Page 16 - FIGI - Use of telecommunications data for digital financial inclusion

Machine learning (ML) algorithms are developed accordance with minimum monthly prepaid airtime
using training data. In MobiScore, an AI system which purchases.
12
develops credit scores from mobile telephone data,
CDRs are used to identify patterns of behaviour that
correlate with unreliable financial behaviour. In paral- Using telecommunications data to enable posi-
lel, credit reports showing actual defaults for the tive credit scoring in Brazil
same individuals are used as a ground truth to train The Brazilian positive credit scoring agency Quod
the user models. Thus, algorithms segment custom- is using telecommunications data of the coun-
8
try’s telecommunications operators in partnership
ers according to a range of behavioural and risk with US fintech company Cignifi. They are offering
assessment registers. For example, US firm Cignifi credit insights into customers, marketing insights
9
worked with Airtel in Uganda to obtain data about into customers’ propensity to certain services, and
the number of calls and text messages made and fraud scores to screen credit applications.
received per day and phone, web and social network While 97% of Brazilians have mobile phones, 30%
usage, and then analyse that data comparatively do not have bank accounts. The use of telecommu-
nications data can thus enable access to credit for
using generic models of behavioural patterns. millions of underserved customers who otherwise
One study of mobile telephone data on loans to would not have had such access, and widen the
banked and unbanked customers in a middle-income range of products to which individuals and small
Latin American country showed that such data out- businesses would otherwise have had access. The
partnership’s product offerings will include cred-
performed traditional credit bureau data: ‘Among it insights to complement Quod’s positive scores,
those with credit histories, if credit were extended fraud scores to screen credit applications and
to the 50% lowest risk prospects according to the on-line transactions, and propensity indicators to
credit bureau the default rate would be 9.7%, where- enable digital marketing initiatives.
as it would be only 8.3% based on our scoring using Source: See https:// www cignifi com/ post/ manage
.
.
phone records. Moreover, if credit were extended to -your -blog -from -your -live -site
those without credit histories whose predicted risk of
default would place them in the top 50% of risk-pros-
pects for those with credit records, the default rate
would be only 6.6%. Our method can identify a group
of good credit prospects from among those with no 4�3 Risk and asset management more broadly
credit history.’ Telecommunications data can be used not only to
10
While industry participants remain sceptical of reduce risk through credit scoring and profiling, but
the ability of any CDR-based model outperforming a to provide information about assets that are being
credit score based on data on past repayment histo- financed or insured. For instance, location data is
ry, there does appear to be a strong opportunity for sometimes used to track leased vehicles using start-
using such models where such historical repayment er interrupter devices. Where the customer fails
data is not available. Safaricom launched M-Shwari to maintain service on his or her loan, the SID may
in 2012 in Kenya, the first digital credit product that communicate not only location but may instruct the
relied on telecommunications data (albeit combined disablement of the vehicle enabling its recuperation
with mobile money usage data) to evaluate risk. by the lender.
Banks, telecommunications operators and insurance Insurance companies are also already using a
companies have also sought to capitalize on tele- variety of IoT data to assess risk. This includes using
communications data to improve delivery of finan- telemetry data (which can track vehicle location and
cial services. Telmex, the largest fixed line operator usage), sensor data from personal fitness devices,
in Mexico and a subsidiary of America Movil, which smoke detectors, burglar alarms and weather gaug-
also owns Mexico's largest MNO, offers small busi- es. Most of this IoT data is transmitted over the net-
ness loans to customers based in part on their phone works of mobile telecommunications operators and,
records. MicroEnsure, a microinsurance firm, has when used with machine learning, can provide more
11
partnered with Telenor Pakistan, the largest MNO in accurate predictions about insurance claims.
Pakistan, to provide free life insurance for users in

14 Use of telecommunications data for digital financial inclusion