics of MSMEs identified through Supervised Segmentation 20 8. Transaction characteristics of Clusters of MSMEs identified through Predictive Modelling 21 Boxes 1. Key results 5 2. Key insights - Predictive modelling and identification of MSMEs 12 3. Key insights - Qualitative MSME segmentation 19 4. Key insights - Supervised segmentation 20 5. Key insights - Clustering of predicted MSMEs 23 3 Executive Summary Micro, Small, and Medium-Sized Enterprises are the Key results across research components are as backbone of vibrant and dynamic economies. But they are follows: sometimes hard for financial institutions to identify because of the methods they use to conduct their transactions. As Supervised segmentation – An initial analysis of mobile a result, many MSMEs do not get access to financing and money transactions identified key segments of highly financial products that are designed specifically to support active and likely businesses. A large number of those businesses. Identifying these MSMEs and addressing their 11,500 potential MSMEs are informal channel workers needs can be very advantageous for digital financial service who are operating businesses using their individual mobile providers. This report discusses predictive data models to money subscriptions. This provides an opportunity for help a mobile network operator in Sub-Saharan Africa1, DFS providers to leverage the networks of these informal identify MSMEs in its market and better understand how to channel workers and to register them as formal agents and serve them. merchants. They are already behaving as such. The MNO has a large market share in the country and Qualitative MSME segmentation (using survey data) – tens of millions of transactions pass its digital financial A survey of 1,275 small businesses that use the partner services channel each month. This report examines those MNO’s mobile money service improved the understanding transactions to determine how many are made by individual of different MSME groups and answered questions consumers and how many are made by entrepreneurs and about the characteristics, needs, attitudes towards, and business owners who use personal accounts to conduct perceptions of mobile money of MSMEs in the country. business. The report postulates that a significant number of These businesses can be segmented into four clusters: Non- MSME owners conduct commercial transactions through Agent Entrepreneurs, Savvy Mediums, Basic Small, and their personal accounts and are therefore not being Mobile Money Agents. These groups differ in business size, identified as business customers and are not being afforded level of technological sophistication, and level of financial the benefits of business customers. inclusion. Their mobile money transaction characteristics mirror the qualitative profiles that emerge from survey data. This research shows that MSMEs can be accurately identified Based on these findings, the paper discusses strategies to with a high degree of statistical confidence. Moreover, the best engage with respective segments. analytic method can be used to segment those MSMEs into more granular business profiles. The segmentation Predictive modelling, identification and behavioral algorithm is driven by patterns of how MSMEs use mobile clustering of identified MSMEs - IFC developed a money. The emerging segments differ in their business robust model for MSME identification from mobile money characteristics and their financial needs. transaction behavior. It has been shown to be 98 percent accurate in identifying which mobile money subscribers Multiple research components generated comprehensive are MSMEs. It identifies 32,600 MSMEs among the MNO’s insights into the MSME segment in the study country. Apart customer base, which represents an important proportion from analyzing mobile money usage patterns, the team also of heretofore unidentified commercial activity segment conducted a survey with 1,275 MSMEs. The survey data was on the DFS channel. Some of them overlap with the 11,500 used to inform the development of an MSME identification potential MSMEs that were segmented in the initial model and to study and profile businesses. supervised segmentation. The predictive model provides a low-cost and fast approach to identifying and monitoring large numbers of MSMEs on a regular basis. 1 Operator and country were anonymized for confidentiality. 4 Identified businesses were further clustered into meaningful To engage with small businesses as customers, it is essential sub-segments based on mobile money transaction features. for DFS providers to be able to identify MSMEs, to know All segments show high levels of mobile money transaction them well and to understand their financial challenges. activity. Above all, Acceptors, Airtime Traders, Service Leveraging combined results from this research supports Providers (see table 2 for definitions) are most active in terms the MNO’s efforts to develop DFS products that are better of transaction count, transaction value, average transaction tailored to the specific needs of MSMEs. Information and amounts, as well as the average number of parties that they results can be used for product development and subsequent transact with on a regular basis. Among MSMEs identified, specific targeting of identified MSME segments. the average revenue per month is approximately 9 USD - this is more than five times higher than the average revenue from normal individual consumers. Meanwhile, for net balances held on the wallet, MSMEs hold an average of 102 USD, as compared to around 30 USD for individuals. Key Results • A significant number of high-value customers on the digital channel are formal and informal businesses that transact using consumer-oriented products. • MSMEs with an individual mobile money subscription can be identified based on their mobile money transaction behavior. IFC developed a robust model that achieves 98 percent accuracy in predicting and identifying which mobile money subscribers are MSMEs. • Data-driven modelling and segmentation can identify MSMEs on consumer channels and sub-segment them further into micro-, small-, and medium-sized tiers. • Profiles and patterns emerge that help to sub-segment MSMEs based on their usage of mobile money, their business characteristics, financial needs, and current use of formal banking services. • Identification and segmentation of businesses that use mobile money services provides valuable information for product design and targeted marketing. 5 Introduction Micro, Small, and Medium-Sized Enterprises are important The core questions that this research collaboration job creators that help advance financial inclusion in Sub- set out to answer are: Saharan Africa, a region expected to add 18 million people to the labor force every year until 20352. However, MSMEs still (a) Can businesses (MSMEs) be identified and segmented in lack adequate access to both financial capital and business the customer base? tools that could help them to prosper and grow. Supporting and engaging them as a customer segment offers a good Can businesses (b) be further sub-segmented into opportunity for digital financial service providers that are MSME tiers? able to tailor services and products specifically for MSMEs. (c) What is the profile/are the patterns of mobile money The value of using data to transform insights into action and usage of those MSMEs segments? to advance financial inclusion has been well documented by organizations across Africa3. With the availability of The research project included a comprehensive survey lower cost computational resources and data applications, of MSMEs who use mobile money. The sample size and companies have been able to sort their customers into geographical distribution were made after preliminary niches and segments, enabling them to better understand analyses of the MNO’s transactional data. Survey results the individual needs of customers. That