A few weeks ago, Swish attended the Fintech Forum in Montréal. Our Machine Learning team lead, Corbin Hudson, held a workshop during the forum to discuss the applications of AI in the fin-tech industry.

If you attended the workshop and wanted more, here is a lengthier version with examples and technical explanations. If you did not have the chance to participate in the workshop, this analysis will provide you with everything you should know about AI in the finance industry.

Fintech?

The financial industry follows technological advancement with keen interest. Big banks such as JP Morgan have been early adopters of disruptive technologies like Blockchain.

Artificial Intelligence (AI) is a paradigm-shifting technology that is seamlessly changing the way we live, move, interact with each other, shop. Finance is no exception, and the industry is just starting to peak at the tip of the iceberg.

Fin-tech is the name given to use-cases of cutting-edge technology to the financial industry.

In this article, we go through ten applications of AI and a subdivision of this technology, Machine Learning, in fintech.

AI for Personal Finance and Insurance

Digital Financial Coach / Advisor

Transactional bots are one of the most popular use cases in AI, probably because the range of applications is so broad - across all industries, at several levels.

In finance, transactional bots can be used to offer users finance coaching/advising services. Think of them as digital assistants helping users navigate their finance plans, savings, and spendings. Such service increases user engagement and improves the overall experience of the user with the financial product they are interacting with.

Digital assistants can be built using Natural Language Processing (NLP), a type of machine learning model that can process data in the format of human language. A layer of product recommendation model can be added, allowing the assistant to recommend products/services based on the transactions that occurred between the algorithm and the human user.

An example of this application has been deployed by Sun Life which created a virtual assistant, Ella, to help users for Benefits and Pension by allowing them to stay on top of their insurance plans. The assistant sends users reminders based on user data like “Wellness benefits about to expire” or “Your child will be off benefits soon.”

Digital assistants can also be used in other finance-related scenarios: dividend management, term life renewals, transaction limit approaching or cheque cashed notifications.

Transaction search & visualization

Chatbots can also be used in banking to focus on search tasks. Managers give access to the bot to the users’ transactional data (banking transactions), and it uses NLP to detect the meaning of the request sent by the user (a search query). Requests could be related to balance inquiries, spending habits, general account information and more. The bot then processes the requests and displays the results.

Bank of America uses such a bot (called Erica) as a digital financial assistant for their clients base. The AI-powered bot was quickly adopted - one million users in three months. The bot offers user-friendly transaction search, enabling users to search in their historical data for a specific transaction with a particular merchant, avoiding them the hassle of looking for these in each of their bank statements. The bot also computes total amounts of credit and debt, a task that users had to do by themselves on their calculator.

Client Risk Profile

A critical part of banks and insurance companies’ job is the profiling of clients based on their risk score. AI is an excellent tool for this as it can automate the categorization of clients depending on their risk profile, from low to high.

Building on the categorization work, advisors can decide to associate financial products for each risk profile and offer them to clients in an automated way (product recommendations).

For this use case, classification models such as XGBoost or Artificial Neural Network (ANN) are trained on historical client data and pre-labeling data provided by the advisors, which eliminates data-induced bias.

Underwriting, Pricing & Credit Risk Assessment

Insurance companies offer underwriting services, mainly for loans and investments. An AI-powered model can provide an instantaneous assessment of a client’s credit risk, which then allows advisors to craft the most adapted offer.

Using AI for underwriting services increases the efficiency of the proposals made and improves the client experience as it speeds up the process and turnaround time of such operations.

Manulife, a Canadian financial service group, is the first player in the country to use AI for its underwriting services, making it “faster for many Canadians to buy basic life insurance, a key to addressing the “protection gap” in Canada.”

The insurance company uses a specific AI, Artificial Intelligence Decision Algorithm (AIDA), which is trained on previous underwriting methods & payouts and can have different classifying processes such as large loss payout or price.

The application of this method is not cantoned to insurance; it can also be used on credit scoring for loans.

Automated Claims Processes

The insurance industry as we know it functions on a standard process: clients subscribe insurance, for which they pay. If the customer has a problem (sickness for health insurance, a car accident for automobile insurance, water damage for a housing insurance), she needs to activate her coverage by filing a claim. This process is often lengthy and complicated.

Transactional bots can transform the user experience into a more pleasant process. Enhanced with image recognition, fraud detection, and payout prediction, the entire user journey is upgraded — less friction, fewer costs for the company, less operational tasks (calls, background checks) and fewer errors all in all. The entire process takes less time and becomes a seamless experience for both customers and the insurance company staff.

What the bot does is to take charge of the entire cycle: it walks the customer through the process, step by step, in a conversational format.

It asks for videos or photos of the damage and uploads them to the database. It takes in all the information required for the processing of the claim. The bot can then run the application through a fraud detection method, looking for anomalies and non-compliant data.

It then moves on to the adjustment model where it provides a range of values for payout. Once all data is set, human intervention can be included for auditing purposes. The bot can at this point calculate and propose payout amounts, based on a payout predictor model it has been trained on.

