Businesses are increasingly interested in how big data, artificial intelligence, machine learning, and predictive analytics can be used to increase revenue, lower costs, and improve their business processes. Machine learning and artificial intelligence are set to transform the banking industry, using vast amounts of data to build models that will enhance decision making, tailor services, and improve risk management.

Badly assessed financial risks were at the core of the financial crisis in the late 2000s. Banks and credit companies used faulty models that did not highlight the real threat of the mortgages granted. When the housing bubble burst, it led to the collapse too big to fail financial institutions and the recession of the entire economy for a few years. All these problems could have been avoided with proper risk hedging tools.


Big Data


Artificial intelligence is increasingly becoming a routine part of our daily lives with the introduction of digital personal assistants, music and movie recommendation services, and cars that can see around corners. Just as smartphones, online shopping sites, and music apps learn and adapt based on our preferences, cognitive computing can be used to teach computers to recognize and identify risk.

Of course, computers have always been able to perform mechanical calculations faster than humans. The difference is that with cognitive analytics, machines can learn as well. The use of artificial intelligence to manage risk is particularly helpful when handling and evaluating unstructured data —the kind of information that doesn’t fit neatly into structured rows and columns. Cognitive technologies, such as natural language processing (NLP), use advanced algorithms to analyze the text to derive insights and sentiment from unstructured data. Given that a 2015 International Data Group study estimates that roughly 90% of data generated today is unstructured, implementing cognitive analytics can place businesses right on the cutting edge. Leaders who leverage cognitive technologies to anticipate and proactively manage risk can gain competitive advantage and use the threat to power their organizations’ performance.

Look at fraud detection as an example. The old method of detecting fraud was to use computers to analyze a lot of structured data against rule sets. For example, fraud specialists would create a threshold for wire transfers at €10k, so the computer would flag any transaction over that amount for additional investigation. The problem is that this type of structured data analysis often creates too many false positives, which require hours of scrutiny. With cognitive analytics, fraud detection models can become more robust and accurate. If a cognitive system kicks out something that it determines as potential fraud, and a human discovers it’s not fraud because of X, Y, and Z, the computer learns from those social insights. Next, time, it won’t send a similar detection your way. The machine is getting smarter and smarter. That’s a huge game-changer.




As these cognitive fraud detection systems continue to learn, they will be able to detect more complex fraud, an advantage that may have the most significant impact on risk management. Cognitive technologies can help unearth emerging patterns that humans could never identify. Then those emerging patterns become a new pattern to look for. In fraud detection, cognitive is expected to make it more accurate and offer stronger protection. These new capabilities are not limited to detecting risk. Cognitive analytics allows businesses to quickly tap unstructured information, personalize services, and reduce subjectivity in decision making. Among the areas where this approach to data is useful are healthcare, retail, and even litigation, where computers are “trained” to discover specific information in millions of legal documents and perform any necessary global language translation. At this stage, cognitive technology is still an assistive technology to help suggest strategies and probabilities of outcomes, and that human expertise is always important. Together, humans and computers will be able to do things that were just not possible previously.


Case Study: Debt Collection



The collection process usually follows a predefined schedule of letters, emails, and phone calls that communicate with increasing urgency the need to repay the debt over time. Ultimately, if the debtor refuses to repay the debt, then legal action can be taken by the collection agency to force repayment. Legal action is expensive and often outside of the collection agency’s control, so it is only viewed as a last resort and avoided as much as possible. In contrary to popular belief, debt collectors generally prefer to cooperate with debtors to repay their debt by offering interest-free extensions, repayment plans, or in some cases waiving parts of the debt if the debtor is genuinely unable to repay. However, this is only possible if the debtor is cooperative and responds to the collectors’ communication attempts (e.g., answers the phone or replies to email). Letters and emails are mostly automated, but phone calls still require human collectors to physically dial a number and have a conversation with the debtor. This is integral to the collection process because debt collection is highly emotional, and an experienced collector can decipher the needs and problems of the debtor and determine the best course of action to maximize the likelihood of repayment. However, debt collection agencies generally have a large number of open cases, and human resources limit the number of phone calls that it can make. Under these constraints, it becomes infeasible to call every debtor, and a method to select debtors to call becomes necessary. It is not calling a debtor who needs human persuasion results in further delinquency and higher risk for non-repayment but calling a debtor who doesn’t require additional persuasion results in wasted effort.


Key Approaches

Most work today falls into two sides of collections and bad debt management:

  • Reactive groups. Using predictive analytics to determine who to invest effort in for gatherings and to prioritize collections activities to rank those most likely to pay as well as successful outcomes
  • Proactive measures. Pre-delinquency management to identify the triggers and events that take place before someone starts missing payments so the business can proactively communicate to those customers and keep them on a path away from delinquency.

The way we’re approaching these areas today is looking at pre-delinquency management – and this is where we marry up the analytics processes with customer engagement. It is all around getting to your customers ahead of time and getting to them with the right messages through the proper channels so that we can change behaviors before they become somebody that the business needs to chase. On the predictive analytics side, it’s taking a lot of the past approaches around ranking, segments, and profiling, and creating predictive models to predict successful outcomes – such as who’s most likely to pay because of a phone call or because of a letter, or who is more likely to pay more significant amounts. From that, we can use predictive analytics to start prioritizing offers and start understanding who might need a different kind of conversation as well. It’s not always a one-size-fits-all in collections management. It doesn’t always have to be a weighty message, because there may be segments of the customer base who would only need an email, not even suggesting that they’re in arrears. Still, just an informative email from your organization to trigger their minds into, “Oh, wait – I don’t think I paid my bill.” Maybe it’s just someone accidentally forgot versus chronic late payers who you’ve got to chase every month or every 60 or 90 days. So it’s all about using analytics to better segment your customers, understand who needs to be differentiated, and how can we prioritize the communications to ensure the most successful outcomes. And indeed, before they become a risk would be ideal.