has made it easier provided ‘ground truth’ data used for developing a model to to manage and deploy strategies that are tailored to specific identify MSMEs based on their mobile money transaction categories of customers. patterns. Survey data also helped to further segment MSMEs into behavioral, geographic, and functional sub- For DFS providers, segmenting the client base opens segments as well as to validate assumptions and hypotheses possibilities for targeted product design, marketing, used in algorithms. and pricing that better meets the respective needs of businesses. Moreover, better serving these segments will drive financial inclusion. Businesses are regular and very active users of mobile money services. Reaching out to MSMEs and offering them mobile money products tailored to their needs has the likely additional benefit of broadening the service provider’s proportion of high-value customers. Given the lack of business-specific digital financial services that are offered, MSMEs and individual customers often use the very same services for business and personal needs. That means that many person-to-person (P2P) transactions may in fact be business transactions. Despite using common products, these segments are likely to transact differently, with different service needs. Data-driven analytics can identify these different usage patterns to inform which users belong to which segment. 2 IMF, 2015. Regional Economic Outlook: Sub Saharan Africa 3 For examples and case studies see the IFC, 2017 (b) Data Analytics and Digital Financial Services Handbook 6 Data Two sources of data drive the analyses presented in this Sampling of MSMEs report: mobile money transaction data of an MNO in Sub- Two different sampling approaches were used following Saharan Africa as well as newly collected survey data with a segmentation and listing of likely MSMEs based on an MSMEs in the same country that use the MNO’s mobile initial analysis of mobile money transaction data. The money product for business transactions through individual list of likely MSMEs was used for a random selection subscriptions. Analyses were conducted with each dataset of 250 respondents with quota for sub-segments that separately for (behavioral) segmentations as well as with came out of the supervised segmentation. For the rest the two datasets merged for the predictive modelling of the survey sample, enumerators randomly contacted exercise. MSMEs in commercial areas with a minimum quota of businesses to be interviewed per survey location Mobile Money Transaction Data - All data was and business size category (Sohos6, Micro and Small encrypted and analyzed using best-practice data Enterprises). The definition for business size categories governance structures. A total of seven months of mobile that was followed for this survey is presented in table money transaction data were included into the final data 1. This definition is in line with IFC’s SME definition in models4 and subscriber level data were aggregated to terms of number of employees. monthly transaction tables per transaction type, with additional geospatial information (cell tower locations). Filter questions for respondent selection The chosen architecture allows future scalability and ease Enumerators selected eligible respondents with four of use of data models. filter questions. Respondents had to be users of mobile money and they had to have an individual mobile money MSME Survey Data - A detailed survey of businesses that subscription that they use for business transactions. The have individual mobile money subscriptions with the MNO minimum transaction frequency was set to at least two provided the ground truth data for the MSME identification transactions per month to make sure the analysis would model. The collected data served as training data (true be able to pick up variation in transaction behavior positives) to identify MSMEs in the individual subscriber when datasets were merged. Lastly, respondents had to base based on their mobile money transaction behavior. have a decision-making role in the company they work The survey data also constitutes the basis for the qualitative for. segmentation of MSMEs to inform the development of mobile money products that are tailored to the businesses’ Topics covered in the survey unique usage patterns and needs. Between June and August The survey instrument covered individual socio- 2018, 1,275 MSMEs were surveyed in the commercial center demographic respondent information, details about (72% of surveys) as well as in other smaller urban and rural respondents’ access and usage of mobile phones, an locations in the study country. assessment of the business(es) respondents are working for, respondents’ banking level and trust in the formal Table 1: Business categories based on banking system, respective mobile money usage for numbers of employees business purposes as well as a section capturing the perception of the used mobile money service and questions testing the appetite for mobile money product features. The survey also included questions to identify the drivers NUMBER OF and barriers of digital financial service usage among MSME CATEGORY interviewed businesses in their country. These questions EMPLOYEES were posed along the six components of a framework 1 employee Soho = Small/Home Office for the identification and description of the drivers and 2-9 employees Micro Enterprise barriers of DFS that were developed based on an IFC 10-49 employees Small Enterprise ethnographic study7 on the perception and attitudes towards mobile money in four African countries. The 50-199 employees Medium Enterprise study country chosen for this research collaboration is the first country that the framework was applied to outside of the original study countries8 (Senegal, Cameroon, DRC, Zambia). 4 The months covered are February, April, May, June, September and December 2017 as well as January 2018. 6 Soho stands for “Small office/Home office”. The acronym is used here for individual entrepreneurs. 7 IFC, 2017 (a). A Sense of Inclusion: An Ethnographic Study of the Perceptions and Attitudes to Digital Financial Services in Sub-Saharan Africa. 8 For more information on how the ethnographic framework was applied, see: Heitmann S., Buri S., Davico G. and Reitzug F., 2018, Operationalizing Ethnographic Research to Grow Trust in Digital Financial Services, EPIC Conference Proceedings 2018 7 Methods The study seeks to answer two core questions: 1) if micro, Two clustering exercises were conducted. One used survey small, and medium-sized businesses can be identified and data (‘Qualitative Segmentation’ - part of Phase II) and the segmented from the transaction behavior of individual other one mobile money transaction behavior of identified mobile money subscribers; and 2) what are the profiles and MSMEs (‘Quantitative Segmentation’ - part of Phase III). mobile money usage patterns of different MSME segments. Reading through the analyses of the three phases and The core questions were assessed during three phases of their outcomes throughout this report may become the research: confusing in parts, since they result in different numbers of identified segments, clusters and MSMEs respectively. For - Phase I: an initial analysis of transaction data and clarification, figure 1 provides an overview of the research segmentation of potential businesses in the individual phases and their resulting numbers of identified and mobile money customer base. interviewed MSMEs. It also lists the different segments and clusters that were identified throughout the analyses and - Phase II: a survey to collect ground truth and behavioral visualizes their partial overlap across phases. characteristics. - Phase III: development of a model that identifies MSMEs based on their transaction behavior . Figure 1: Overview of Research Phases Phase I Supervised Segmentation Development of expert-based MSME definitions & segmentation