This application is a three in one machine learning solution that holds the potential to relieve a high pain point in the industry.

It is what Lemonade, a New York-based insurance startup, has set as a mission. On the homepage of their website, they ask users to “forget what you know about insurance” clearly announcing the disruption they are bringing to the industry through the use of AI. The company raised USD 180 million since its creation in 2015.

Read more about the applications of AI to the Insurance industry in this analysis by Swish.

Cross-industry

Contract Analyzer

Contract analysis is a repetitive internal task in the finance industry. Managers and advisors can delegate this routine task to a machine learning model.

Optical Character Recognition (OCR) can be used to digitize hard copy documents. An NLP model with layered business logic can then interpret, record, and correct contracts at high speed.

Business logic is a sort of conditional formatting similar to what one can find on Microsoft Excel. Formulas can be added to the model such as “if this box is checked then this one should be blank.” The model can be trained on existing contracts and learn how to behave with such content. In this case, the accuracy of the model’s outcome is remarkably high because of the repetitive nature of contracts.

JP Morgan has harnessed the power of this application of AI, leading to freeing 360,000 hours (yearly) from its employees load in only a few seconds.

These solutions support contract-related analysis, while blockchain-based smart contracts, a paradigm-shifting upgrade to contracts management, are being more widely adopted.

Churn Prediction

Churn (or attrition) rate is a key KPI across all industries and businesses. Companies need to retain clients, and to do so, predicting coming churn can be extremely helpful to take preventive actions.

AI can support managers in this mission by providing a prioritized list of clients who show signs of considering to cancel their policy. The manager can then address this list accordingly: give a higher degree of service or improved offering.

The model, in this case, is based on explainer variables to the churn effect, based on customer behavior data. Explainer variables can be the number of times statements have been downloaded, the occurrence of user reading account policies, unsubscription to newsletters and mailings, and other indicators of churn behavior. By processing consumer data, banks can serve them better by adopting their offering and pricing.

The model used in this case is a classification one trained on historical data of clients who have canceled their policy and others who have remained after considering leaving the institution.

A research paper about customer churn prediction for the banking industry showed the importance of consumer research versus mass marketing for this specific industry: “The mass marketing approach cannot succeed in the diversity of consumer business today. Customer value analysis along with customer churn predictions will help marketing programs target more specific groups of customers.”

Algorithmic Trading - the most advanced ML you’ll never see.

Most applications of algorithmic trading happen behind the closed doors of investment banks or hedge funds.

Trading, very often, comes to analyzing data and making decisions, fast. Machine learning algorithm excels in analyzing data, whatever its size and density. The only prerequisite is to have enough data to train the model, which is what trading has in abundance (market data, current and historical).

The algorithm detects patterns usually difficult to spot by a human, it reacts faster than human traders, and it can execute trades automatically based on the insight derived from the data.

Such a model can be used by a market-maker looking for short-term trade based on quick price movement. Such operations are time-sensitive, and the model provides the speed needed.

An example of this is trading individual stocks versus price movements in the S&P 500 index, which is a known leading indicator (i.e. stocks follow the index). The algorithm takes the price movement from the index and predicts a corresponding move in the individual stock (ex: Apple). The stock is then bought (or sold) immediately with a limit order placed at the prediction level, in hopes the stock reaches that price.

Augmented research tools

In investment finance, a large portion of time is spent doing research. New machine learning models increase the available data around given trade ideas.

Sentiment analysis can be used for due diligence about companies and managers. It allows an analyst to view at a glance the tone/mood of large sets of text data such as news or financial reviews. It can also provide insight into how a manager reflects their company performance.

Satellite Image recognition can give a researcher insight into many real-time data points. Examples of such are parking lot traffic in specific locations (retailer shops, for example) or freighter traffic in the ocean. From this data, the model and the analyst can derive business insights such as the frequency of shopping at specific stores of the retailers mentioned above, the flow of shipments, routes, and so on.

Advanced NLP techniques can help a researcher analyze a company financial reports quickly. Pulling out key topics that are of most interest to the firm.

Other data science techniques can also format and standardize financial statements.

Valuation Models

Valuation models are usually applications for investment and banking in general. The model can quickly calculate the valuation of an asset using data points around the asset and historical examples. These data points are what a human would use to value the asset (ex: the creator of a painting), but the model learns which weights to assign to each data point by using historical data.

This model was traditionally used in real estate where the algorithm can be trained on previous sales transactions. For financial firms, it can use financial analysis data point, market multiples, economic indicators, growth predictions; all to predict the value of company/assets.

Such models are used as an internal tool by investment banking teams.

This article was brought to you by the Machine Learning team at Swish.

We build products at the forefront of machine learning and artificial intelligence to give an edge to businesses. If you want to equip your business with the future of tech, we are here to make it happen. Let’s chat

Do you have questions or comments on this analysis? Head to our Medium page and leave us a note there.