How fighting fraud and boosting sales have turned from a dilemma into a must-have set

Online businesses have been for years struggling to effectively fight fraud of various kinds. Along with the rapid growth of the ecommerce industry and the emergence of numerous new payment methods, preventing fraudulent transactions has become an extremely challenging task. Despite all the efforts of greatly skilled and knowledgeable risk managers supported by diverse IT solutions, many companies still mistakenly consider fraud prevention as a limiting factor for sales. They believe that false positives – erroneous transaction denials that lock sales and discourage customers from buying – are the price they have to pay for protection against chargebacks. Actually, nothing could be further from the truth! In the following article, I explain how the combination of advanced Machine Learning and diverse data gathering and enrichment methods not only enables effective fraud prevention (nearly completely free from false positives) but also provides unique business insights and accurate predictions that merchants can leverage to increase sales and outperform competitors.

Rules vs AI

First of all, it is crucial to understand the difference between the deterministic rule-based approach to fraud prevention and the probabilistic Machine Learning approach. The former is possible only thanks to the extensive knowledge and analytical skills of risk management professionals. Risk managers analyse the stream of incoming transactions, historical data and many other factors, and define special anti-fraud filters – sets of designates of fraud attempts, the appearance of which equals a fraud alert. This approach requires frequent updates of rules and in the face of the present-day’s diversity of payment methods, variety of customer habits and sophistication of fraud, becomes more and more vulnerable to false positives. Each single rule, to be effective, cannot include more than 5 variables. Otherwise, managing a group of rules becomes extremely hard. Since the rule-based approach is deterministic, for a rule to be activated, a transaction must meet all the conditions defined by that rule. In effect, even a very subtle change in a fraudulent operation can make the rule helpless. In order to avoid situations of this kind, risk managers oftentimes define broad rules that in some cases might, unfortunately, mistake healthy transactions for fraud attempts. The probabilistic Machine Learning approach, in turn, relies on rich data and special anti-fraud models that automatically detect interdependencies between apparently unrelated variables and thereby define the designates of fraud. Unlike rule-based systems, where the number of variables is strictly limited, Machine Learning models take into account thousands of parameters and the more data they are provided, the more effective they get. The models analyse the stream of transactions (and surrounding events) in real time and calculate the probability of fraud in each single case. In this approach, a risk manager turns from a person who manually handles the stream of incoming payments into the one who coordinates the automated Business Intelligence process and closely cooperates with data scientists responsible for the system maintenance. Thanks to the large number of parameters, fine-tuned Machine Learning models are able to predict even the most sophisticated fraud attempts. They predict them with outstanding accuracy, minimising the ratio of false positives to the level unattainable for rule-based solutions.

360o view – 100% satisfaction

Machine Learning anti-fraud solutions gain effectiveness in proportion to the quantity and quality of data they analyse. Accurate verification of the cardholder’s identity is at the heart of fraud prevention. Therefore, in order to tell whether a customer is who they declare to be, the anti-fraud system must collect, enrich and analyse huge amounts of data associated with that person and their interaction with the website or application. The User Profiler created by Nethone, for instance, collects over 5000 data points in each single case. The system takes into account such variables as hardware and connection specs, software settings, and the user’s behaviour (including both, actions and activities). In fact, while preventing fraud, the merchant obtains information that can be effectively leveraged for purposes reaching far beyond protecting the bottom line. Rich data allows online merchants to thoroughly x-ray their customers and adjust the overall shopping experience to their specific collective as well as individual needs, expectations and preferences. The customised customer experience refers to numerous aspects. Merchants can optimise their websites/applications for the types of devices used by their customers.  Furthermore, they can adjust the UI to the way their customers browse the content. Additionally, they are enabled to provide truly accurate shopping recommendations that not only match the previously purchased/viewed items but also meet the shopper’s taste and style. The possibilities are nearly infinite.

Bespoke models and sales-boosting insights

The above mentioned examples of using rich data to go beyond fraud prevention and actually increase sales are possible thanks to bespoke Machine Learning models created for particular purposes and being applied simultaneously with the fraud-fighting ones. As explained earlier, Machine Learning solutions deliver highly accurate probability assessments – predictions based on extensive networks of interdependencies identified by models. Those predictions, as explained above, might as well refer to fraud as to other phenomena. In fact, a solution like Nethone can provide merchants with truly accurate business forecast, embracing whatever indicators they need to predict. Through the use of customised models, a merchant can predict expansion, contraction, MRR, churn rate, LTV, and many other indicators – depending on whatever they need to know to outperform their competition.

No more dilemmas

To cut the long story short, fraud prevention becomes truly effective thanks to the probabilistic Machine Learning approach to the problem. Bespoke, fine-tuned Machine Learning models detect fraud without interfering with the flow of healthy transactions which equals reduction of the false positives ratio to minimum. Being fuelled with rich data, solutions like Nethone can provide merchants with truly accurate predictions of diverse business indicators, giving them significant competitive advantage. As a result, an anti-fraud solution becomes a precious asset for merchants. Thanks to Machine Learning solutions like Nethone, fighting fraud and boosting sales shift from an either/or relation into a must-have set… and go hand in hand.

Meet the Author


Aleksander Kijek, VP of Operations Aleksander Kijek is a Python programmer and Linux enthusiast fascinated by FinTech and Neuroscience. At Nethone, he is responsible for business and product development, coordination of the workflow and assuring comprehensive operational excellence of the company. My-LD-Profile-logo      

Meet Nethone

n_logo Nethone provides AI-based Anti-fraud and Business Intelligence solutions for ecommerce companies of any size. The company was founded in 2015 by a team of experienced data scientists, risk managers and security specialists with rich merchant background to help online businesses turn threats and challenges into profits. Nethone operates globally, making commerce safe and prosperous.