Predictive modelling will advance financial inclusion as the rural business case improves: a case study in Uganda

Christian Ruckteschler
30 Jan 2018

By Christian Ruckteschler

  1. Introduction

A key aspect of responsible finance is inclusion: ensuring that the financial system works not only for the wealthy and urban but for all members of society. Yet achieving financial inclusion in the poorest and most remote rural areas remains a difficult feat. The lack of economic viability keeps banks and other financial institutions out of remote areas, and even mobile money agents are currently far and between. This may be about to change. With a quickly rising mobile phone penetration rate, rural areas are moving into the focus of business interest. Mobile Money operators are increasingly focusing their growth strategy on an expansion into the more remote areas. A case study in Uganda suggests that MNOs see themselves in a competition to service new areas and hence new customers. In this post, we argue that they are right: our analysis shows that an increasing share of potential high-value customers is located in rural areas. Hence, financial inclusion goals and business incentives are becoming more aligned. We expect access to mobile money to reach many more remote rural areas in the near future.

This analysis is a result of research by IFC’s Applied Research and Learning Program under the Partnership for Financial Inclusion and further elaborates on a case study presented in the Handbook on Data Analytics and Digital Financial Services. Through a long-term research engagement with Airtel Uganda and in collaboration with the Bill & Melinda Gates Foundation, the team analysed two six-month periods of anonymized Call Detail Records and Airtel Money transaction records for the entire Airtel Uganda customer database (Nov 2014 – Apr 2015 and Oct 2016 – Mar 2017). In this post, we present findings from a predictive model we built using machine learning methods to identify high-potential mobile money customers. Using the results of the model, we find that the geographic distribution of high-potential customers is much more spread out in 2016 than it was in 2014.

  1. A key user group to focus on

To better understand mobile money users, we first segmented the customer database and followed customers over time. A helpful tool for this analysis were diagrams that visualize the flow of customers between different segments over time, known as Sankey diagrams. This revealed the existence of a small group of key customers which set themselves apart along several dimensions: they transact more frequently, generate more revenue than any other segment, and are more stable over time. Consequently, this small group of “Highly Active Users” accounts for a large share of the service’s revenue. In addition, their active usage makes the network more useful and attractive to other customers. MNOs would thus be well-advised to focus on growing this segment, and to work on identifying non-users with a high probability of falling into this segment if activated.

  1. Identifying High Potentials

To identify high-potential non-users, we built a machine learning model that predicts a GSM user’s Airtel Money customer segment. The first significant finding is that this was possible: using only call detail records, the model correctly predicted a user’s customer segment in 85% of cases. The model generates a prediction score between 0 and 1 for each GSM user, indicating the user’s probability of becoming Highly Active. Ordering GSM users with no mobile money activity according to this score provides a ranking of the most high-potential users to target. Comparing user activity two years down the line to the predictions of the model shows that it works remarkably well. Converting the highest ranked customers to mobile money users through targeted marketing could significantly grow the Highly Active segment, boosting revenue and overall business performance.

Linking such a prediction model to geographic information can inform business decisions and growth strategies by uncovering which geographic areas have the highest density of high-potential users. This is particularly important for MNOs as they look to expand into rural areas. One aspect of responsible finance is to serve clients reliably and sustainably. Taking a data-driven approach can help MNOs identify where this is feasible.

  1. Financial Inclusion: Business Interests and Development Goals Increasingly Aligned

To learn more about the urban-rural split of high potential GSM-only users, we plotted their density on a map of Uganda, using cell tower locations to approximate their whereabouts. For reasons of confidentiality, we use customer data from 2016 to illustrate here. Predictably, many high potentials are located in and around Kampala, the capital city. Yet less predictably, a significant share of them is located far from the capital and in more remote areas, as Figure 2 illustrates. This represents a marked change from two years prior (2014), and implies that Airtel should have a growing strategic interest to increase their mobile money presence in remote areas. Such an expansion would substantially increase financial inclusion.

Figure 2 – Density of high-potential non-users (2016 data)


We find that a marked shift has taken place in the attractiveness for mobile money operators in Uganda to serve rural and remote areas. While the business case previously led MNOs to focus on urban areas, more and more high potential users are now located in rural areas. They can be found by using the data analytics methods discussed here and in our published handbook. MNO’s expansion strategies suggest that those in charge are aware of this trend. We suspect that Uganda is not unique in this regard but that the attractivity of rural areas to mobile money operators is increasing in many countries across Sub-Saharan Africa together with mobile phone penetration. This is a good message for responsible finance: we can expect business interest to drive financial inclusion of remote areas in the next years. Preliminary indicators suggest that the Highly Active segment is moving down market, too: not only are High Potentials more geographically spread, they are poorer than before, too.

This case study also demonstrates the significant potential for big data analytics to inform expansion strategy and enable effective targeted marketing campaigns. Predictive models can identify not only High Potentials as demonstrate here but other user categories as well. For instance, we have built similarly accurate churn prediction models to target customers even before they leave the service. MNOs in Sub-Saharan Africa are only beginning to tap into this potential.


Adapted from a case study presented in the Data Analytics and Digital Financial Services Handbook (June, 2017), this post was authored by Christian RuckteschlerIFC-Mastercard Foundation Partnership for Financial Inclusion, for the Responsible Finance Forum Blog January2018. We thank Nicolais Guevara and the team at Cignifi Inc. for their tremendous work without which this analysis would not have been possible.