There are many students applying for a loan but only a few complete the process. Diosa Jamila Angsioco, Kathlyn Cacho, and Hanceely Villa were motivated to make a study regarding student loans to improve access to education.
Tropang South was able to determine the bottlenecks in the application process and as a result, they provided better marketing strategy for Bukas Finance Corp.
The scholars performed classification and clustering algorithms to address the drop off issue of many interested applicants on an online student loan company. In addition, they gathered significant information regarding the personas of the applicants and recommended ways to gather leads for the company.
2. PHILIPPINE CONTEXT
1.2M
Senior High
School
graduates
per year
3 in 5
students
can’t afford
to go to
college
Higher
education
annual cost
up to
₱230k
2
Sources:
https://www.philstar.com/headlines/2018/05/11/1814173/senior-high-school-results-exceeded-expectations-deped
https://businessmirror.com.ph/2015/03/30/3-of-5-high-school-graduates-cant-afford-to-go-to-college/
https://www.ecomparemo.com/info/cost-of-colleges-and-universities-in-the-philippines/
13. Click apply
Loan Info
Address
Education
Accept Initial Offer
Guarantor Info
Guarantor Address
Guarantor
Income
Guarantor
Loan Info
Docum
ents
Submit
Guarantor’s
Other Loans
Personal Info
2000150010005000
14. Click apply
Loan Info
Address
Education
Accept Initial Offer
Guarantor Info
Guarantor Address
Guarantor
Income
Guarantor
Loan Info
Docum
ents
Submit
Guarantor’s
Other Loans
Personal Info
2000150010005000
-30%
-26%
-36%
-22%
-44%
19. BUILDING A
CLASSIFIER
19
○ Cross Val: 79%
○ Accuracy: 82%
○ Precision for True: 84%
○ Precision for False: 64%
...What else can we get from
this classifier?
21. STRATEGY
21
○ Only include observations with
education information-- that’s
912 in total
○ Derive standardized features
based on existing ones
□ ie: How does the requested principal
compare to other applicants with the
same course category?
○ Use KNN clustering algorithm
to segment data and then get
comparison and insights
between groups.
24. USER SEGMENTATION
When exactly are
they leaving?
TARGET
BORROWERS
WINDOW
SHOPPERS
Guarantor’s
basic info
Guarantor’s
financial info
Supporting
documents*
* Variation due to response on required questions and differences in specific document uploaded
25. USER SEGMENTATION
What are their courses?
Hospitality
219
Business and Accountancy
168
Engineering
141
IT
96
Communication Arts
and Media
90
Social Sciences
and Law
84
Health Sciences
64
Education
28
Others*
29
25
QUANTITY OF LEADS PER COURSE CATEGORY
* Mostly foreign relations and criminology
26. USER SEGMENTATION
What are their courses?
26
Hospitality
219
Business and Accountancy
168
Engineering
141
IT
96
Communication Arts
and Media
90
Social Sciences
and Law
84
Education
28
Others*
29
Health Sciences
64
QUALITY OF LEADS PER COURSE CATEGORY
* Mostly foreign relations and criminology
27. User Segmentation
How complete are
their profiles?
TARGET BORROWERS ARE MOSTLY FROM
COURSE CATEGORIES WITH QUALITY LEADS!
THESE ARE COMMUNICATION ARTS AND MEDIA,
HOSPITALITY, ENGINEERING, AND OTHERS.
28. USER SEGMENTATION
Do they have siblings*?
* Assumed siblings because most guarantors are parents
TARGET
BORROWERS
WINDOW
SHOPPERS
Majority are only child and less likely to have
siblings.
Majority are only childs but more likely to have
siblings than target shoppers.
This supports the idea that children from bigger
families are less likely to receive higher education
due to limited income. Hence, the need to look for
other options.
29. USER SEGMENTATION
How much do they need?
TARGET
BORROWERS
WINDOW
SHOPPERS
Their requested principals are more
consistent and closer to the mean per course
category.
Whereas window shoppers have a broader
range of requested principal.
It might be useful to check if they are
graduating or summer class students (having
less units) or with laboratory classes (needing
to pay more fees than usual).
30. USER SEGMENTATION
Can they pay?
TARGET
BORROWERS
WINDOW
SHOPPERS
Guarantors have more capacity to take care of
monthly repayments.
Guarantors are earning ~30k and below.
It is possible that this was a consideration for
the student-- realizing the guarantor may not
be able to pay, hence not finishing their
student loan application.
31. USER SEGMENTATION
Why prioritize target
group?
IN ADDITION TO HAVING MORE COMPLETE
PROFILES, THEY ARE MORE CAPABLE OF
PAYING-- WHICH IS WHAT A LENDING COMPANY
SHOULD PRIORITIZE!
36. RECOMMENDATION
36
On journey optimisation:
Find ways to reduce questions asked
and total pages during the
registration.
Since calling has good conversion
(based on limited applications that
are on hold), it can be repositioned.
Perhaps after initial offer so Bukas.ph can also ask if
they understand what a guarantor is and if their
potential guarantor is nearby. Takes advantage of
foot-in-the-door phenomenon.
On targeting:
Advertise online with specific audience
This works best for schools 5, 6, and 7.
For window shoppers, find more ways to engage them beyond
online posts.
Target courses with high quantity and high quality leads.
Engage with student groups more creatively.
Such as by tapping into student organisations and getting student ambassadors to
raise awareness within the school.
Once traction is improved, explore other channels to engage
guarantors specifically (PTA groups).
Focus campaigns on areas close to partner schools.
Most applicants are from Metro Manila, the same as schools.
37. 37
On being smarter about data:
Gather more info about the students with less effort from them.
○ track where they come from (specific post)
○ explore how many times they log in (measure interest)
○ use location info to see if they’re registering while on school grounds (as a
product of booths) or at elsewhere (confirm if at home and would have more
access to guarantor)
○ explore the possibility of keeping contact with parents who have younger
children who can also apply in the future
Gather more behavioral data
Other lending companies refer to most used apps (with more preference to applicants
who use banking ones than those with a lot of games)
Replicate the same strategy on bigger datasets.
More likely to define specific personas from bigger datasets where combinations of
features are more balanced and fully exhausted.
RECOMMENDATION
39. User
39
Registered
User
Sign Up Initial Loan
Form
Call User
Edit Form
Approve or
Reject
Release
Loan
Data
Loan
Student
Education
Yes
Continue?
No
Completion
Form
Data
Guarantor
School Files
End or contact after 3
months
Call User
OPTIMISE JOURNEY
Suggested User Flow
40. User
Registered
User
Sign Up
Initial Loan
Application Form
Contact User
Edit Form
Approve or
Reject
Release Loan
Data
Loan
Student
Education
Yes
Continue?
No
Completion Form
Data
Guarantor
School Files
end/contact
later
Contact User