based on transaction data 11,500 segmented MSMEs 5 resulting segments: 1. Acceptors 4. Cash-In / Cash-Out Agents 2. Service Providers 5. Bulk Sender Agents 3. Airtime Traders Phase III Phase II Predictive Modelling & Quantitative MSME Survey & Qualitative Clustering Clustering K-means clustering using ‘ground truth’/ survey data MSME Predictions using an Average Trading Days Composite Index & subsequent K-means clustering 1,275 interviewed MSMEs 32,600 predicted likely MSMEs 4 resulting clusters: 1. Basic Smalls 3. Mobile Money Agents 6 resulting clusters: 2. Non-Agent 4. Savvy Mediums 1. Acceptors 5. Medium Value Bulk Entrepreneurs 2. Services Providers Senders & Receivers 3. Airtime Traders 6. High Value Bulk 4. Low Value Bulk Senders & Receivers Senders & Receivers 8 SUPERVISED SEGMENTATION During the supervised segmentation, transactional data was aggregated and weighted based on four key features: In Phase I, mobile money transactions were analyzed using a composite measure of the average trading days per expert-based definitions of segments to identify highly transaction type and per month9; the average count of active users. Based on their transaction behavior, most of second parties10; the average monthly transaction count11; these users were classified as informal channel workers and the average monthly transaction amount12 of individual who appeared to be running businesses through their mobile money subscribers. personal mobile money channels. This initial analysis set the definitions of what types of After a detailed analysis of mobile money transaction data, MSMEs might be identified later in the individual subscriber interactive sessions were held with the MNO operational dataset. The resulting segments and their definitions and management teams, during which the transactional based on transaction metrics are shown in table 2. Results characteristics of businesses were discussed. Thresholds were so clear that labels or “personas” could be allocated and transaction-based filtering criteria were defined to the segments. Segmentation patterns are consistent to segment potential MSMEs using local domain and throughout different phases of the analysis and therefore operational knowledge and previous insights. The process given labels are also meaningful and largely consistent of defining and labelling potential business segments based with the ones from the segmentation of MSMEs identified on an initial analysis of transaction data in collaboration through the predicted modelling in Phase III. with the MNO is referred to in this report as Supervised Segmentation. Matching the transaction data with ground truth survey data and the development of a predictive model to identify MSMEs was only done at a later stage of the project. Table 2: Definition of Segments resulting from Supervised Segmentation AVG AVG COUNT SEGMENT TRADING TRANSACTION NUMBER OF SEGMENT DESCRIPTION OF 2ND PARTY LABELS DAYS PER TYPES TRANSACTIONS SUBSCRIBERS MONTH Acceptors Acceptor accepts P2P for > 10 >5 Cash-in or payments of goods or services. P2P Received Bulk Sender Bulk Senders use their > 10 >5 Cash-out or Agents subscriber account to do P2P Sent bulk P2P transfers to multiple people. Cash-in/ Cash-in or Cash-out Subscribers > 15 >4 Cash-in or > 10 Cash-out facilitate rebalancing liquidity of Cash-out Subscribers other agents by doing Cash In/ Out Transactions. Airtime Airtime Traders do large > 19 Airtime > 50 Traders amounts of airtime sales and purchases. > 250 Airtime transactions per month Service Service Provider Agents >3 >9 3rd party Provider facilitate service provider remittances Agents payments for multiple 2nd (e.g. paying bills) parties 9 Average trading days per month = AVG across transaction types (AVG Number of trading days of a subscriber per relevant transaction type per month) 10 Average count of 2nd Parties = AVG (Count of distinct 2nd parties a subscriber transacted with per month) 11 Average transaction count = AVG (Number of transactions per distinct transaction type per month) 12 Average monthly transaction amount = AVG (Total value of transactions/ Total number of transactions) 9 MSME CLUSTERING USING SURVEY their individual mobile money subscriptions. The fact that they emerge so clearly as a cluster and that their answers DATA show that this agent group better understands and uses mobile money more actively than other segments, confirms A survey of MSMEs that use individual mobile money the meaningfulness and interpretability of the identified accounts for business transactions provided the ground four clusters. Results of the qualitative segmentation truth data for the subsequent predictive modelling. A are further empirically validated through the transaction sample of respondents was selected from the list of informal behavior of each segment that was observed after matching channel workers identified in Phase I. Clustering businesses survey and mobile money transaction data. using information from the survey data later allowed to describe characteristics of four business profiles for MSME Matching survey data with mobile money transaction data sub-segments and to develop guidance on how to best helps to compare the four business clusters that emerge approach them in terms of marketing and product offering. from the survey data with the segments from the supervised segmentation discussed previously. Figure 2 shows that the Analysis using K-means clustering led to the identification biggest proportion (63 percent) of interviewed businesses of four primary segments among the 1,275 interviewed among those that match with the supervised segments are MSMEs and mobile money customers. The segments are mobile money agents. spread across demographics such gender, location, and company size. Identified clusters reflect characteristics This was expected since large parts of the supervised such as economic maturity, technology awareness, and segments were identified as informal mobile money risk-aversion. The four clusters are labelled “Basic Smalls”, channel workers that behave in part as formal agents. A “Non-Agent Entrepreneurs”, “Mobile Money Agents”, and large group of matching businesses (23 percent) classify “Savvy Mediums”. therefore appropriately as Service Provider Agents in the supervised segmentation (top grey horizontal flow bar in Although being a mobile money agent was not used as a Figure 2). Other Mobile Money Agents classify as Acceptors, segmentation characteristic, segments emerged that are Airtime Traders, or Cash-In/Cash-Out Agents accounting strongly aligned with those that are formal or informal for 12 percent of the matching businesses respectively. mobile money agents. The majority of entrepreneurs in Supervised segments from Phase I were largely found to be the Mobile Money Agents cluster are not registered agents informal mobile money channel workers. with the chosen MNO. However, they are often informal businesses that found a business value proposition by using Figure 2: Sankey Diagram - Overlap of Segments based on Survey Data (Qualitative Segmentation) and Segments from the Supervised Segmentation SURVEY SUPERVISED SEGMENTS SEGMENTS Service Provider Agents 28% Mobile Money Agents Acceptors 63% 17% Airtime Traders 20% Savvy Mediums Cash-in/ 19% Cash-out Agents Non-Agent Entrepreneurs 30% 7% Basic Smalls 11% Bulk Sender Agents - 5% 10 PREDICTIVE MODELLING FOR Figure 3: Logistic Regression – Receiving MSME IDENTIFICATION & Operator Characteristic (ROC) Curve BEHAVIORAL CLUSTERING OF MSMES 1.00 After an initial expert-based segmentation in Phase I and the collection of ground truth data as well as a qualitative 0.75 segmentation of interviewed businesses in Phase II, the True positive rate core analysis and development of the predictive model was done in Phase III building upon insights and results obtained during the previous phases. 