71304 BRIEF Can Digital Footprints Lead to Greater Financial Inclusion? Whether in Mexico, Ghana, or Pakistan, millions of poor people wake up to the glowing light of their mobile phones. Poor people have mobile phones, but not formal financial services. CGAP and GSMA estimated that close to 2 billion people will have a mobile phone but not a bank account in 2012. Poor people’s use of their mobile phones to rise as it becomes cheaper and easier to access generates data that leave what can be called a Internet services.1 digital footprint. These data are among a handful of 4. Financial transactional data: As the volume of sources of electronic information that exist on poor mobile money transactions continues to grow, people. This information is potentially powerful but providers will have access to a deeper well of data. has not yet been used in ways to radically impact financial access for poor people. CDR and other basic data are passive data, whereas data provided by people in response to surveys, on This Brief highlights some early experience on the social media sites, and so on, are considered active potential of digital footprints from mobile phone data. While the term active data more accurately use. Most of this initial work is experimental. As describes the person who generated the data, both long as consumer interests are protected and passive and active data have potential value. For privacy, security, and ethical use concerns are instance, Massachusetts Institute of Technology addressed, these data may become a useful way researchers were able to predict with 74 percent to reach unbanked poor people with a range of accuracy the sex of mobile phone users based financial products. primarily on an analysis of CDR data (Eagle et al. 2009). Data behind digital footprints The duration for which these data are available There are four basic kinds of data generated by impacts the quality of the analysis. For instance, mobile phone use: according to Wired (2001), the four major U.S. MNOs keep basic CDR data anywhere from four 1. Timing, location, and duration of voice and text- months to three years or longer. Regulators may or message and airtime purchase: Mobile phone use may not specify “data retention� rules. In Pakistan, for voice calls and text messages generates a call carriers are required by the telecommunications detail record (CDR) that is recorded by mobile regulator to keep data for at least three years. 2 network operators (MNOs) to accurately bill In addition, some operators do not attach a customers. new subscriber identity module (SIM) card to an 2. Use of value-added services, such as ringtones, actual subscriber name, making it difficult to track text-messaging-based services: People also account-level relationships with a given customer, download ring tones, play games, subscribe to especially someone with two SIM cards. text-messaging-based information services (e.g., sport scores, agriculture pricing, health alerts, etc.) As shown in Figure 1, the variables from these data and respond to text-messaging-based surveys. can be placed on a continuum from direct financial 3. Internet use: While the poor have traditionally not transaction data and other data that map a user’s used the Internet much, the rate of usage among use of money, to data that is nonfinancial and less those living at or near the poverty line is expected obviously related to financial services, but could 1 In Kenya, 90 percent of Internet usage occurs via cell phones, and 31 percent of Internet users are on Facebook. In June 2012, Airtel Ghana and Facebook launched Facebook’s text-only mobile site called “Facebook Zero,� which allows free access to Facebook. 2 See http://www.pta.gov.pk/images/stories/kashif/apc_rules.pdf. The data include “all books and accounts pertaining to payments made or received…and the telecommunication services to which such payments relate, including call detail records and itemized billing data.� July 2012 2 Figure 1: Value of variables for financial services Overall • Mobile wallet use data Weighting (e.g., average daily balance, type, size and • Monthly airtime usage frequency of payments, net remittance sender or receiver) • Number of unique • Value-added services calls and text • Level of airtime at time of messages • Information text airtime purchase messages • Time of usage during • Purchases via mobile the day • Response to surveys wallet; m-commerce • Location information • “Social graph� from • Subscriber tenure social media use Directly relevant for financial services Less directly relevant for financial services be insightful. The challenge is determining how to zero could be rated “low risk,� while a customer weight the variables to build an accurate profile of who had just activated his SIM and reloaded a low the customer relevant for financial services. number of minutes now followed by periods of inactivity might be rated “high risk.� Use of digital footprints for financial services In the end, CGAP was unable to test the model, but companies such as Cignifi and Experian have In partnership with an MNO, CGAP set out to gone further. Cignifi built a credit scoring model test some hypotheses on how digital footprints using CDR data and tested it in Tanzania and could be used to deliver credit to the unbanked. Brazil. It built a model in Brazil using 50 variables For example, we hypothesized that people who from 2.3 million prepaid customers of MNO Oi’s purchased airtime frequently and in a consistent mobile business and back-tested the model against pattern (similar amount or at similar times) historical lending data from approximately 40,000 demonstrated predictability in income and better borrowers of Oi’s lending business, Oi Paggo. The planning, which might impact their ability to repay test showed the model was an accurate predictor a loan. We also hypothesized that people who have of default—its scores were