Get onboarded - KYC & Credit scoring

Onboarding Process

Onboarding Customer - Mobile app

Onboarding - Customer

1. eKYC Products:

>IQA Image quality assessment - uploading of ID, IQA opens, guides client to make sure that image is ok, take photo, pass to ID OCR. This checks whether the image provided has sufficient image quality requirements. Intended to be used by the App SDK when scanning the customer’s ID before being sent to the backend for OCR & assessment. The scanned image is sent to the AAI, and can later be accessed by returned IDVID reference. https://doc.advance.ai/sg/global_image_quality_check.html

> ID forgery - physical, face value - checks whether the ID on the provided image is forged or not: https://doc.advance.ai/sg/global_id_forgery.html#global-id-forgery

> ID OCR - extract details on the ID - intended to be used during the onboarding to pre-fill the customer data with extracted fields from the ID card. Different ID = different pre-populated details. Refer to https://doc.advance.ai/ph/#ocr-lite

Acceptable IDs as of Sept 20, 2022. To be listed in the app as is:

1. UMID Old/New Version

2. Driver’s License

3. PH National ID

4. Passport

5. PRC ID

Under “SHOW MORE”:

6. Premium Postal ID - 1st version may not have a very good accuracy due  to limited no. of samples. Continuously collecting samples to improve the accuracy.

7. PAGIBIG ID

8. Voters ID

9. SSS ID

10. TIN

> Liveness Detection - open camera, instruct to blink eyes etc, selfie image taken, with score liveness %: https://doc.advance.ai/ph/#liveness-detection

• If the user’s liveness score is more than 50, that is regarded as normal behavior, and the suggestion is pass.

• If the user’s liveness score is less than 50, that is regarded as abnormal behavior, and the suggestion is manual check.

> Face comparison - match two different set of images (use cases: Onboarding and Step up verification - eg higher transaction, go for another selfie/vKYC): https://doc.advance.ai/ph/#face-comparison

• If the faces similarity is more than 70, the suggestion is pass.

• If the faces similarity is from 55 to 70, the suggestion is manual check.

• If the faces similarity is less than 55, the suggestion is rejection or using other methods to verify the user’s identity.

> Face Search - feed images to Face search, database, input face image code - insert face (image, reference id, id number) then search, FS return is Hit or no hit, if hit, details of face image reference ID and id number: https://doc.advance.ai/ph/#face-search

2. Device Fingerprint: FS: Device Fingerprint

• with Private API in OSP

• get DF ID and store for DFP Scoring in the OSP Workflow (https://safibank.atlassian.net/wiki/spaces/ITArch/pages/129630780 )

• same device, 2 numbers - reject

3. Application Form: FS: Customer Application Form

• Based on FIGMA:

Email

Phone number

ID (details in ID) - name, ID number, birthday, address, etc

Gender

Date of birth - for loans 21 years old to 59+ (should not exceed 60 at loan maturity)

Place of birth

Nationality - Filipino citizen only

Location - vs device GPS, should not be within the Negative locations set by the bank

Complete address - vs location in GPS

Account purpose -

• savings/investment,

• salary,

• business,

• other

Source of income - if not salaried - no initial loan available

• salary/pension,

• investment,

• inheritance,

• remittance,

• other

Occupation - high risk / low risk, if unemployed - no initial loan available

• Private company employee,

• Part-time employee,

• Government employee,

• Business owner,

• Student,

• Freelancer,

• Retired,

• Unemployed

Industry - high risk / low risk

• Wholesale and Retail,

• Agriculture,

• Construction,

• Production and Manufacturing,

• Transportation and Warehousing,

• Military or Police,

• Self-employed,

• Finance, Telco, Communication, IT,

• BPO (Call Center),

• Real Estate & Renting,

• Electricity, Gas and Water Distribution

• Education

• Health and Social Work

• Hospitality / Hotels & Restaurants

• Media

• Government

• Others

Monthly income - set threshold

Employment duration - less than 6 months - no initial loan available

• Less than 6 months

• More than 6 months, less than 1 year

• 1-3 years

• 4-5 years

• More than 5 years

Onboarding Workflow - OSP decision engine

Onboarding - One Stop Platform

Below are the several available APIs in the OSP. Not yet listed in chronological order.

1. Identity Check (CIBI) API - FS: CIBI Identity Check ; API doc https://doc.advance.ai/ph/identity_check.html#identity-check

INPUT:

RESPONSE:

DECISIONS

Pass

Not Pass = request for other IDs, Identity Check only recognizes TIN, SSS, and GSIS (all three IDs are prone to fraud)

2. AML Screening API - FS: Anti-Money Laundering (AML) ; API doc https://doc.advance.ai/aml_watchlist_search.html#aml-watchlist-search , https://doc.advance.ai/aml_watchlist_profile.html

INPUT:

RESPONSE:

DECISIONS:

AML Screening Result is EMPTY

AML Screening Result is NOT EMPTY

3. AML Screening and Monitoring API - API docs https://doc.advance.ai/aml_watchlist_search.html#aml-watchlist-search , https://doc.advance.ai/aml_watchlist_profile.html , https://doc.advance.ai/aml_watchlist_monitoring.html#create-a-new-item

INPUT:

RESPONSE:

DECISIONS:

AML Screening Result is EMPTY

AML Screening Result is NOT EMPTY

4. Email Detection API - The service gets the most comprehensive social media lookup available with data from 20+ social media sites, including user avatar, bio and profile info. It also confirms that the address is valid, gets detailed domain whose info and data breach lookup.

