Real world Machine Learning in Fintech

Marc Andreessen made the famous comment — “Software is eating the world” almost 7 years ago. With the increase in the structured and unstructured data, coupled with the power of cloud computing, one can relate to the comment made by Marc. Although the use of analytics and cloud has been on the rise across many sectors, the financial services industry has seen this trend of disruptive software more than any sector in the last decade. This has led to a rise in new products and offerings in the Fintech sector. Goldman Sachs made a prediction that Fintech could displace $4.7 trillion in revenue for financial service firms. The world economic forum released a framework against six functions of financial services and eleven clusters of innovation within the Fintech industry in 2011, which helps us to see the diverse areas of change visually.

The info-graphic shows a broad range of business scenarios. For the scope of this article, we will look at the three financial functions and the underlying Machine Learning architectures used for solving business problems.

Insurance

Traditionally, Insurance industry was known to be a sluggish industry where there are inherent delays whether it is getting a new policy (because of the manual complex underwriting process) or the manual delays associated with filing a claim. But, the customer of today demands more from this Fintech industry. Consumer life styles have evolved with rise in the use of mobile devices, smart homes, connected cars etc. Consumers want the ability to transact without extensive paperwork. This has resulted in new digital business models such as Lemonade which promises zero paper work and no hassles while managing insurance.

Here are some examples where Deep Learning Architectures as applicable in Insurance.

Image Recognition using Deep Learning

Convolutional Neural Networks have been used for Image Recognition and have been proven to be very successful as they have evolved over the last few years. The below scenarios are relevant in our immediate topic.

• Home Insurance: Use of drone images to capture pictures of home roofs and performing Image Classification on them to determine their age and condition (either during policy creation or for a claim say during thunderstorms)
• Health Insurance: Using selfie images to determine BMI classification.
• Auto Insurance: After a car accident, the adjuster takes pictures of license plate, VIN, odometer in addition to the pictures of the damaged body area on the car. These values can be entered through a Machine Learning powered app vs entering them manually to prevent manual errors.

Time series Data using Deep Learning

Long Short-Term Memory (LSTM) networks are used to learn important current and past behaviors that have an impact in making future predictions. This neural network is useful for making predictions on a stream of input data as we notice in the two examples listed below.

• Auto Insurance: There is a lot of data collected via smart cars and mobile apps created to track driver behavior. This time series data can be used in a LSTM network to determine competitive insurance premiums.
• Health Insurance: The use of wearable devices and clothes has been on the rise in the last few years. This time series data is a good candidate to be used in a LSTM network to analyze and predict health patterns in an individual.

Lending

Credit Approval using Natural Language Processing

The appetite for risk for financial institutes in lending business has changed substantially over the past ten years. The change in regulations due to the financial crisis on 2008 has triggered a far more rigid look at the way lending is carried out. There are several factors that determine a bank’s tolerance for lending. The 5Cs of credit — Character, Capacity, Capital, Collateral, and Conditions tell us the breadth and depth of evaluation a credit application can be subjected to before being approved. In addition to reviewing the financial health by reviewing the credit score, assets, and liabilities, Deep Learning technique such as Natural Language Processing (NLP) is being used to get a 360-degree view of a customer by reviewing publicly available data that is either structured or mostly unstructured (social media feed, images, video, audio, public records etc).

Investment Management

Personal Finance using NLP

NLP is extensively used within Chat Bots used for Personal Finance. The bot has access to the configured accounts and provides feedback and input on spending habits and opportunities for savings. There are many options currently available, but here is a snapshot from the Penny app.

Machine Learning helps users manage user’s personal finance by using supervised learning algorithms that look at the past transactions and user inputs. It is equivalent to having an analyst working on your behalf managing your finances.

Conclusion

In this article, we have seen how CNN, LSTM, and NLP are being used to solve practical problems in the Fintech industry. In the subsequent articles we will look at Crypto currency and it’s impact on Fintech.