0.50 Mobile money transaction data was ultimately matched with survey data of MSMEs (true positives13) to train models 0.25 on business related transaction behavior. The final model predicts and identifies with high accuracy who the MSMEs are among individual mobile money subscribers. MSMEs that were identified through the predictive modelling 0.00 0.00 0.25 0.50 0.75 1.00 were clustered and sub-segmented into six clusters based on their mobile money transaction behavior for further False positive rate insights. Logistic Regression Different methods were applied for the predictive segmentation. It is important to use more than one algorithm to predict the outcome of data and to compare the results of multiple algorithms and models to ensure consistency of Results are conclusive. The model can predict whether a results. Using just one algorithm can predict false positives subscriber is an MSME by calculating the composite index (identifying subscribers as MSMEs that aren’t). of average trading days per subscriber and per month. Applying this algorithm to the transactional data of the The same transactional metrics that were used for the MNO, identified 32,585 likely MSMEs in January 2018 (based initial supervised segmentation were also tested as features on the latest available transactional data) with almost 98 for the predictive modelling. Different machine learning percent accuracy (Figure 3). Figure 3 shows the accuracy of algorithms were used and compared for the predictive the model and confirms that the true positive versus false modelling. First, meaningful features were identified using positive rate is extremely good (blue line is far away from correlation analysis. The team then moved towards more the green dotted line). complex algorithms, such as a decision tree and a random forest algorithm, using the same features. A composite Figure 4 clearly shows the correlation between average index of customer activity that includes different measures trading days and MSMEs. The greater the average number of average trading days per month was identified as the of trading days across transaction types, the more likely most important and the average transaction count as the that a subscriber is an MSME. It was decided to stick with a next most important feature. strict threshold of 15 days for MSME identification to ensure high model accuracy. A logistic regression algorithm was ultimately employed to identify and predict who were MSMEs. Consistent The 32,585 users that were identified this way as very likely results on key features across previous analysis steps and MSMEs, were clustered based on their mobile money algorithms provided the confidence for going forward with transaction behavior using K-centroid clustering. The a composite index of the average trading days per person ideal number of clusters identified using this method was to predict the probability of a subscriber being an MSME14. six. These clusters all had similar attributes as the initial This straightforward design of the final model allows easy supervised segmentation in Phase I. interpretation, replication, and application of results by service providers. Succinctly, the model identifies businesses in terms of high-active transaction users consistently over a rolling window of time. 13 715 interviewed MSMEs could be identified in the transaction database. They represent the true positives. A random sample of 1,000 individual mobile money provides the true negatives (non-MSMEs). As the dataset of individual subscribers consisted of about 4 million subscribers, first, 6 different random sample of 1,000 were used to train 6 different datasets. The machine learning algorithms were trained using these 6 combinations of data. Resulting model accuracies were all within 1 percent of each other which provided the confidence to ultimately only use one random sample. 14 The outcome was also tested with a second feature, which only added a marginal contribution to the outcome of the algorithm. 11 Figure 4: Logistic Regression: Probability to be a business based on the Average Number of Trading Days 1.00 COMPOSITE INDEX OF THE PROBABILITY OF A AVG NUMBER OF TRADING SUBSCRIBER BEING DAYS PER CUSTOMER AND AN MSME 0.75 MONTH Probability of SME 8.9 Days 25% 0.50 10 Days 50% 11.1 Days 75% 0.25 12 Days 85% 10.0 15 Days 98% 11.1 8.9 0.00 0 6 12 18 24 30 Average number of trading days Key insights – Predictive modelling and identification of MSMEs • MSMEs with an individual mobile money subscription can be identified based on their mobile money transaction behavior. • IFC developed a robust model for MSME identification that achieves 98 percent accuracy in predicting and identifying which mobile money subscribers are MSMEs. • The model identifies 32,585 MSMEs among the customer base of mobile money users of an MNO in Sub-Saharan Africa. 12 Results Both the qualitative segmentation based on survey results All of the entrepreneurs who were interviewed use mobile as well as the quantitative segmentation based on the money for business operations. MSMEs show a high level predictive modelling of mobile money transaction behavior of trust in mobile money services and financial institutions; provide valuable information about MSMEs and their 76 percent express their trust in banks and MFIs. An even specific DFS needs and usage patterns. higher percentage trust MNOs (88 percent). More than 60 percent of respondents agree that mobile money services are meant for businesses like theirs. Mobile money is QUALITATIVE MSME SEGMENTATION generally perceived as a non-discriminatory option to USING SURVEY DATA manage money. Adoption of technology does not seem to be a barrier for entrepreneurs, and they say they have a Based on survey data, four MSME segments were identified: good understanding of mobile money services.More than basic smalls, non-agent entrepreneurs, mobile money 50 percent of respondents said they think that banks are agents, and savvy mediums. They can be distinguished the safest place to save money. In comparison, 38 percent of through their size, technological sophistication, mobile respondents said mobile wallets are the safest place. More money usage patterns, and their perception and usage of than 70 percent of entrepreneurs who were interviewed traditional financial services. Among the businesses that believe that using mobile money agents means less privacy. were interviewed, non-agent entrepreneurs make up the biggest proportion (35 percent in Figure 5), followed by Businesses have been using mobile money on average for 3.5 mobile money agents and savvy mediums. Basic smalls years. The main reasons they started using mobile money constitute with 14 percent, the smallest survey segment. was the speed of service, convenient pricing, and the ease of use of mobile money services. Issues with mobile money that have been experienced by the biggest proportions of respondents are poor geographic coverage, missing Figure 5: Distribution of transaction receipts, as well as liquidity issues with agents. interviewed MSMEs across Entrepreneurs look for safety, speed, and simplicity when Clusters using mobile money for conducting business transactions. More than 75 percent of entrepreneurs are interested in business-oriented DFS products for conducting transactions 35% such as paying salaries, receiving payment from clients, paying suppliers, collecting money from retailers, and transferring funds to and from bank accounts. 23% The following profiles describe each segment’s 28% characteristics, including mobile money transaction behavior and use of financial services. The profiles also 14% include guidance on how to engage different business segments in digital financial services. Non-agent entrepreneurs Savvy Mediums Basic Smalls Mobile Money Agents The emerging segments from the clustering that used survey data are mostly consistent with the other transaction data-based segmentations presented in the report section on ‘Behavioral quantitative MSME Clustering’. Interviewed businesses that were segmented as Mobile Money Agents appear, for example, in the transaction data-based segmentations as High Value Bulk Senders and Receivers as well as Service Provider Agents. 