positively correlated an inactive prepaid account or one that consistently with default across the lending portfolio. The runs to zero airtime balance before their next score could be a useful complement to a credit airtime purchase may not be strong planners. underwriting effort even if it would not replace it. Experian Microanalytics did a similar trial in While no one variable in isolation is likely to be the Phlippines with MNO SMART and its partner, adequate to profile a customer’s credit risk, our mBank. premise was that certain variables in combination might be able to do so. For example, a customer While the initial experience with digital footprints who has been active for three years, reloads the has been primarily with credit, there is growing same amount of airtime minutes every Friday, interest in developing models for other products. and rarely lets his prepaid balance run down to Digital footprints could help match people to types 3 of insurance and help providers tailor premium data supports what are known as “two-sided levels and payment methods to fit people’s abilities business models,� where the data themselves are and needs. Indeed, digital footprints from mobile considered a bigger source of revenue than direct phone use could yield two basic types of models: charges to customers. For example, a “freemium� (1) predictive models to design financial products business model in mobile money would offer basic and (2) “propensity� models to be used primarily transactions for free to build transactional volume; for marketing. For instance, providers could use once you build sufficient volumes of transactional the propensity models to match savings products data, analysis of that data could yield revenues in with poor people more likely to save based on their excess of the revenue foregone by providing the mobile phone usage. services for free (Kumar and Mino 2011). Businesses using digital Protecting consumers’ privacy, footprints for financial services security, and ethical use of their digital footprints As shown in the Figure 2, companies involved in the use of digital footprints can be categorized based Even though there are opportunities for innovation on the type of data—financial or nonfinancial—and by providers, there remain significant privacy, whether data are used for predictive or propensity security, and ethical use concerns with the use models. Because nonfinancial data cover more of these data. It is also unclear which regulatory people and are less explored so far, the opportunity body—the telecommunications regulator, banking for continued innovation is on the right-hand side regulator, or another—has relevant jurisdictional of Figure 2. oversight. On the bottom left of Figure 2, the opportunity Some countries clearly regulate the ownership, for innovation is in business models. The potential use, handling, storage, and transfer of personal revenue from analysis of financial transactional data. Regulations in Mexico differentiate between Figure 2: Example of businesses that use digital footprints for financial services Predictive models for financial products (e.g., credit risk profiles) Cignifi: analyzes Experian and other credit CDR and related data scoring companies that use financial transactional data Lenddo: analyzes social graph Financial data Experian Microanalytics: analyzes Nonfinancial data both financial and CDR data Real Impact: analyzes mobile Jana, Esoko, Mobile Accord: wallet transactional data for companies that have “active� data MNOs and platform companies but not used for financial services Propensity models for marketing and driving adoption 4 July 2012 All CGAP publications are available on the customers as “data owners� and providers as • More openness to experimentation by providers, CGAP Web site at “data custodians.� However, in most countries, either in partnership with existing analytics www.cgap.org. data privacy laws are not well developed. It is companies or through investments in their own unclear if poor people even have rights to the home-grown analytics teams CGAP 1818 H Street, NW digital footprints data generated from their own • Improvements in regulation and clear guidance MSN P3-300 mobile phone usage and exactly how they would to providers on protection for consumers when Washington, DC consent to the use of these data. Unless there are it comes to data privacy, security, and ethical use 20433 USA adequate protections in place, poor people’s data Tel: 202-473-9594 may get commercialized without their consent or References Fax: 202-522-3744 knowledge. Eagle, Nathan, et al. 2010. “Who’s Calling? Email: cgap@worldbank.org Even the world’s most sophisticated data Demographics of Mobile Phone Use in Rwanda.� companies struggle to protect data they have AAAI Spring Symposium 2010 on Artificial © CGAP, 2012 from criminals looking to exploit that information Intelligence for Development (AI-D). http://ai-d. for financial gain. There are also two sides to the org/pdfs/Blumenstock.pdf ethical use question. On the one hand, consumers may game the system (e.g., change prepaid airtime Kumar, Kabir, and Toru Mino. 2011. “Can Mobile purchase patterns) if they are aware of how data Money be Free?� CGAP Technology Blog. http:// are being used. On the other hand, poor people technology.cgap.org/2011/05/20/can-mobile- may be denied financial access based on analysis money-be-free/ that might treat them unfairly. Pakistan Telecom Authority. 2004. Access Conclusion Promotion Contribution Rules. Pakistan: Pakistan Telecom Authority. The following will need to happen for digital footprints from mobile phone use to make a Wired. 2011. “Which Telecoms Store Your Data difference to the unbanked poor: the Longest? Secret Memo Tells All.� http://www. wired.com/threatlevel/2011/09/cellular-customer- • Better understanding of the availability and quality data/ of data AUTHORS: Kabir Kumar and Kim Muhota