AI has been proven to be pivotal in many industries and sectors. Besides being applied and trained to make search suggestions and speed up writing and video making, AI has also been capitalized on to rid human workers of tedious tasks and has a huge potential to help solve pressing matters such as cancer detection or financial inclusion.
Nam Ma, Head of AI Labs at VNG Corporation joined us from Vietnam to talk about leveraging AI for financial inclusion in developing countries. Read on to catch the highlights of his speech.
4 main obstacles to digital financial services in Southeast Asia
According to Nam, there are 4 main obstacles that Southeast Asian countries are facing in terms of digital financial services:
-
Lack of digital personal identification
-
Unreliable credit rating
-
Regulatory constraints
-
Cash resilience in consumer and merchants
He believed that AI would be useful in solving the first two problems.
Solutions provided by AI
Digital Biometric Identity
This identification is a process to show who is the customer by collecting the customers' credentials and verifying and affirming their credentials. Once we have established their credentials, we can do the authentication, which is to confirm if the customer is whom they claim to be.
Traditionally, human resources are needed for this task, which is time-consuming and errors are likely to occur. Not to mention, the likelihood of identity fraud is higher.
But with an AI-supported application, it takes only one or two minutes for a user to open a bank account anywhere and anytime as well as using Face Match, Deduplication, and Anti-spoofing to prevent unauthorized access. Moreover, we can detect the ID forgery typically overlooked by human reviewers.
Risk Assessment
The traditional risk/credit scoring approaches are not effective in Southeast Asia, according to Nam Ma.
Traditional scoring uses rule-based methods and flags too many transactions with high false positive rates, leading to unnecessary friction for customers in transactions. Additionally, most transactions are not flagged in real time, causing more damage. Then, there is a large number of unofficial or self-employed workers with no historical transaction data to form credit scores in developing countries, resulting in a lack of data when it comes to risk scoring.
Now, developers have learned to combine device data like the OS, the model, or the geolocation with the transaction data and the user data like KYC or a network of frequent recipients to build a deep learning neural network. They also do the network analysis to identify cluster and user actors.
This technology allows banks to alert abnormality, risk profiles of users, and alert new attack patterns. Even if the consumers gave very limited transactions and limited financial data, financing products can still be given to them based on alternative data. This is why Nam believed that AI and machine learning could bring significant value.
3 implementation challenges
Unsurprisingly, real-life implementation of AI in banking is easier said than done. Nam stated that practical AI implementation requires a combination of AI modeling and MLOps, which often encounters three challenges:
Long tail problem
There are a few long-tail problems as claimed by Dr. Nam Ma:
-
ID card templates and regulations are constantly updated with mistakes from issuing body
-
It’s difficult to collect sufficient training data for all cases
He proceeded to present a couple of solutions:
-
AI: Data augmentation to generate training data
-
DataOps: feedback loop with partners to collect sample data and label from actual customer population
Power law/systematic fraud
The problem with this is fraud types constantly evolve to bypass AI checks. The solutions proposed include enhancing Liveness Detection and Face Deduplication to identify spoofing and monitoring abnormal behaviors.
Bias in AI
Little data on ethnic minorities causing inaccuracy in name OCR or Face Match and data labeled according to bias against certain race/sex are a few examples of biases in AI. To solve this, it’s important that teams:
-
stay aware and detect cases of potential bias
-
loop human to review in an objective manner
-
retrain model
While there’s still a long way to go, AI/ML is proving its important role in financial inclusion in developing countries and AI practice within more financial institutions needs to be largely encouraged.