SFF GREEN SHOOTS SERIES PRESENTS FINAIM: CREDIT LENDING POWERED BY AI
May 14, 2020
On 14 May, the SFF Green Shoots series presented the first FINAiM (Finance, AI & Marketplace) Forum by MAS. A big thank you to our speakers: Dr Yanming Fang, Director & Head of Credit Risk Model at Ant Financial, Chun Dong Chau, Co-founder of ADVANCE.AI and Shameek Kundu, Chief Data Officer at Standard Chartered Bank - who shared insights on our panel discussion on the application of AI to address lending challenges during the COVID-19 pandemic. The session also featured a keynote by Dr Fang Yanming on applying deep learning to detect fraudulent default on Ant Credit Pay, as well as Dr Li Xuchun, Head of the AI Development Office at MAS on MAS' approach to sustaining a vibrant AI ecosystem in the financial sector.
- Q&A with Speakers: check out our speakers’ responses to your burning questions, below.
- If you missed the session or would like a replay, view this on our YouTube channel here. Do follow us for more videos!
- Want more? Check out upcoming Green Shoots sessions on our SFF website here.
Q&A with Speakers
Question: What is the % gain of fraud detection with AI vs general rule-based fraud detection - from your experience?
Response: Dr Yanming Fang: More than 50%.
Question: In a much publicised incident involving Apple and its new credit card being accused of having a "gender bias", where it pretty much had Apple pointing the finger to its provider, Goldman Sachs, while the latter pointing its finger towards the machine learning algorithm. However, neither party was able to show nor demonstrate how the ML algorithm came up with its decision or recommendation. In such cases, would it benefit all parties from a regulatory as well as compliance perspective, if there is a way or methodology to explain how the ML came out with its decisions or recommendations as opposed to having it all be like a "black-box"?
Response: Dr Li Xuchun: MAS has published the FEAT principles two years ago to address this AI adoption guidance (such as AI transparency). MAS currently is working with an industry consortium to develop solutions (methodology, metrics and open source code) to implement the FEAT principles. We hope the Veritas solution could help disclose the AI explainability, transparency etc.
Question: How can we obtain advantages from AI, if we don't have enough data?
Response: From a tech perspective, you can try semi-supervised machine learning to enhance your small-training-data-set issue. From a business perspective, you can explore alternative data.
Question: Shameek, I agree that we should not be frightened with the term “AI”, what do you suggest is the best way to describe AI assisted tools?
Response: Shameek Kundu: we have been using "alternative lending approaches" in some of our projects.
Question: Hi Chun, we are using ADVANCE.AI for face recognition, EKTP and liveness. What is the model/pattern behind this?
Response: Chun Dong Chau: Basically there are two steps: face recognition, and face comparison. We use different deep learning algorithms to get it as accurate.
Question: Could you please share a few drawbacks/downsides/business risks/reputation risks etc. (if any) of implementing AI/ML?
Response: Chun Dong Chau: Think about AI/ ML as a technique for better prediction. Predictions have been done for decades (eg by normal extrapolation or regression model). I think the scary or uncomfortable part is what are we predicting with AI? And what are we doing with the result of the predictions?
Question: Can you provide examples of how AI can help with SME corporate lending as opposed to invoice lending? Are you seeing good instances of this especially given the lack of consistent and clear data in Asia specifically with regards to assessment of credit risks for these SMEs?
Response: Chun Dong Chau: Lots of SME lending including invoice financing is really about the ability to cross check. AI can speed it up and automate, for example: computer vision can be used to see whether the product is delivered (eliminate execution risk).
Question: I'm more asking about AI for credit scoring and credit assessment vs KYC, verification etc. which is commonly seen in invoice financing. Do you see machine learning or AI in corporate / SME credit assessment?
Response: Chun Dong Chau: In my opinion for SME lending, you need to understand 1 particular vertical/industry very well. For example, Kaggle started with lending doing to eBay sellers only, where they understand those sellers and the industry/ecosystem very well!
Question: How accurate are the AI data points & how is the accuracy measured?
Response: Chun Dong Chau: we have different metrics for that, for example like: auc/gini/ks; or false/true positive rate.
Question: How can we use AI in processes like reject inference when building credit-scoring models?
Response: Chun Dong Chau: reject inference can be reduced by better sampling, meaning sampling on an unbiased population.
Question: It’s a fantastic dashboard product to detect fraud. In your view, what is the biggest challenge in adopting such an AI solution? What’s the key takeaway you would give to those who want to adopt AI solutions for their organisation?
Response: Dr Yanming Fang: You need to have a full test before you actually push anything into production.
Question: Do you think banks/FIs will be keen to work with FinTechs to implement AI/ML to improve credit-lending process?
Response: Chun Dong Chau: Yes they are! Standard chartered is piloting with many start-ups (including ADVANCE.AI).