13 Non-Agent Entrepreneurs Non-Agent Entrepreneurs seem to behave more opportunistically in their businesses and are therefore also Segment Characteristics - None of the MSMEs in this more likely to value flexibility in mobile money services. segment are mobile money agents. These MSMEs are Across segments, they are the least likely to have a bank most likely engaged in informal businesses for themselves, account; only seven percent reported having a bank such as Small office/Home office businesses, or with a account. Among the banked non-agent entrepreneurs, 45 small team. Non-Agent Entrepreneurs have on average six percent have had a bank account for less than four years. employees. Their lines of business are trade, e-commerce, manufacturing, and services. Strategic advice for engaging Non-Agent Entrepreneurs - When reaching out to non-agent entrepreneurs, the Use of Mobile Money and other Financial Services - recommended strategy is to grow their current engagement Non-Agent Entrepreneurs have a relatively low engagement and usage of mobile money financial services by focusing on with mobile money at present. That presents MNOs with business-specific features and key concerns like latency or an opportunity to unlock revenue by targeting those security. Offering them low-cost incentives for enrollment businesses more specifically. and increased usage of the channel is another way to engage them and maintain activity levels. Transaction frequency should be a core indicator to measure and monitor this segment. Table 3: Mobile Money Transaction Profile of Non-Agent Entrepreneurs MOBILE MONEY TRANSACTION PROFILE OF NON-AGENT ENTREPRENEURS AVG trading days per month 2.6 AVG monthly transaction count 3.3 AVG monthly 2nd party count 1.5 Monthly transaction value per subscriber 25.3 USD AVG transaction amount 23.4 USD Main transaction types used • Airtime purchases (7 per month) Average mobile money account balance 46.2 USD ARPU 15 1 USD 15 Approximate Average Revenue per Users (ARPU) = Total value of fees paid per month and segment / MSISDN Count per respective segment 14 Savvy Mediums They are the largest segment that already have a bank account; 78 percent report having a bank account. Savvy Mediums Segment Characteristics - Savvy Mediums show a higher have varied financial transaction needs for their business level of technological sophistication and are more receptive to operations that mobile money can help them with. They use new technology and services for business purposes. They are more mobile money because they value the service above all for its likely to be mid-size companies with an average of 25-26 employees speed (32 percent) and convenient pricing (25 percent). and they have more formal and established practices. Their lines of business are construction, real estate, transportation, storage, Strategic advice for engaging Non-Agent Entrepreneurs - logistics, trade, e-commerce, and services. Savvy Mediums are a more formal category of business. Messaging to them should reflect that. Partnerships with Use of Mobile Money and other Financial Services - financial and business service providers for targeted products Savvy Mediums have greater exposure to traditional business and promotions are one possible engagement strategy. institutions and workforces. They are more likely to value Business supplier networks are also extremely high-value services that integrate with transactional financial services communities that can be targeted with tailored products. and credit facilities. Table 4: Mobile Money Transaction Profile of Savvy Mediums MOBILE MONEY TRANSACTION PROFILE OF SAVVY MEDIUMS AVG trading days per month 3.6 AVG monthly transaction count 7.8 AVG monthly 2nd party count 2.8 Monthly transaction value per subscriber 162.6 USD AVG transaction amount 42.9 USD Main transaction types used • Airtime purchases (20 per month) & • P2P transfers (15 per month) Average mobile money account balance 61 USD ARPU 2.3 USD 15 Basic Smalls Smalls complained about prices being too high; 42 percent said the service was not often available; and 39 percent Segment Characteristics – Basic Smalls are more likely express concern about the security of the service. to be technologically averse and have difficulty with new technology. There is comparatively less opportunity to Strategic advice for engaging Non-Agent Entrepreneurs - upsell products to this segment. These MSMEs average To engage Basic Smalls, emphasize ease of use for seven employees and have lower turnover. Their lines of essential functions such as sending and receiving money, business are manufacturing, trade, and financial services, and monitoring account balances. Other engagement offering among other things, mobile money or transfer strategies can be to conduct prolonged messaging on very services. specific value propositions of mobile money. Topics could be safety and ease of paying wages. Generally, engagement Use of Mobile Money and other Financial Services – and marketing towards Basic Smalls should focus on peace Seventy-nine percent of Basic Smalls do not have a formal of mind, emphasizing network security, support, and clarity bank account. They are very sensitive to cost and cash-flow of transaction confirmations. arguments and less receptive to value-add. Asked about the main disadvantages of mobile money, 45 percent of Basic Table 5: Mobile Money Transaction Profile of Basic Smalls MOBILE MONEY TRANSACTION PROFILE OF BASIC SMALLS AVG trading days per month 4.2 AVG monthly transaction count 7.7 AVG monthly 2nd party count 3.3 Monthly transaction value per subscriber 191.1 USD AVG transaction amount 52 USD Main transaction types used • P2P transfers (21 per month) & • Cash Withdrawals (12 per month) Average mobile money account balance 57 USD ARPU 4.4 USD 16 Mobile Money Agents the reason they use it; and 16 percent say it allows them to improve their income. Ninety-four percent of informal Segment Characteristics - Mobile Money Agents are mobile money agents are interested in becoming formal very independent and are likely to be micro or Soho sized agents. Most of them, 67 percent, have no formal bank enterprises with two-to-three employees on average. account, and the majority say banking services are for rich Their lines of business are financial services, mobile money people and larger, formal businesses. services, and transfer services. Agents in this segment can be both formal agents as well as informal channel workers. Strategic advice for engaging Non-Agent Entrepreneurs - Strategically, for MNOs they may be more relevant as a Use of Mobile Money and other Financial Services - channel than as individual users themselves. Most are Mobile Money Agents are very familiar with a variety of informal agents that found a business case in using their mobile money networks. More than 80 percent use different individual mobile money subscriptions and promotions. mobile money services at least once a day and 86 percent There is an opportunity to sign them up as formal agents say they understand mobile money marketing material. and to leverage their networks of customers. These informal Eighty-nine percent of mobile money agents say that they mobile money agents can be targeted with promotions know how to activate a mobile money account; 70 percent such as new customer sign-up rewards. Messaging should know what to do when a transaction fails; and 89 percent focus on concrete value propositions. Key messages could know how to make a transaction using a cell phone. As focus on cost of transactions, time to transact, and security non-agent entrepreneurs, they are frequently involved in of transactions. Loyalty programs are another engagement informal business as primary or secondary income. Twenty- strategy to encourage