FS: Seon (Email Detection) ; API doc https://doc.advance.ai/ph/email_detection.html#email-detection

INPUT:

RESPONSE:

DECISIONS:

Email status is equal to VALID

Email status is equal to INVALID

5. PH Tele-Status Check - The service verifies current status of the mobile number in real-time, which helps prevent fake applications. It can be used for risk management of pre-loan and post-loan, and improves the efficiency of collection.

FS: Tele Status Check ; API doc https://doc.advance.ai/ph/tele_status_check.html

INPUT:

RESPONSE:

DECISIONS:

Reachable

Unreachable

Disconnect

Invalid Number

6. Social Media Check API - The service verifies whether the mobile number is registered with WhatsApp/Facebook/Instagram/e-commerce. It can help identify if the mobile number is actively used and if the applicant shops online regularly, which will help institutions identify potential fraud risk.

FS: Seon (Social Media Check) ; API doc: https://doc.advance.ai/ph/social_media_detection.html

INPUT:

RESPONSE:

accountDetails:

Notes: - If a service fails, registered return null。

DECISIONS:

Facebook, WhatsApp, Telegram, Viber, Instagram

Blocked, Not registered, Registered = ??

7. Blacklist API - The blacklist continues to accumulate data obtained from actual loans and loan behavior tracking through the industry's joint prevention and control and big data laboratory mechanism, mainly covering the overdue data of microfinance 15d+, and returns whether it is a hit.

The service will check the ID number, DOB, name and mobile number of the applicant against Advance AI’s (Atome) blacklist database which has millions of records. Based on the result, financial institutions can know the applicant better, improve their awareness of fraudulent and malicious behaviour, and make better credit decision.

• clients with 15 dpd+, no buckets classification, no updating/removal of name in the list if paid, HIT or no HIT only (ID Number, Phone Number, Name and DOB), no access with the list

https://doc.advance.ai/ph/blacklist.html

INPUT:

Note : md5PhoneNumber or phoneNumber, at least one of which cannot be null

NOTE

• Currently, only the idType of SSS, GSIS, PHILHEALTH_ID, PRC, UMID, TIN, DRIVER_LICENSE, VOTERS_ID, PASSPORT, POSTAL_ID_PREMIUM, STUDENT are supported, if your idType is not in this supported list, please set the idtype as OTHERS

• SSS：It has to be a number.

• GSIS：It has to be a number.

• PHILHEALTH_ID：It has to be a number.

• PRC：It has to be a number.

• UMID：Must be a 12 digit number, fill 0 in front of digits if not enough

• TIN：Must be a 12 digit number, fill 0 after digits if not enough

• DRIVER_LICENSE：The first letter is uppercase and the rest are pure numbers

• VOTERS_ID：No VIN, only letters and numbers, keep letters uppercase

• PASSPORT：Only letters and numbers, keep letters uppercase

• POSTAL_ID_PREMIUM：There is no “PRN” in the first place, the last digit is not english, only letters and numbers, keep letters uppercase

• STUDENT ：Only letters and numbers, keep letters uppercase

• OTHERS：Only letters and numbers, keep letters uppercase

RESPONSE:

DECISION:

hitIdNumber true OR hitPhoneNumber true OR hitNameAndBirthday true

Else

** Final HIT response should not constitute outright reject

8. Telco-Behavior Score - This telco scores are statistical models built on various Mobile Network Operator(MNO) data sources – Data and Voice usage, Top-up Patterns, Location and Device Data and many more. Derived by the most advanced machine learning algorithms, the scores have significant predictive power and can be used as either standalone models or in combination with existing internal Social-Demographic and/or Credit Bureau models.

