Page 19 - FIGI - Use of telecommunications data for digital financial inclusion

5�2 Data sharing proprietary product model The proprietary data sharing model can also make
use of new technologies, including “federated learn-
5.2.1 Data interrogation analytics without ing,” to improve the accuracy of artificial intelligence.
sharing model Federated learning is a form of machine learning
An alternative to the partnership model occurs where artificial intelligence learns algorithms from
where the MNO does not share raw data, but instead distributed datasets. In other words, the data sets
collects and processes its subscribers’ data to gener- themselves are never shared, they are only used to
ate a stand-alone credit scoring product that can be train the artificial intelligence algorithms. In China,
sold to financial institutions or other third parties. Tencent's WeBank has used federated learning arti-
KT Corp., South Korea’s largest telecom operator, ficial intelligence to develop small business credit
provides a good example of how this model oper- risk models based on distributed data in individual
ates. KT Corp. also owns the first internet-only bank invoice centers. In this case, the raw invoice data in
in Korea and is active in the fintech, AI and big data the individual invoice centers is never actually shared
analytics spheres. KT Corp. has developed the “K-Tel- with WeBank, but the data is used to develop the
co Score” and the Credit Rating Delivery Platform artificial intelligence algorithms used by WeBank in
(CRDP) as alternative credit rating products using its business. Using this approach enabled WeBank to
telecom big data analysis. cut its loan defaults in half. 16
The K-Telco score uses a variety of telecom data,
including CDRs, subscriber data, usage data and 5�3 Data interrogation analytics without sharing
application data. KT has also commercialized a cred- model
it reporting delivery platform (CRDP) that produces Under another model, a telecommunications oper-
the K-Telco score from raw data. CRDP is updated on ator may permit a financial institution to interrogate
a daily basis with telecom data, which is synced, pro- the telecommunications operator’s data without
cessed through the application of big data analysis supplying that data to the financial institution. This
and machine learning, and packaged into an updated widely used approach involves the financial institu-
K-Telco score and profile. tion installing its business rule engine within the tele-
KT markets the K-Telco score and CRDP to tele- communications operator’s premises, and running
com operators globally as a way to monetize their analytics on it for various use cases. This results in
data. For example, an MNO operating in sub-Saharan a computed value or output, such as a credit score
Africa could purchase the CRDP and use the platform or confirmation as to regular locations. The financial
to generate credit scores based on their subscriber institution might aggregate the output with other
data. Under the business model, when a prospective data about the customer, such as KYC data it has
customer applies for credit, they will provide permis- already obtained, or credit bureau data to which it
sion to use personal data for credit purposes. The has access. Typically, the telecommunications opera-
creditor will request credit data from a credit bureau, tor will require the customer to consent to his or her
which will in turn seek telecom-related credit infor- data being accessed in this way.
mation. This will be provided by the MNO in the form Telecommunications operators may be remuner-
of the K-Telco score. ated on the basis of a revenue share. The details of
There is less transparency to the proprietary prod- such revenue share arrangements may be complex
uct model, but it appears that in the K-Telco score to determine, particularly where the financial insti-
and CRDP appear to seek to make use of the full tution combines the data output with other data
range of data available to MNO. This would include before using it, or where the financial institution uses
telecom data, subscriber data, and potentially other the data for purposes other than immediate reve-
data as well. It also isn’t clear whether, in the case of nue-generating services, such as for digital advertis-
the CRDP, KT would have access to the raw data of ing.
the MNO using the CRDP.

Use of telecommunications data for digital financial inclusion 17