Skip to content Skip to navigation

Seminar: Big Data Analytics in FinTech and InsurTech: Industry Trends and Use Cases

Analytics Seminar 31st August 2017 by Dr. Vishnu Nanduri

Big Data Analytics in FinTech and InsurTech: Industry Trends and Use Cases



MOTIVATION FOR FINTECH


What is FinTech

It is essentially the idea of offering better financial services with the aid of technology. With the advent of Blockchain and Bitcoin and the likes, the revolution of Financial Technology has been significantly accelerated. Back in the days, banks traditionally required a physical building and bank tellers to service customers at the counters. This results in inefficiency and long queues especially when there are insufficient bank tellers to attend to the customers. More recently, digital banks such as Atombank and Openbank are emerging as potential alternatives. With the aid of technological advancements, traditional and primitive modes of financial services are changing dramatically as companies seek to go online and digitize their services. FinTech enabled the integration of financial services onto mobile platforms, creating ease and efficiency, allowing people to access these services anytime, anywhere.


Where are we in the FinTech Journey

Based on APAC FinTech Financing Activities report, investment in FinTech was estimated to be $9 billion by the end of July 2016 in the Asia Pacific region with China and Hong Kong leading the way and India close behind them. While banks have significantly invested in FinTech over the past forty years, we still have a long way to go in the FinTech journey and have yet to see the new Facebooks, Amazons, and Googles of the post Bitcoin era. This leaves us with lots of room to grow and invent new financial technologies.


Why is FinTech Important

FinTech has the capability to address and solve the basic needs of customers which mainly revolve around trust, access, and speed. These are the three fundamental concerns of most customers, where customers trust that banks would safeguard their money, allowing them to keep their savings and withdraw money easily when required, offer them credit lines so that they can make purchases easily and speedy transactions and cross-border remittances.

Trust:

In 2008, when Lehman Brothers shuttered, hundreds of thousands of citizens of the United States lost their savings in their 401(k) accounts that they planned to use for retirement. Additionally, the Dow Jones industrial average fell close to 50% in a short span of two-three weeks and subprime mortgage crisis caused numerous homes to be foreclosed. However, this was also the start of the FinTech revolution which had been silently taking place in the background. Bitcoin came into the picture in 2009, shortly after the 2008 Financial Crisis. This started to change people’s perspective on ways of finance and technology as these two are seen in a whole rather than in isolation.

Access:

According to the World Economic Forum report, the figures of those without bank accounts are:

  1. ¾ of the world’s poor
  2. 59% of those in developing countries
  3. 77% of those adults making under $2 per day

This is in stark contrast to statistics in developed countries where one can (relatively) easily open and have bank accounts, gaining access to high-tech financial services such as digital wallets. On the other hand, those without a bank account in developing countries face significant financial challenges especially when they are trying to get credit from banks, pay loans, purchase farming equipment for agriculture or even take up loans to start a business.

Speed:

With technological advancements, monetary transactions such as transferring money from one country to another and conversion of currencies can now be done online, safely, securely and quickly without much hassle. However, such speedy transactions usually come at a cost and sometimes the transaction fees could be relatively high. The remittance costs to developing countries could be higher. Several new FinTech firms are now cropping up to address the cross-border remittance challenge at much lower costs or no costs using the bitcoin blockchain or other blockchain protocols.


How Do Fintech Solutions Work?

The use of Blockchain is one of the many solutions used to address customers’ needs of trust, access, and speed. Below is a short, simplified description of how Blockchain works:

  1. Person A living in Country X wants to send money to Person B living in Country Y. Cryptocurrencies such as Bitcoin and Ethereum can be used.
  2. This transaction is represented online as a Block.
  3. The Block is then broadcasted to every party in the network.
  4. Those interested in earning Bitcoin, would then work to verify the legitimacy of the transaction using a process called mining (essentially solving a hard mathematical problem).
  5. Once the transaction is verified as legitimate, it will be added to the previous Block where it becomes indelible and no further changes can be made to it, ensuring transparency of the transaction made.
  6. The money is then transferred from Person A to Person B.

This method takes cares of concerns of trust, access, and speed simultaneously as there is no need for the presence of a centralized trust institution for this transaction to take place. Without the “middle-man”, this eliminates the problem of trust. This also cuts down the time taken significantly for performing a transaction.

