Machine Learning In The Payments Industry
How are Machine Learning Models going to change the Payments Industry?
It wasn’t so long ago that CEO’s and large commercial banks were convinced that more bank locations would always be necessary to service and acquire new customers. However, in the last ten or five years we have seen an emergence of Digital Banks, that have never and will probably never own a physical location, but still manage to grow their user base and add additional services including insurance, mortgages, and loans.
In the Banking industry, we have seen companies like First Bank of Nigeria, United Bank of Africa, Zenith Bank, Guaranty Trust Bank dominate for well over twenty years. However, just like the digitization of banking has forced incumbents to change their strategies, the digitization of payments has provided companies like Flutterwave, Paystack, Remita and lately even Korapay to take up some of the market shares, not by focusing on traditional businesses, but by focusing on startups who have grown to overshadow and sometimes even bankrupt traditional businesses. Think of Alaba International market versus music streaming sites, Taxi’s & “Danfos” versus Uber & Taxify.
But as more and more consumers and businesses are understanding that digital is the new traditional, I’m out here thinking, what’s next for the payments industry?
Just like the companies who were open to adopting computers and databases in the early 2000s, or the companies who understood that the internet would influence the way customers decisions in the 2010s, I believe that companies who understand that machine learning is going to change the financial industry and are actively investing in it are most likely to come out of the 2020s as top dogs.
Machine learning is steadily making its way through the payments and financial services world. In recent years many successful machine learning applications have been developed, ranging from data-mining programs that learn to detect fraudulent credit card transactions, to information-filtering systems that learn customers’ buying behavior, to the most popular application of AI in financial services — and perhaps the most limited — the chatbot, a program that converses with customers through text or speech.
It’s important to note that before businesses can start using machine learning applications, it’s absolutely necessary that they digitize their processes or operations first. Even though many companies already digitized their processes or are seeking ways to, now is not the time to rely on traditional business procedures because with the lockdown and quarantine measures enforced by the government to combat the spread of Covid-19; the importance of digitization of processes and operations has never been more evident.
MACHINE LEARNING MODELS
Machine Learning Applications are “Models”. Models are Built upon Data. So when we talk about Machine Learning, we talk about the process of using digitized labeled data which is stored as a data set in a database, and using automated processes to create analytics from which users can derive information. This a process that most companies have already gone through but this is only the first step into Machine Learning.
Whenever a Data Set (X) can be used to create analytics from which users are deriving information, the next step is to use Machine Learning with the help of algorithms to create a New Process changing the data to (X=X+1), which in itself provides a new Data Set completing the Machine Learning Model.
Payment systems generally generate a lot of data, because of industry standards, and the good thing to know is much of this data is well structured which reduces the effort needed to prepare it for use in machine learning algorithms. In machine learning, these data columns are referred to as “features” while getting the data ready to use is called “feature engineering”.
Looking at this payment flow above, you see the various points in a transaction cycle and you get to understand and appreciate the digitized processes in completing a transaction. For each a junction in the transaction journey, data is created, sent, or stored for later analysis or probing.
To illustrate, imagine a Payment Service Provider who processes millions of transactions a day. Each transaction that is submitted by a merchant contains transaction-related pieces of information such as BVN (Bank Verification Number), CVC (Card Verification Code), BIN (Bank Identification Number), Expiry Date, Customer Name, etc. Through the browser, the PSP is able to collect additional data like the date and time, device fingerprint, Browser-type and Version, IP-Address, and some other data points (X).
As the transactions are being processed they are stored in a “database”. Most PSPs will use this data to provide their Merchants with a standard report of the transactions of a particular day. Some might even go as far as aggregating the data to provide a summary of the data. Some, such as Korapay, go on to develop dashboards and graphs that are accessible through a UI, to show how transactions are progressing over time.
With the advancements in computing power, cloud computing, and distributed storage technology, a lot of companies including PSPs have been experimenting with ways to improve existing processes. For example, many PSPs make analysis on a large set of historical data that could help them wipeout false declines in payments. By using Machine Learning, companies with the help of data scientists could create an algorithm that is able to use the many variables that are part of a transaction and predict the likelihood of a newly submitted transaction to be legitimate.
How will Machine Learning affect Payments?
Currently, it seems that the majority of Machine Learning solutions for payment processing are focused on fraud detection and prevention. Some companies claim to offer straight-through processing software as well but for me’ “presently”, I only see two key areas when thinking about Machine Learning solutions to offer to banks and payment processors. They are;
- Fraud Detection
- Payment Optimization
The use of machine learning in payments is far more than just preventing fraud. Costs, conversion, connectivity, billing, and payouts are some of the areas where companies can use Machine Learning to make a difference. As more and more companies are getting used to the commoditization of digital payment companies, we all should agree that there’s more than one way to stay successful as a Payment Service Provider. It shouldn’t be by focusing on reducing fraud alone but also thinking of means of driving down the cost, speed, endpoint connectivity, etc. A lot of value points can be provided to businesses with accurate analysis and use of data, this in turn can be sold by payment providers to get more businesses.
The example above was just one use-case of how Machine Learning could be used to improve existing payment processes, and the interesting thing about the example is that according to Ajay Bhalla, the president of enterprise risk and security for Mastercard; consumers are being inconvenienced as a result of the problem, merchants are losing sales and banks are losing transactions. This just proves there is a current need in the payment industry for this and there’s an opportunity for data scientists to solve this.
Thanks for reading ;), if you enjoyed it, hit the applause button below, it would mean a lot to me and it would help others to see the story. Let me know what you think by reaching out on Twitter or Linkedin.
