March 10, 2017

Machine Learning at Heart

by Payment Fraud Prevention, Industry Trends and Technology, Point-of-View, Cash Management, Payments and Financing

Originally published in OpusCapita Journal.

Data Scientist Ali Faisal gets enthusiastic when talking about the transformative capabilities of machine learning, advanced analytics, and other technologies utilizing artificial intelligence. He has been researching the areas for years, and now he focuses on finding ways these disruptive technologies can propel finance and procurement solutions and services to the next level.

Machine learning is extremely powerful for exploring large amounts of data and, for instance, making predictions, suggestions, and categorizations based on it. For many of us, an everyday demonstration of this power is the helpful voice of Siri on our iPhones, the customer service agent at PayPal, the spot-on movie picked for us on Netflix, or the interesting items recommended to us on Amazon.

“These big players have turned machine learning into the hype it is today, as they revealed to the world the potential of the technology that had been maturing in academia for a couple of decades,” says Ali Faisal, who himself carried out postdoctoral research on the subject at Aalto University before joining OpusCapita.

The widespread adoption of machine learning in different industries has been further spurred by the increase of cost-efficient high-performance computing and also by the increasing amount of available data. Faisal points out that the terms ‘machine learning’ and ‘artificial intelligence’ often get used interchangeably.

“To achieve smart behavior, an artificial intelligent system can utilize programmable rule-based logic, robotic process automation, or intelligent machine learning algorithms,” he clarifies.

“For me, machine learning is the most challenging and also the most rewarding part of artificial intelligence – after all, it is the part which truly makes a machine intelligent and adaptable.”

Revealing the secrets of data

What else should those of us, who are unfamiliar with the finer points of these technologies understand? To start with, Ali Faisal explains that, as a technology, machine learning is at the intersection of computer science and statistics. Computer science focuses on designing programs that solve problems, whereas statistics deals with data and what can be understood from it.

“Simply put, a machine learning algorithm is a program that itself defines the logic needed to solve a certain problem and learns more through experience. It is not hard-coded like computer programs nor does it have pre-set rules, like robotic process automation.”

There are hundreds of different machine learning algorithms, but Ali Faisal broadly differentiates three major paradigms – namely supervised machine learning, unsupervised machine learning, and reinforcement learning. The mainstream applications today utilize supervised machine learning, which means the algorithm is given labelled training data.

“In practice, training data can be historical data for a process, with ready labels. These labels identify which data belongs to which category. A supervised machine learning algorithm learns the logic based on the training data in order to correctly carry out categorization in the future,” Faisal explains.

In unsupervised machine learning, the algorithm is not told in advance what is what; rather, it just goes on to explore the data available to find a solution for the given task.

“Machine learning is very efficient in revealing hidden patterns in the data and discovering new associations that may have previously been unknown to us humans. Fraud detection and different recommender systems are possible applications of unsupervised machine learning.”

The third approach, reinforcement learning, includes a human in the loop.

“When a machine makes a decision, for instance when categorizing the data, it can be either rewarded or penalized based on how it performed. The algorithm then adjusts the rules accordingly, and learns through trial-and-error.”

Intelligent automation for finance and procurement

OpusCapita has been exploring the possibilities of artificial intelligence and data science thoroughly over recent years and has identified several areas where machine learning algorithms are being developed to enable intelligent automation and deliver superior customer experience.

“In the finance and procurement departments, there are still many manual, repetitive tasks that are done every day. Increasing automation with machine learning is not just about cost cutting and the efficiency of these processes, but also about elevating the quality of these functions and supporting the business even more,” Ali Faisal describes.

One area where machine learning is already applied is purchase invoice posting. The algorithm predicts the posting dimensions – GL account, cost center, VAT code, and the inspector of the invoice, for instance – with high accuracy.

The development of automated anomaly and fraud detection is also well on the way.

“The algorithm recognizes payments that are unusual in some way and can flag them for inspection. Similarly, machine learning can be used to classify payments into different spend categories or for supplier product data validation in the procurement solution.”

“What is, in my opinion, one of the most beneficial aspects of this kind of intelligent automation is that it is not static but based on constant progress and evolution of the logic. The predictive models adjust themselves over time and do not become outdated if the business logic changes.”


Read more blog posts about Industry Trends and Technology

This website uses cookies to ensure you
the best experience on our website.