recommendations over competitors. eight percent of Mobile Money Agents use mobile money mainly because it is fast; 16 percent cite convenience as Table 6: Mobile Money Transaction Profile of Mobile Money Agents MOBILE MONEY TRANSACTION PROFILE OF MOBILE MONEY AGENTS AVG trading days per month 5.1 AVG monthly transaction count 11.1 AVG monthly 2nd party count 4.1 Monthly transaction value per subscriber 509.2 USD AVG transaction amount 52.8 USD Main transaction types used • P2P transfers (61 per month) & • Cash Withdrawals (22 per month) Average mobile money account balance 67.3 USD ARPU 5.9 USD 17 Transaction behavior of qualitative MSME segments Non-agent entrepreneurs have the lowest level of transaction activity, both in terms of frequency and volume Matching survey data with mobile transaction data helps (blue bars in figure 6). Basic Smalls have slightly higher to compare the transaction behaviors16 of the businesses levels of transaction activity across different metrics than that were interviewed and validates the meaningfulness of Savvy Mediums (orange and grey bars in figure 6). the clustered segments. Transaction activity in comparison – As expected, mobile money agents are the most active of the four segments. They transact with the highest frequency and value and have the most second party contacts (yellow bars in figure 6). Figure 6: Transaction Activity of MSME Segments identified based on Survey Data 509 12 500 11,11 10 400 8 7.82 7.68 300 6 5.1 210 4.12 4.2 200 4 163 3.63 3.32 3.29 2.56 2.78 100 53 2 1.51 52 25 23 43 0 0 AVG trading days per AVG monthly transaction AVG monthly 2nd party Monthly transaction value AVG transaction month count count per subscriber amount Non-agent entrepreneurs Savvy mediums Basic smalls Mobile money agents Average Revenue Per User17 of segment members that were identified through the predictive modelling – For DFS providers, these small businesses contribute and quantitative clustering. Across the qualitative survey about twice the revenue per user as individual mobile segments presented in this section, the ARPU is highest for money subscribers. Nevertheless, they generate less the Mobile Money Agents Segment and the lowest revenue revenue from monthly transaction fees than informal per user comes from Non-Agent Entrepreneurs. channel workers and any of the other business segments 16 Out of the 1,275 MSMEs interviewed, 715 businesses could be matched with the mobile money transaction database. Across segments, 106 Basic Smalls, 237 Non-Agent Entrepreneurs, 227 Mobile Money Agents as well as 155 savvy mediums were matched. 17 The ARPU is calculated here as the total value of fees paid per month and segment divided by the number of subscribers in the respective segment. 18 Key insights – Qualitative MSME segmentation • MSMEs that use mobile money services can be segmented into four meaningful clusters: Non-Agent Entrepreneurs, Basic Smalls, Savvy Mediums, and Mobile Money Agents. Asking for key characteristics when signing up new mobile money customers may allow DFS providers to identify and classify new business clients directly into these groups. • Non-Agent Entrepreneurs are most likely to be engaged in informal businesses for themselves or with a small team. Their level of financial inclusion and mobile money usage is the lowest across the segments. • Savvy Mediums are more likely to be involved in medium-sized enterprises with more formal and established practices. Most of them already have a formal bank account. • Basic Smalls are small companies, both in real terms (few employees) as well as in business terms (low turnover). They have low levels of financial inclusion and are very price sensitive. • Mobile Money Agents are mostly individual entrepreneurs or are engaged in micro-sized companies. They are often working as informal businesses but are interested in becoming formal agents. Despite their mastery of mobile money services, most are not formally financially included and think bank accounts are for larger, formal businesses. • Transaction characteristics of these segments mirror the qualitative profiles that emerge from survey data. Non- Agent Entrepreneurs have the lowest levels of mobile money transaction activity whereas Mobile Money Agents clearly outperform other segments in terms of transaction activity. 19 BEHAVIORAL QUANTITATIVE They are informal channel workers that are informally volunteering as agents. These entrepreneurs are already MSME CLUSTERING USING MOBILE actively behaving as agents by providing agent-like MONEY TRANSACTION DATA services to earn informal commission value. This finding has important implications for DFS providers in terms of agent roll-out and strategic engagement. The presented Results of the Supervised Segmentation segmentation provides an opportunity to identify, engage, and formalize the role of informal channel workers as The supervised segmentation that resulted from an initial formal agents and merchants. Leveraging already existing diagnostic analysis of mobile money transaction data and a customers to become formal agents or merchants that way set of expert-based definitions18 of likely MSMEs, returned a could support the often expensive and costly task of rolling total of 11,500 potential MSMEs in the transaction database. out a functioning active agent network. They can be sub-segmented into five categories of likely businesses - Acceptors, Bulk Sender Agents, Cash-In/Cash- Based on the supervised segmentation in Phase I, only a first Out Subscribers, Airtime Traders, and Service Provider Agents. subset of informal channel workers and likely businesses Among those likely businesses, Cash-in/Cash-out Agents and could be identified compared to the list of MSMEs that was Airtime Traders constitute the largest proportions. later identified with the help of the more comprehensive and rigorous predictive modelling (results presented in the The supervised segmentation also generated other important next section). Table 7 below shows the characteristics and insights, such as potential misuse of the current value key metrics of the mobile money transaction behavior of proposition of airtime and P2P transactions of some individual each segment. mobile money subscribers. Informal agents, entrepreneurs, and businesses were found in the subscriber data that used their individual subscriptions to gain more income. Table 7: Transaction characteristics of MSMEs identified through Supervised Segmentation AVG AVG AVG AVG TOTAL TRADING MONTHLY MONTHLY TRANSACTION TRANSACTION DAYS TRANSACTION 2ND PARTY AMOUNT IN VALUE IN JAN PER MONTH COUNT COUNT USD 2018 IN USD Acceptors 16 37 17 28 1.5 million (do Cash-Ins or receive P2Ps) Bulk Sender Agents 18 70 34 22 0.9 million (do Cash-Outs or send P2Ps) Cash-in/ Cash-out Subscribers 18 31 12 41 8.2 million (do Cash-In or Cash-Outs) Airtime Traders 25 208 1 1 0.6 million (do Airtime Purchases and Sales) Service Provider Agents 10 28 1 78 4.8 million (do 3rd party remittances e.g. paying bills) Key insights – Supervised Segmentation • Segmenting highly active mobile money users based on expert-based definitions and exploratory analysis of their mobile money transactions helps identify five segments of likely MSMEs – Acceptors, Bulk Senders, Cash-In and Cash-Out Subscribers, Airtime Traders, and Service Provider Agents. • A large number of identified likely MSMEs are found to be informal channel workers that already act actively as informal agents and merchants using their individual mobile money subscriptions to gain more income. There is opportunity for DFS providers to leverage their networks and to sign those businesses as formal agents and merchants. 18 See section ‘Supervised Segmentation’ - p.9 - for a reminder of the definitions that were developed and defined to segment MSMEs for a reminder of the definitions that were developed and defined to segment MSMEs. 