Introduction

Machine learning, big data, and artificial intelligence applications have become a part of our daily lives. We start the working day at the office by reading articles selected specifically for us based on our search history and our social media activity. Each article users come across is suggested to them after a complex algorithmic analysis of their activity. In other words, such recommendations are offered due to the use of artificial intelligence, machine learning, and big data. These technologies provide the best possible results based on the quantity and quality of data available.

These intelligent machine-learning algorithms are also used as a helpful tool in Fintech. They can be used to achieve several purposes in the field, from shortening timeframes and taking crucial decisions, to detecting fraud or even helping customers through a simple chatbot.

Nowadays, banks actively use AI in their daily office work, helping employees make intelligent decisions. The typical areas involving machine learning are:

• Customer segmentation
• Debt collection
• Customer support
• Fraud detection
• Credit scoring

On the other hand, customers can interact with Online Banking solutions through a friendly AI chatbot that will automatically answer FAQs with no immediate human interaction.

In this article, we will review some opportunities and the most critical problems within the financial technology industry —without delving deep into coding or logic arguments. This article is a topic starter to reflect on the possible solutions for a fintech startup.

Areas where we can usually implement AI

Customer segmentation

Thanks to customer segmentation, companies can have a better understanding of people’s activities, behaviors, and needs, and this enables them to group people according to their interests. Such segmentation is possible after a thorough analysis of a large amount of data, which can then be studied from different perspectives. 

Overwhelmed customers

• Without ML
  Overwhelmed customers who are bombarded with too many irrelevant ads that do not match their preferences will simply start ignoring all of them. Consequently, clients will lose potential new business opportunities.
• With ML
  Targeted promo choices will be offered to customers based on their specific interests.

Real-life use case

By using Machine Learning to analyze customers’ social networks and internet activity,  we can determine a customer’s profile. For instance, we can establish that a customer is a “techie” because they usually search for smart home appliances with Wi-Fi (IoT – Internet of Things) or a “sports lover,” since they usually search for and buy good running shoes.

Debt collection

How can we make the debt collection process less stressful for customers and agents?

False Positives

• Without ML
  Sometimes a long-standing customer is burdened with a debt collection process. Said customer will probably get upset and disappointed if treated as a common defaulter.
• With ML
  Evaluate each customer and group them with others with similar behaviors, like good and bad customers. By doing so, companies can prevent stressed collectors from harassing good customers and reduce the load on the collection team, making the process more efficient.

Consuming a lot of resources

• Without ML
  Human resources professionals often become debt collectors, making phone calls to each customer. Such practice is time and money-consuming.
• With ML
  Using machine learning with the previous categorization helps companies improve the debt collecting performance by sending debt collectors only to actual defaulters.

Real-life use case

An e-mail notification will be enough for good customers who have incurred a debt for the first time. Thus, the collector will be only appointed to actual debtors.

Customer support

Give better support with less work and time invested.

Normal questions

• Without ML
  Customer support agents usually have to answer the same simple questions over and over again for several customers; therefore, their tasks are repetitive. 
• With ML
  Chatbots can automatically answer common questions and, therefore, agents can focus on those things which really require human interaction.  

Complete satisfaction

• Without ML
  Each support agent is appointed to a specific area, where the ticket resolution is a repetitive task across each agent, but sometimes the customer arrives at the incorrect area looking for help. In those cases, the agent must transfer such support chat to the correct area.
• With ML
  Systems will predict which area should answer each incoming issue, requesting some specific answers from the user.

Real-life use case

Each new customer support request will be dealt with by a general chatbot. This chatbot can address two possible scenarios. 

• For common questions and problems, for instance, things that customers can solve by themselves, the chatbot will provide some steps to be followed. 
• For more complex situations, the chatbot will collect issue details to assign the correct agent to assist customers according to their needs. 

Fraud prevention

Anticipating common problems simplifies the management of company finances. Thanks to ML, companies can process multiple signals for each individual user, and therefore, they can better understand the risk involved on each transaction or operation.

• Without ML
  If fraud takes place, the company could be exposed and lose resources. In addition, fraud detection usually consumes a lot of time and resources.
• With ML
  Including machine learning in the process helps humans calculate the risk of each operation for each customer. Consequently, companies minimize errors and work smartly, consistently, and objectively.

Real-life use case

If someone uses a stolen credit card, makes unusual purchases, tries to buy many products, or does some small operations on several sites, it will then be categorized with a “high fraud score.” Agents will review the user activity before making a decision.

Credit and customer scoring

Using ML, we can score our customers individually based on their operations history and other data available online.

Requiring a new credit

• Without ML
  When a startup, young person, or venture requests a small loan, banks typically demand many years’ worth of tax returns and loan history.
• With ML
  You can now get the user’s “score” based on a large amount of user data. This includes their loan history, transaction history, credit card use, online shopping information, expected future income, and a lot of background information that will help the company identify good borrowers more accurately.

Improving the decision time

• Without ML
  Bank employees usually require several hours to make credit decision.
• With ML
  We can determine how much a user can borrow based on multiple factors. This decision will be faster and more accurate than a human decision. Once available data has been processed and analyzed, agents can make a final decision.

Real-life use case

Bearing the previous example in mind, let’s analyze this situation:

With ML and big data, customers get a score between “0” (zero) and “100,” where less than “60” is a low-scoring result. Therefore, customers scoring under 60 should not have their loans approved. Customers scoring between 60 and 90 should be reviewed carefully. Finally, those scoring above 90 are customers to whom a loan should be easily granted. 

Conclusion

Nowadays, we can improve our organization through the correct use of technology. Machine learning, big data, and AI are necessary partners at decision-making time.

We can retrieve relevant information from the smallest piece of data within an organization. Either input quality or quantity are key factors for success. To put it simply, try not to underestimate the importance of the information available within your organization. It is essential to make clever decisions on a daily basis.