Page 15 - FIGI - Big data, machine learning, consumer protection and privacy

velocity (the “three Vs” – sometimes expanded to Many countries’ telecommunications laws and
28
four Vs by the addition of “veracity”). The advent licences include clauses expressly prohibiting licens-
29
of big data techniques arises from developments ees from using, disclosing or recording any communi-
in how data is collected, stored and used. Data is cation or content sent using an electronic communi-
collected using numerous applications and sensors cation service or information relating to such services
which record consumers’ communications, transac- provided to others. This is increasingly extended to
tions and movements. Distributed databases store metadata. For example, the EU’s ePrivacy Directive is
the data, and high-speed communications transmit being replaced with the ePrivacy Regulation, which
it at high speed, reducing the cost of data analytics. fleshes out data protection themes of the GDPR fur-
Advanced analytical processes are applied in numer- ther specifically for electronic communications ser-
ous contexts. vices, addressing both personal data and metadata,
such as call detail records (CDRs). However, this is
31
2�2 What kind of data is used? not universal, and many countries do not prevent use
In the financial services context, historically, data of metadata. Even where it is prohibited, it may be
used for decision making might have included formal permitted with the customer’s consent, enabling the
representations by an applicant for a service, some operator to generate credit scores that may be used
personal knowledge by the local bank manager or to extend digital loans.
insurance broker, and a broader range of organized
data held, analyzed and profiled through credit refer- Mobile money and other payment data
ence bureaus. Today, big data includes alternative Telecommunications companies may also hold data
data, i.e., data that is not collected and documented about the use of related services that are carried
pursuant to traditional credit reporting but from a over telecommunications networks. For instance,
wide range of other digital sources. many mobile network operators provide a propri-
etary mobile payment service to their customers. As
Telecommunications data a result, they have access to data about when, how
An important source of alternative data being used to regularly and by how much a person tops up his or
extend financial services is derived from telecommu- her mobile money wallet, the average balance he or
nications network operators’ services. Telecommuni- she maintains, who he or she makes payments to or
cations companies are typically constrained in their receives payments from and the amounts of such
ability to collect and use data about their custom- payments. By analysing the regularity, amounts and
ers, particularly the content of their telephone calls. recipients (e.g., family, utility invoices or school fees)
These have been protected by legislation on lawful involved, data analytics can form a picture of the
interception with themes similar to the laws protect- scale and reliability of a person’s cash flows (both
ing postal communications that prohibited the open- income and expenditures), his or her social network,
ing of envelopes without a lawful basis. However, and ultimately enable assessment of creditworthi-
while telecommunications companies may not use ness. Regular payments of utility bills or school fees
the content of their customers’ communications, may indicate a regular cash flow and generally posi-
they also have access to (and are often required by tive approach to payment of debts.
regulation to retain ) metadata. Access to such data is proving to be a useful
30
Metadata are data about the customer’s use of means of introducing people hitherto excluded from
their communications services, including who com- financial services – due to lack of information about
municated with whom at what time, for how long, them – to digital financial services. Mobile network
and the location from where the call was made, the operators have in many cases partnered with banks
combination of which can help profile an individu- to facilitate mobile lending using credit scores devel-
al’s relationships and cash flows. Regular topping up oped using the mobile network operator’s data
of prepaid phone credit may imply a stable income. about the customer. The operator might not share
Calls to and from abroad may imply access to an the raw mobile money data or call metadata with the
international network, and potentially greater afflu- banks, but will often apply algorithms to it to pro-
ence. Regular calls during the working day in a dense duce a credit score.
urban area may imply a steady job, and calls made To take one example , one mobile network oper-
32
or received at the same location in the evenings may ator uses 48 parameters over a 6-month period and
indicate the location of the individual’s home, and so information collected in the individual’s registration
economic or social class. (KYC) process to produce a scorecard and buckets

Big data, machine learning, consumer protection and privacy 13