FS: Telco Behaviour Score (FinScore) ; API doc https://doc.advance.ai/ph/telco_behavior_score.html#telco-behavior-score

INPUT:

RESPONSE:

Notes: Values of city,cphoneNumber and telcoBehaviors will always be null if phoneOperator is Globe

DECISION:

• User Phone Number Risklevel = is equal to / Higher than / Higher than or equal to / Lower than or equal to / Lower than = Highest / High / Medium / Low / Lowest / Inactive

• Customer Loyalty = is equal to / is not equal to = New SIM(1-365 days) / Young SIM( 1-2 years) / Matured (2-3 years) / Established (3-5 years) / Loyal (5+ years)

• Subscription Type = is equal to / is not equal to = Smart Prepaid / Smart Postpaid / Sun Cellular Prepaid / Sun Cellular Postpaid / Talk'n'Text / Broadband Device

• SMS Usage = is equal to / is not equal to = Lowest(<20% users) / Low( 20% < users< 40%) / Mid( 40% < users< 60%) / High( users> 60%) / Highest( Top 20% users) / Newly Registered/Fake Number

• Top Up = is equal to / is not equal to = Lowest(<20% users) / Low( 20% < users< 40%) / Mid( 40% < users< 60%) / High( users> 60%) / Highest( Top 20% users) / Newly Registered/Fake Number

• Mobile Data Usage = is equal to / is not equal to = Lowest(<20% users) / Low( 20% < users< 40%) / Mid( 40% < users< 60%) / High( users> 60%) / Highest( Top 20% users) / Newly Registered/Fake Number

• Avg Daily Activities = is equal to / is not equal to = Lowest(<20% users) / Low( 20% < users< 40%) / Mid( 40% < users< 60%) / High( users> 60%) / Highest( Top 20% users) / Newly Registered/Fake Number

• Contacts Messaging Frequency Rank = is equal to = In Top10 / Not in Top10 / No Rank

• Contacts Calling Frequency Rank = is equal to = In Top10 / Not in Top10 / No Rank

• Contacts Airtime Amount Rank = is equal to = In Top10 / Not in Top10 / No Rank

• City Stayed Period = is equal to / is not equal to = Most City in All Time / Most City in Night and Weekend / Most City in Day and Weekend / Most City in Day and Night / Most City in Weekend / Most City in Night / Most City in Day

** Scoring threshold

8. High-Level Risk Check

9. Customer Deduplication Deduplicate customers (Onboarding step 2)

• microservice - with Private API in OSP - to prevent fraud or the creation of multiple customer instances for the same person

DECISION:

• FULL_NAME duplicates = NOT EMPTY / IS EMPTY

• FULL_NAME error = NOT EMPTY / IS EMPTY

• BIRTHDAY duplicates = NOT EMPTY / IS EMPTY

• BIRTHDAY error = NOT EMPTY / IS EMPTY

• ID_CARD duplicates = NOT EMPTY / IS EMPTY

• ID_CARD error = NOT EMPTY / IS EMPTY

• DEVICE_FINGERPRINT duplicates = NOT EMPTY / IS EMPTY

• DEVICE_FINGERPRINT error = NOT EMPTY / IS EMPTY

• FACE_SEARCH duplicates = NOT EMPTY / IS EMPTY

• FACE_SEARCH error = NOT EMPTY / IS EMPTY

• NETWORK duplicates = NOT EMPTY / IS EMPTY

• NETWORK error = NOT EMPTY / IS EMPTY

• EMAIL duplicates = NOT EMPTY / IS EMPTY

• EMAIL error = NOT EMPTY / IS EMPTY

• PHONE_NUMBER duplicates = NOT EMPTY / IS EMPTY

• PHONE_NUMBER error = NOT EMPTY / IS EMPTY

• status = is equal to / is not equal to / is contained in / is not contained in = UNIQUE / POSSIBLE_DUPLICATE / DUPLICATE

Scoring Services

The features here are tied to the various scoring services (e.g. scorecards), that will be used to provide a score and other parameters to be used in other services (e.g. loans product recommendation). In short, for the various scores generated , if it should be used as a feature, then it will be stored in its respective entity_type_id in the score_fs featurestore

1. Scorecard for Device FP - https://safibank.atlassian.net/wiki/spaces/ITArch/pages/129630780

2. Scorecard for SEON - Scorecard to asses new customer based on the SEON features. https://safibank.atlassian.net/wiki/spaces/ITArch/pages/130843063

________________________________________________________________________________________

** Action items:

• Make a table based on the ff suggestions from Anti-Fraud User e75ec :

** Can accept ID nos. 6-11 - allow opening an account without overdraft, but client will need to submit other valid ID via vKYC (top 1-5) if he/she will avail of overdraft/loan product.

** IDs w/o address (Passport/PRC/PAGIBIG) - allow opening an account with overdraft, but client will need to submit POB/other ID w/ address prior availment of PL

** To be discussed further.

• New Driver’s license included in the OCR accuracy testing
• IQA/OCR trial allowed limit/threshold, result if IDs are invalid/blurred

Other comments/suggestions/action items from User e75ec :

• Existing number log in/sign up - generic prompt error message
• Atome’s blacklist database - get from Ronel

Others:

• Blacklist/Watchlist of SAFI https://safibank.atlassian.net/wiki/spaces/ITArch/pages/77070551
• Risk rating from Compliance c/o Atty. Ruth
• Credit risk guidelines for Overdraft and PL. Income documents requirement?
• As per Pavel: High level strategy for Overdraft: generally if we have not much of idea or even the risk score (finscore, other sources) not looking good they will get 200php - minimum, the very good clients will get the higher end, and the rest something between