Page 16 - FIGI - Big data, machine learning, consumer protection and privacy
P. 16
of customers exhibiting similar characteristics. Sit- Figure 2 – Smartphone app permission settings
ting above the credit scores are ‘business rules’ set
by the bank that will determine the actual limits that
may be offered to customers. These include a cut-
off score which a customer must reach and maintain
in order to qualify for a loan, limit caps for different
customer segments, individual customer limits (e.g.,
credit limits determined by a formula applied to their
average monthly mobile money wallet cash-inflow),
and entry barriers such as blacklisted customers
due to previous default or negative credit reference
bureau status.
Firms that offer value added payment services are
also increasingly able to use data about payments to
reach decisions on credit. For instance, Kopo Kopo
facilitates merchant access to Safaricom’s M-Pesa
payment system in Kenya, setting up APIs enabling
merchants to receive payments and managing the
receipt and accounting for payment receipts. This
affords the company a unique window into its mer-
chant customers’ cash flows, putting it in a strong
position to evaluate their creditworthiness and so to
develop a lending business.
Data on online activities
Beyond a customer’s use of the mobile network
operator’s services, large quantities of data from
activities on web browsers and mobile phone apps Broader types of data
are collected and shared, often without being There are numerous other sources of data about a
subject to standard opt-in policies. For instance, person that may be combined for the purpose of big
a recent Oxford University study of about 1 million data operations. These may be collected from retail
Android apps found that nearly 90 per cent of apps shops where a person makes purchases, from credit
on Android smartphones transfer information to card companies used for transactions, data passively
Google. Customer internet usage may be swept up collected from Bluetooth detection devices in shops,
33
along with location data, contact information and images of a person gathered on video cameras, car
text messages (see Figure 2). number plates collected by video cameras, medi-
The data market allows web tracking as well as cation information gathered from pharmaceutical
cross-device tracking that makes it possible to link a purchases, recordings made by toys with installed
person’s use on a smartphone to his or her computer microphones and cameras, and a vast number of
and tablet. As the internet of things develops, data other sources. One adviser to investors in big data
from devices a person uses at work, home or on market players lists the following sources of alterna-
their body will increasingly be linked. As a result of tive data available in today’s market: 34
this wide range of linked data sources, it is possible
to track a user’s location via mapping apps, brows- • Data from financial aggregators
er and search history, whom and what they “like” • Credit card data
on social networks, videos and music they have • Geospatial and location data
streamed, their retail purchase history, the contents • Web scraping datasets
of their blog posts and online reviews, and much, • App engagement data
much more. Companies such as Branch, Tala and • Shipping data from U.S. customs
Jumo have developed substantial digital credit busi- • Ad spend data
nesses in Africa relying on such alternative data. • Data made available through APIs
• Location/foot traffic data from sensors and rout-
ers
14 Big data, machine learning, consumer protection and privacy