Another example would be M-Pesa, a mobile phone-based money transfer, financing and microfinancing service launched in 2007 by Vodafone for Safaricom which is used in several countries in Africa. Currently, the company has approximately 18 million active users in Kenya and it was estimated that there were 614 million M Pesa transactions processed during December 2016 and 6 billion transactions over 2016. With more than 287,400 agents worldwide providing services to its users, this helped lift around 2% of Kenyan households out of extreme poverty by providing them with access to mobile money services and bank credits.


What Role Does Analytics and AI Play?

Banking – Operational Efficiency

  • Applications
    1. Opening Accounts
    2. Identity and access
    3. Routine transactions
    4. Check deposits
  • ML Algorithms
    1. Deep learning
    2. Natural language understanding
    3. Decision trees
    4. Random forests

With the use of Artificial Intelligence, the possibility of one opening a bank account using a bot is highly likely and in some cases, already happening. A proposed idea involves deploying a bot to obtain the user’s credentials, user’s particulars, user’s photograph and match it with the passport photograph for verification. If these are verified and valid, the bot will automatically open a bank account for the user, saving the user time of going down to a physical bank to open an account. The use of AI technology is becoming a common sight now and companies such as Amazon Web Services and Microsoft Azure are providing web services to allow users / developers to be able to easily deploy and integrate bots into their offerings. Mathematical and statistical algorithms such as Deep-Learning, machine learning, natural language understanding underpin these fundamental advances being made in facial, image, voice, and text recognition. One such example is Azure, providing users the service to upload photos onto it and using deep-learning and other methods, the service can understand, comprehend, and verbalize the content of the picture before generating a caption for the users.

Underwriting

  • Applications
    1. Risk Profiling
    2. Quote generation
    3. Automated premium adjustments
    4. Pay by how you live – Companies producing wearables to track human activities tie up with insurance companies to profile their customers using a set of algorithms before deciding on the premium that they should be charged. To put it simply, this translates to healthier people paying less for insurance premiums as compared to those leading unhealthy lifestyles.
  • ML Algorithms
    1. Natural language understanding
    2. Decision trees
    3. Random forests
    4. Support vector machines
    5. Clustering

Machine learning, data science, and analytics are commonly used in underwriting. Analytics is used in risk profiling, enabling insurance companies to adjust premiums automatically according to each of its customers. An example would be automated car premium adjustments in the United States where some insurance companies require its customers to plug a USB device under their cars’ steering wheels which would record telemetry data of the car such as rates of acceleration, rates of deceleration, and number of hard-brakes and soft-brakes etc. Putting this data through an algorithm, insurance companies would be able to analyze, score and profile their customers accordingly, based on driving habits and behavior. These “pay-by-how-you-drive” premiums allow safe drivers to pay lower premiums than those who are scored as unsafe drivers, making car premiums fairer and more tailored to an individual’s driving behavior.

Customer Acquisition and Retention

  • Applications
    1. Social media mining
    2. Next best offers
    3. Geolocation based services
  • ML Algorithms
    1. Text analytics
    2. Clustering
    3. Recommendation engine

Tapping on social media, companies can find out what their consumers are liking on Facebook, whether they are Tweeting about the company and what kind of groups are their customers are linked to. An example of monetizing location based data is by using foursquare data, a check-in service, customers can be given location-based offers. Geo-location services allow businesses to keep existing customers happy by providing them with offers and assurance that they are well-taken care of. This requires machine learning and text analytics to help companies better understand which group of customers should be clustered together for better organization and allowing easy and targeted dissemination of offers / discounts to their customers.

Fraud detection

  • Applications
    1. Anomaly detection
    2. Outlier detection
  • ML Algorithms
    1. One-class support vector machine
    2. Synthetic minority oversampling – Over sampling the less represented class and build a model based on that for better prediction
    3. Logistic regression – Binary based to detect fraudulent transactions

Fraud detection is of key importance in FinTech, enabling companies to understand when and where frauds might occur, flagging them out in real-time and taking the necessary steps to prevent their occurrence. An example would be using a one-class support vector machine to detect anomalies in the data sets. Since fraudulent transactions makes up only a small proportion of all transactions and they come in a myriad of forms, detecting them manually is extremely difficult, hence using a machine learning mechanism would help address the issue of class imbalance, making it a more effective and efficient method.

