The future of finance

Machine learning in finance: From buzzword to mainstream

Ilya Kislenko
machine learning
© Shutterstock / wanpatsorn

The financial sector is a late adopter of machine learning. Top applications include fraud detection, customer care, and risk hedging. See how the mass adoption of machine learning can apply even to the most conservative sectors.

Finance is one of the most conservative sectors and one of the latest technology adopters for a good reason: it requires stability, predictability, and risk hedging.

Even if there are significant opportunities presented by technology, financial companies will be lagging behind because they should wait for a technology to enter its maturity stage as well as for somebody else to test it first and absorb associated risks. However, once a technology is stable enough and has gained mass adoption among businesses, it becomes appealing for the finance sector. This is the case of machine learning (ML), which has ceased to be just a buzz word now.

SEE ALSO: Meet Manifold: Uber’s machine learning model debugging tool goes open source

Machine learning uses statistics to recognize patterns and associate them with entities or cases. It does so by analyzing thousands of labeled examples and looking for similarities and differences. The more diverse data points it uses to train, the better the algorithm becomes and the more false positives it can prevent. It’s already used by social media platforms as well as stock market trading software. Here are a few ways, as pinpointed by machine learning consultants, in which it can boost finance.

Fraud detection

Putting in place the right mechanisms to prevent fraud makes clients feel safer, protects the reputation of financial institutions, and demonstrates care for the customers. It relies heavily on real-time interaction and a history of what is considered normal for every client. Any deviation is carefully analyzed.

ML looks at transaction patterns and identifies those which don’t follow the business-as-usual trend. This includes but is not limited to transferring large amounts of money, accessing an account from a different country or device. The advantage is that the algorithm learns how each user behaves, for example, a frequent traveler will not fall under the location red flag, whereas a client who always uses their home desktop for banking transactions will trigger the system if they want to send over a substantial sum while abroad.

The same algorithms can be used by authorities to fight money laundering. This works by tracing transaction routes and identifying cycles of interlinked companies.

Risk hedging and investment strategies

Financial companies create added value by finding the right balance between risk and safe bets in their portfolios. In the context of economic volatility, safe means foreseeable, not flat. ML algorithms can help predict future trends and create more solid portfolios.

When it comes to risk management, ML models are currently used to compute credit scores for credit applicants. In the future, we can expect to see more factors included in the mix along with the payment history. China is already experimenting with a social credit score system, which has numerous ethical concerns and implications.

Algorithmic (fast) trading

The days when stockbrokers were shouting on the floor can be a thing of the past. Computers can successfully replace their bids. Machine learning can help fund managers automate most trading processes. This option also has the advantage of gradual order release to the market, called “child orders,” to avoid the shock of a larger order.

Better client service

Although it seems counter-intuitive, technology can personalize client experience, as long as it doesn’t try to solve the problems for which a human operator would be better.

Rule-based chatbots are most of the time too simple and offer a frustrating experience. However, machine learning can turn a chat window into a helpful selling tool. Chatbots built with this technology can “think” and compose answers from bits of information they put together instead of serving canned responses.

Training such chatbots requires hundreds of hours of customer service logs, but once the algorithm learns, it can successfully replace a lot of call canter agents without losing its temper during long working hours.

Robo-advisors: personal finance, investments and pension plans

One of the most challenging aspects of personal finance is budgeting and staying within spending limits. Another challenge is that most of the time people aren’t able to tell with certainty how exactly they spent their money.

Wallet management software based on machine learning can help users track their spending habits and make better choices. By automatically classifying shopping into categories, such software can set triggers whenever the user is getting close to the spending limit.

Investments and pension plans can also be managed better with the help of machine learning. An automatic algorithm can allocate your savings in a portfolio matching your risk appetite. Once the initial set-up is complete, it can periodically re-evaluate the assets and make the necessary changes to keep it profitable by rebalancing it.

Insurance and underwriting

These are some of the areas that can benefit from machine learning automation the most. The only setback is that for smaller companies there is not enough training data. Still, national companies can use the anonymized records of millions of users to evaluate the associated risks and create valuable models.

The initial costs and time to create well-calibrated algorithms could be quickly recovered when these are implemented as they are poised to save millions of dollars.

SEE ALSO: 3 out of 5 developers contribute to open source, new Slashdata report claims

Final thoughts and future directions

Depending on the application, machine learning can be supervised or unsupervised. Supervised learning is better suited for tagged data that requires precise inputs and known outputs, while unsupervised models are better for exploratory research. There is also the opportunity to combine the two methods in a semi-supervised approach. This can uncover new patterns in data and provide insights into risks.

When it comes to machine learning adoption in the financial sector, there is a paradox: the most eager companies to try it out are typically fintech startups. Still, the most prepared ones in terms of training data ownership are large banks and insurance companies.


Ilya Kislenko

Ilya Kislenko is the Head of Mobile Department at Mbicycle. With more than 5 years of proven track record in Android development, he’s also worked on large teams in different roles, such as Senior Android Developer, Team Leader and Technical Consultant. Clean code and professional growth are two of his biggest priorities.

Inline Feedbacks
View all comments