Fintech Lecture 2 : AI and AI in Finance
Fintech is the technology of our time materially affecting finance. Inclusion of new technology, not phone or internet but AI and ML that may materially change the provision of finance.
In Fintech, there are two sides of AI, the consumer-facing side and data analytics. The consumer-facing side is chatbots using NLP saving the cost and time of enterprises to give better services and the other side is Data analytics which uses consumer’s financial and alternate data to give better credit scores and more.
Financial data was always there. It’s the technological advancement that is disrupting the market. Most startups/disrupters made advancements around data or maybe standardizing it. Cloud is a great example that is helping us provide better computation power at a low cost.
Technology stack in Finance :
Money and Ledger -> Internet -> Mobile -> Cloud -> AI/ML -> OpenAPI -> BlockChain
Artificial Intelligence(AI) — Machine or computer mimic human behavior
Machine Learning (ML)— Machine solves a problem without human intervention while learning throughout the journey. Machines learn from data. The more data we feed into problem-solving/pattern recognition algorithms, it gets better.
Deep Learning (DL)— ML and DL both can extract correlations. But DL is a subset of ML powerful enough to find patterns within patterns. It works on multiple layers of connection and pattern recognition, Neural network. It often requires more data and computation power.
Example: Face recognition and autonomous vehicles.
Companies are using ML/DL to identify patterns in consumer finance data to commercialize it better but when it comes to credit scoring they use classic linear regression because it is cheap and easier to explain to regulators and customers.
Regularization and Fraud detection is the most developed financial area in terms of ML/DL. With the help of ML models companies are using user's transactional data along with some alternate data like app usage, browsing history, geolocation, social media data, work history and more to work on cash flow underwriting and other.
Data is facilitated by a lot of startups, getting access to that which the incumbents already have.
Big firms give these financial data to startups, on which they apply their ML model. So in this case even when the data initially came from big firms, startups are charging them money. By providing this data to startups, banks receive the benefit of valuable input which help them serve their customer better and modify their business model.
Even when few fintech startups are free for users, have billions of dollars valuation because of their ability to apply data analytics and monetization on it.
NLP — Taking human language or natural language and processing it down to computer language and vice-versa.
ChatBot — “A chatbot is a software application used to conduct an online chat conversation via text or text-to-speech, in lieu of providing direct contact with a live human agent.” Wikipedia
Chatbots give faster replies and help businesses provide better services. But in the end, not all chatbots solve our problems or answer our questions. That’s why we look for a human because we believe that a human can understand us and answer better. When chatbots will be smart enough to answer all relevant questions, we might not need humans in the end.
Yes, it’s uncomfortable to see an automated system without humans at the end. Even for autonomous vehicles, not the company owners, regulators or public is comfortable enough to see them on the road without humans, YET.
Why Bank’s virtual assistants are not as refined as Alexa or Siri?
- Not many users are using the bank’s virtual assistance so they are not improving it.
- Banks are used to solving queries through in-person conversation and over the phone so they might have fewer data.
- Less human capital and experience building that kind of tool.
Past: Banks — Specialized in all services like payment, credit, trading etc
Present: Fintech — Specialized in one service of the financial world, payment, credit, trading, insurance etc
Future: Specialized in one market of one financial service, loans for MSMEs, student loans, payment services for restaurants etc
As a society, we are not perfect. We have biases and while we are collecting data we are embedding biases too. You might not be aware of them. Even if you have a perfect data set, the protocols themselves might sort of build some biases on top of them. Even in financial data.
Most of the credit card companies and financial firms have been told not to be biased on ethnicity, color, race, gender and more. Financial firms or credit card companies have to answer WHY they have rejected someone’s credit to prevent bias and explainability. Explainability is not just about holding the data but to be able to explain it.
Biases are the reverse of inclusion. More inclusion and fewer biases is Fairness.
Buying second-hand material is acceptable in all races, colors and geographies but buying a second-hand product is a sign of lower-income which can lead to high credit risk and in this case it’s not about any bias related to race, ethnicity, color or gender. In some cases, you might not have bought anything second-hand, you might have just searched for something. This is how AI uses our digital footprint as alternative data in finance. It’s not all about patterns but parameters that lead to decision-making.
Q. If systems are designed to be biased based on gender information and as a user, I do not disclose my gender and still feel the bias, will that be breaking the law?
Another trade-off in fintech is Privacy. Companies can track our shopping patterns, identify changes in lifestyle and alter credit scores.
Imagine a future where some big tech or big financial firm comes to a less developed country and dominates everything and now everywhere we have the same kind of credit scoring system and it’ll be less resilient.
FICO is a credit score standard used in 30+ countries for more than 50 years now. But every few years they release new versions of the same. Recently they launched V10. Maybe by the end of this decade, it’ll be V14 or V15 and that’ll look like an ML/DL type of model.
From actually working on machines to work on plug-n-play tools for ML like tensorflow, we have evolved a lot in just 50 years.
By automating or digitizing the financial world, we can give better service, include more people and can be more targeted. But it comes with the trade-off of Human presence because here we are dealing with money. We have done 80% and need 20% more for human touch. Again, this is the case of 2021, we don’t know what it’ll look like in 2030.
Lecture 1 — Introduction — — — —— — — — Lecture 3 — Open APIs