20 Results from the Clustering of Predicted MSMEs The predictive model expands the analysis of the supervised Table 8 lists the six segments that resulted from the segmentation to identify – or predict – expected MSMEs from behavioral clustering and characterizes their behaviors the overall transactional customer base. The model identified through key metrics of transaction activity. The cluster about 32,600 likely MSMEs the MNO’s mobile money descriptions below show how usage patterns of segments customer database. They were clustered into six segments vary not only in terms of general transaction metrics but also based on their mobile money transaction behavior. The six regarding the types of transactions that they conduct. The clusters -- Acceptors, Airtime Traders & Service Providers, clusters can be used to target new MSMEs for segmented Service Providers, as well as Low, Medium and High Value Bulk value propositions. Senders and Receivers – all transact very differently regarding the frequency, volume and number of contacts they interact with and their use of specific transaction types. Bulk Senders and Receivers constitute 89 percent of identified businesses (blue clusters in figure 7). Figure 7: Distribution of identified MSMEs across Clusters 5% 3% 3% Acceptors 14% Airtime traders & service providers Service providers Low value bulk senders & receivers 42% 33% Medium value bulk senders & receivers High value bulk senders & receivers Table 8: Transaction characteristics of Clusters of MSMEs identified through Predictive Modelling MONTHLY AVG AVG AVG TRANSACTION AVG TRADING MONTHLY MONTHLY VALUE TRANSACTION DAYS 2ND TRANSACTION PER AMOUNT IN PER PARTY COUNT SUBSCRIBER USD MONTH COUNT IN USD Acceptors 20 33 7.5 2,850 40 Service Providers 19 54 6.6 1,690 21 Airtime Traders & Service Providers 25 157 11.9 2,330 16 Low Value Bulk Senders and 26 66 2.6 190 3 Receivers Medium Value Bulk Senders and 21 35 2.1 180 5 Receivers High Value Bulk Senders and 17 26 2.4 210 8 Receivers 21 Use of Transaction Types by Cluster - Acceptors mainly expert-based definition from the supervised segmentation. conduct high volumes of 150 to 200 P2P transactions per Similarly, Service Providers that emerge from the predictive month. Service Providers are providing services to their modelling are mostly likely to be defined Service Provider informal customers by making transactions for them. Agents, according to the supervised segmentation. The Service Providers are doing less P2P transactions but same holds for the predicted Airtime Traders and Service conduct predominantly third-party transactions. They Providers. Identified MSMEs overlap with the supervised conduct on average 150 third party remittances per month segments of Airtime Traders and Service Provider Agents. as well as 100 airtime transactions and bill payments. The Lastly, the identified clusters of Low, Medium and High cluster of Airtime Traders and Service Providers’ conduct Value Bulk Senders and Receivers from the predictive an average of 100 third party remittances per subscriber modelling are likely to be among the Bulk Senders and and their average number of airtime transactions is with Receivers as classified in the supervised segmentation in 300 transactions per month also very high. They also do Phase I. However, their more varied use of transaction types about 200 P2P transactions a month, which means that might indicate more diversified MSMEs. Low Value Bulk this cluster’s behavior is very similar to what we would Senders and Receivers, for example, also have high airtime expect from a business. Airtime Traders and Service transaction volumes. Medium and High Value Bulk Senders Providers demonstrate intensive usage of mobile money and Receivers also accept mobile money payments. across different transaction types. Low Value Bulk Senders and Receivers handle large transaction volumes but Average Revenue Per User of cluster members - lower transaction amounts both for their P2P and airtime In terms of value that DFS providers may get out of the transactions. Medium Value Bulk Senders and Receivers different clusters of businesses, the ARPU of these segments have subscribers who conduct fewer transactions than is consistently higher than the average revenue from the Low Value Bulk Senders and Receivers, but they conduct 35 qualitative segments discussed earlier (see end of section percent larger P2P amounts. High Value Bulk Senders and ‘Qualitative MSME Segmentation using Survey Data’ - p.18). Receivers have the biggest cluster of identified MSMEs, The average revenue through mobile money transactions 42 percent. These subscribers are showing 20 percent less from identified businesses is more than five times higher activity than Medium Value Bulk Senders and Receivers, but than the one from normal individual consumers. 35 percent larger P2P amounts. Clusters of identified MSMEs overlap with Supervised Segments - The approximately 32,600 MSMEs that were identified through the predictive modelling in Phase III overlap meaningfully with the 11,500 MSMEs classified during the supervised segmentation in Phase I. MSMEs identified as Acceptors in the predictive modelling are also most likely to be classified Acceptors according to the Figure 8: Clusters of predictive modelling - ARPU in USD 40 Across Segments 30 29.13 Acceptors 25.18 Small airtime traders 20 18.57 Service provider ‘agents’ Low value bulk senders 10 7.20 9.00 6.97 7.00 Medium value bulk senders High value bulk senders 0 22 The average revenue per month across identified businesses provider perspective, account for 11 percent of identified from the predictive modelling is 9 USD (grey bar in figure 8). MSMEs, as shown in figure 7. In contrast, the different The highest value segment are Acceptors with an ARPU of clusters of Bulk Senders and Receivers have comparatively more than 29 USD (green bar), followed by Service Provider low average revenues per users, ranging from 6.97 USD to Agents that bring an average revenue of 25.18 USD (yellow 7.2 USD, as illustrated in figure 8. They constitute the largest bar), and Smaller Airtime Traders with an ARPU of 18.57 USD share of identified MSMEs. (purple bar). Together these high value clusters, from a DFS Figure 9: Clusters of predictive modelling - AVG monthly account balance in USD 160 140 136.5 Across Segments 120 110.0 102.3 98.4 Acceptors 100 89.3 89.9 Small airtime traders 80 62.1 Service provider ‘agents’ 60 Low value bulk senders 40 Medium value bulk senders 20 High value bulk senders 0 Mobile money account balances - Interestingly, USD - blue bars in figure 8) have account balances on their businesses in clusters that have a high average revenue mobile wallets between 89 USD and 137 USD (blue bars in per user are not necessarily the ones that have the highest figure 9). Low Value Bulk Senders and Receivers, shown in account balances on their mobile wallets. Bulk Senders and Figure 9 as the light blue bar, is the cluster that holds the Receivers that had comparatively low ARPUs (below 7.5 highest amounts on their accounts, in excess of 136 USD. Key insights – Clustering of predicted MSMEs • Predictive modelling identified 32,600 MSMEs among the customer base of mobile money subscribers of an MNO in a country in Sub-Saharan Africa. These businesses can further be clustered into the six meaningful subgroups – Acceptors, Service Providers, Airtime Traders and Service Providers, as well as Low, Medium and High Value Bulk Senders and Receivers. • Identified clusters of MSMEs overlap with the groups from the initial supervised segmentation. Hence, validating identified business user groups. • All segments show high levels of mobile money transaction activity. Acceptors, Airtime Traders & Service Providers are the most active in terms of transaction count, transaction value, average transaction amounts, and the average number of second parties that they transact with on a regular basis. • The approximate average revenue per customer per month among identified MSMEs is on average more than five times higher than the one from normal individual mobile money consumers. They hold an average balance of 102 USD on their mobile wallets. 