As in all models related to big data, the primary problem is related to data cleaning. Since it’s a matter of garbage in, garbage out, before making any prediction, the company dealing with this task needs first to build the pipeline to bring in the date, clean it and use it for training the neural network. Another challenge might be related to various personal data protection regulations and privacy matters. Beyond data, some of the other problems around analytics include:


Being able to attract and retain the right types of people and the right skill sets and being able to have enough people to keep up with an organization’s analytic demands.


It can turn all of that data into real insights and into practical information that can be used by the business.


The ability to take action on that data. It’s not going to solve any problems having great models and insights if there’s no way to deploy them into business processes and into the channels that involve customer engagement to influence and change behaviors.




With these challenges also come opportunities. With data, the question here is, can you get at it? What do you have access to today? It might be operational data, billing data, maybe transactional or usage data as well. And that might exist in many different types of systems. The message here is to make sure you’ve got something – something to go on. As far as people go, again, it’s a matter of skills and volume of resources. Tools are all around bringing in the right technologies that enable efficiency and enable the processes to get us from point A, which is data, to point B, which is process change and insights. And finally, the action piece is making sure that the analytics that you’re chasing is the analytics that are going to solve the business needs and can be used within the processes that exist today for customer engagement.

How all of these pieces come together ultimately, if we were able to bring all of this together, we’d have a very robust customer view that would enable a lot of analytic insights around usage and satisfaction and billing data and marketing data, and even demographic and grid data as well. But the one that everybody has is the billing and payment data. Typically, because it involves money, this is always the cleanest and most readily accessible data, as well.

Predictive Analytics

Predictive analytics, it’s all around profiling, visualizing, segmenting, and ultimately prioritizing the conversations and messages that are going to go out to customers.

Self-Service Analytics

Self-service analytics, which is a newer trend. It is all about putting into the hands of the business the ability to ask their questions, the ability to get their counts and start making decisions on the type of data that they typically have to go to an IT or business intelligence team. By moving some of that light querying and reporting back to the business users, we’re freeing up some of those resources—those hard-to-find resources who are more statistically inclined and more driven to predictive modeling—to build valuable predictive models that are going to help transform the business.

Message Execution

Message execution and this is all around taking those predictive models and information learned and feeding them into the channels so that delivered messages are relevant, targeted, and part of an automated multi-step, multi-touch, multi-channel communication strategy. So it’s not about trying to do everything manually and worrying about resources. It’s about getting your best practices and best communication strategy automated and connected into analytics, so the customers are getting that right message at the right time through the right channel with the right offers.




Imagine if a piece of software could tell you the repayment probability both for current, but also for future clients. Of course, that is what the FICO scoring model aims to do, but as we’ve seen, it has not been entirely successful. FICO stands for FI (Financial Accounting) and CO (Controlling). Models based on predictive analytics which use big data could have a better chance of foreseeing repayment chances. Yet, most companies have not adopted these tools. Not only can predictive analysis tell you which clients are the highest risks for your business, but it can also identify when it’s best to get in touch with them for maximum results. For example, if they work in shifts, it’s best to call them when they are not at work or resting, which could be outside regular business hours.

Showing your clients that you care about their habits and lifestyle improves your chances of them listening to your call agents and ultimately raising the collection rates. Of course, to train the system, you need the logs of past conversations. Since the late 80s, the FICO score has been the gold standard for evaluating loan application and creditworthiness. It even comes in a few different flavors, including auto FICO, and healthcare FICO. Machine learning and specifically predictive analysis can take this process beyond a single number and create a 360-degree portrait of the client, taking into consideration more than just the credit history and current debts. Now, it can include data from social media, spending patterns, and more. Such a tool would be great for foreign clients who have no previous FICO score but would be great business partners, like foreign investors. It would also offer a fair chance to recent college graduates or other young people. By taking into consideration a broader array of input data, the accuracy of the prediction improves consistently, and it can also be refined to a very personal level. The new outcomes can go as far as setting an individual credit score limit to minimize potential damage.


Concluding Thoughts


Get as much data in your models as possible, but don’t wait until it’s perfect. You can always start small and grow it. Use your data and ensure you’ve got the tools and the people available to gain value from it. Start pushing the organization to become customer-centric, so marketing doesn’t need to be separate from credit and recovery. And finally, start moving towards the ability to take immediate action on analytic findings, so you don’t sit with a model for six months or nine months or 12 months. Instead, take those findings, get them into the processes, and get them integrated so that actions can be taken immediately. It can boost conversion rates by targeting the right people at the right time and reduce or eliminate financial risks. Of course, an in-depth analysis could go as far as identifying the message which would influence the client the most. Targeting the right clients also means more productivity for call center agents and less wasted work hours. Having a mathematical model behind the decisions eliminates bias and makes the process fair. Through carefully considering and adopting useful predictive analytics and customer engagement practices, FinTech can start to reduce risks and improve customer satisfaction.