Blockchain Analytics

  • Applications
    1. Legitimate V/S nefarious transactions
    2. Pattern of transfer of funds
    3. Identifying entities of interest
  • Approaches
    1. Network and graph mining
    2. Real-time analysis and visualization

The number of monetary transactions on the bitcoin blockchain and other blockchains have increased significantly. While Blockchain allows consumers to use cryptocurrencies, making their identity details secret, it is crucial that financial service providers are still able to track down the origins and destinations and paths of these transactions. Companies could use real-time analysis and visualization to determine if the person on the receiving end is a good entity based on historical data and transactions. There are some companies addressing such important questions already.


Start-ups That Addresses Fintech Challenges

Trov

An Australian insurance company that provides on-demand insurance. Customers can take a photo of an object that they require to be insured and choose the period that they want it to be insured for. This benefits customers who are travelling and may save on insurance premiums since they do not require the entire house to be insured while they are away. Selective insurance provides consumers increased ease and flexibility, allowing them to only insure items they deem valuable.

21.CO

A start-up from California, functioning like LinkedIn, allowing people to reach out to others for expertise and pay for the services provided in Bitcoins.

TALA.CO

A start-up from Africa that helps solve the problem of access to credit. It uses non-traditional data sources to determine appropriate credit scores for customers using advanced analytics algorithms. Using their mobile app, the company extracts information of its customers, analyses it using proprietary algorithms, and give them a credit score, which is then used to to determine the amount of grant or loan issued. It is an example of a very successful startup using analytics to solve a pressing FinTech challenge.

Chainanalysis.com

Blockchain analytics which gives real-time visualisation of transactions to help determine potential fraudulent transactions.


Executing Analytics Project

CRISP-DM (Cross Industry Standard Process for Data Mining)

  1. Data
  2. Build business understanding --> Data understanding (See if data is sensible and filter out junk data) and vice versa
  3. Data preparation – 60% - 70% of one’s focus should be in these 3 steps
  4. Modelling
  5. Evaluation (If the model is not good enough, return to business understanding and repeat the process)
  6. Deployment

Mythbusters

Myth 1: AI will take all our jobs

Truth: Sky Net is not coming anytime soon. While some jobs will be automated, they are merely for efficiency purpose, we still have a long way in technological advancements and making most jobs obsolete in the next decade is not that likely to happen.

Myth 2: AI is possible only for the Googles and Facebooks

Truth: Any company can do AI if the company has the data and knows how to fully utilize it.

Myth 3: Analytics is magic

Truth: It is math and statistics and can’t make pigs fly. Behind analytics, it is just data and math and statistics must be applied to it to drive analytics.

Myth 4: Analytics, AI, Machine Learning, Deep learning are mutually exclusive

Truth: They are just a philosophy. Machine learning is a subset of data science, which is a subset of AI and AI is a subset of analytics. To put it simply, they are not exclusive and should not be viewed in isolation. Analytics is essentially better decision making using data and mathematical and statistical techniques.


How to Succeed In The Analytics / Data Science / AI Industry

  • Pick up at least 2 skills: SQL / Python / R python
  • Pick up at least 1 skill: PYSPARK / h20.ai / manhout
  • Pick up at least 1 skill (Deep learning): TensorFlow / Caffe / mxnet / PYTORCH / theano

While having these skills are important the most crucial skills are still written, communication and presentation skills to effectively communicate with business and other stakeholders.


How to Hire Analytics and Data Science Professionals / How to Get Hired?

  • Attitude to Skill Ratio – 60:40
  • Hire for potential
  • Be Nice

Top Data Science and Machine Learning Books

  1. An introduction to Statistical learning
  2. Mining Massive Datasets
  3. Deep Learning

 

DISCLAIMER: The comments made during the talk are STRICTLY for educational purposes and should not be construed as financial or investment advice and must not be commercially distributed. Also, the opinions expressed are entirely those of the speaker and not those of his current or any former employers. The statistics presented are obtained from openly available sources in the internet and must not be attributed to the speaker.

Last updated on 30 Oct 2017 .