23 Implications Results from both the qualitative and quantitative In a competitive DFS landscape, having the ability to modelling done during the project have implications for DFS understand and anticipate the needs of MSMEs gives a providers, above all MNOs attempting to service MSMEs DFS provider a competitive market advantage. The model and their growth. Results contain a wealth of information that was developed through this project identified more regarding the MSME market. than 32,000 potential MSMEs among the partners MNO’s individual mobile money subscribers. That is approximately Additional model validation - For additional model 16 percent of the country’s total MSMEs. Although a rough validation, it is recommended that the MNO conducts estimate, this number is an indicator of the potential follow up calls or visits with a sample of the MSMEs that were impact that segmentation and data driven analysis can identified through the predictive modelling. During this have on the market and the continuous monitoring and validation process, additional information can be gathered. measuring of transactional data might provide an MSME The MNO should first confirm that the algorithm did not pipeline for future growth. predict a false positive, and it should also confirm that the MSMEs identified are not already using the MNO’s agent Having information like this available for DFS providers to and/or business profiles. Additionally, the service provider act upon, supplemented by qualitative survey insights, is a should enquire why respective MSMEs use an individual substantial market differentiator that can aid in increased subscriber profile to transact so frequently instead of using product offerings, retention and financial inclusion for other available value propositions. Possible reasons might currently underserved and underbanked MSMEs. be that they have found a better value proposition by using an individual subscriber profile, that they are not aware of other value propositions, or even that the current value propositions are not tailored to their specific needs. Product development – Understanding the above as well as combining and leveraging the qualitative information obtained about businesses from the survey and the quantitative information about the transaction behavior of different segments will aid with product design. Incorporating the results and using design approaches to then develop new business-specific DFS products helps to offer a better value proposition to micro, small, and medium enterprises. The integration of survey results is crucial in the design phase. The survey results contain valuable ethnographic factors and behavioral metrics that can assist providers to understand, for example, individual MSME’s barriers to use DFS, risk behavior, as well as their trust in the different products and financial institutions. Targeted outreach to MSME segments – Modelling and segmentation of MSMEs helps to identify business clusters among existing mobile money clients as well as support the acquisition of new MSMEs. By collecting key information about new subscribers, DFS providers can directly identify and segment businesses during registration. By knowing which customers are MSMEs, DFS providers can effectively target them with tailored marketing messages that encourage continued active usage of mobile financial services. 24 Conclusion This research showed that MSMEs with an individual mobile In Phase III, IFC built on information garnered from the first money subscription can be identified based on their mobile two phases to ultimately develop a robust model that can money transaction behavior. They can even be further predict with 98 percent accuracy which individual mobile sub-segmented into meaningful MSME tiers. Profiles and money subscribers are MSMEs. Using the latest data patterns emerge that help to sub-segment MSMEs based on available to the team (January 2018), the model identifies their usage of mobile money, their business characteristics, more than 32,000 MSMEs among existing mobile money and their financial needs. users, a sizable proportion of the country’s MSMEs. In Phase I, the initial segmentation based on transaction The next step for the MNO is to use and operationalize analysis and expert-opinion could already identify probable the results by integrating them into product design for MSMEs as well as different types of informal channel MSMEs, taking into account segmentation characteristics workers that are using the mobile money channel with when signing up new mobile money customers, to develop individual subscriptions for business purposes. segment-specific marketing messages and to reach out to identified MSMEs through phone calls or client visits In Phase II, clustering the more than 1,200 MSMEs that for additional model validation. Apart from additional were interviewed and that are mobile money users helped model validation, areas of potential future research include to describe business profiles that can inform product studying and mapping the suppliers and client networks design and targeted marketing. DFS providers can use that MSMEs transact with through mobile money. This this information in the future to filter and classify new will help to further advance the understanding of how customers as traditional consumers, agents, or other these businesses operate and help DFS providers identify MSMEs. Depending on the sub-segment and profile they networks of potential users in the ecosystem to leverage fall into, MNOs can then provide tailored products and and engage with. messaging that address their specific concerns and needs more effectively. 25 References Heitmann S., Buri S., Davico G. and Reitzug F., 2018, Operationalizing Ethnographic Research to Grow Trust in Digital Financial Services, EPIC Conference Proceedings 2018 - https://www.epicpeople.org/wp-content/uploads/2019/02/EPIC2018Proceedings.pdf IFC, 2017 (a). A Sense of Inclusion: An Ethnographic Study of the Perceptions and Attitudes to Digital Financial Services in Sub-Saharan Africa - https://www.ifc.org/wps/wcm/connect/15e6158a-8e52-444b-9103-391547cb1730/ IFC+A+sense+of+Inclusion+DFS+Ethnographic+Study+2017.pdf?MOD=AJPERES IFC, 2017 (b). Data Analytics and Digital Financial Services - https://www.ifc.org/wps/wcm/connect/region__ext_content/ifc_external_corporate_site/sub-saharan+africa/resources/dfs- data-analytics IMF, 2015. Regional Economic Outlook: Sub Saharan Africa – Navigating Headwinds, Chapter 2: How can Sub-Saharan Africa Harness the Demographic Dividend? World Economics and Financial Surveys. SME Finance Forum – MSME Finance Gap https://www.smefinanceforum.org/data-sites/msme-finance-gap AUTHORS Sinja Buri is a data operations analyst in the applied research and learning team of the Partnership for Financial Inclusion, based in Dakar, Senegal. Her research projects focus on understanding how different customer segments take-up and use digital financial services in Sub-Saharan Africa as well as the resulting business implications for financial service providers. Morne van der Westhuizen is a data science consultant with 12 years of data science experience in the financial industry spanning 16 countries. Soren Heitmann leads the applied research and learning program of the Partnership for Financial Inclusion. His background is in data science, development economics and cultural anthropology. ACKNOWLEDGMENTS IFC and the Mastercard Foundation Partnership for Financial Inclusion are grateful to the Mobile Network Operator that participated in this study for its collaboration in realizing this research project. Thank you also to the supporting market research company for their work and input as well as Gary Seidman and Lesley Denyes from IFC for editorial support. April 2019