Business

Fraud, detected: Why Every Ecommerce merchant Should Get Well-Acquainted With Machine Learning

Machine learning – the capability of computers to process and learn from information – holds the promise of revolutionizing every facet of business, especially e-commerce.

Although the sci-fi image of truly sentient machines is not being realized any time soon, recent machine learning advances such as deep learning are already having an impact in e-commerce. From improving the relevancy of search results on an etailer’s website to better product recommendations, the ability of machine learning to draw connections from past correlations in historical data and apply them to useful predictions in the present is helping online merchants boost revenue and slash losses right away.

Boosting revenue, avoiding chargebacks

One facet of e-commerce which is already reaping the benefits of machine learning is card not present (CNP) fraud detection. One vendor of solutions for fraud protection for ecommerce, Riskified, uses its machine learning-based system to enable its customers to safely approve more online orders while also accurately detecting and declining fraudulent orders. This improves customer retention, boosts online order revenue and cuts losses from chargebacks and their fees and penalties.

Riskified’s platform frees merchants from the conundrum forced on them by previous, less intelligent fraud prevention tools, such as overly strict rules. Those inadequate tools would sometimes do a decent job of stopping the fraudsters, but at the cost of rejecting lots of legitimate orders. For every falsely declined order, revenue is lost – and so is a potential repeat customer, compounding those losses over time. This severely hinders a merchant’s growth, since new customers are unlikely to ever return to an etailer that has labeled them as a criminal.

machine learning in business

A closer look under the hood

Many machine learning algorithms are trained using a very large set of pre-labeled data. In Riskified’s case, they fed their algorithms a huge set with each one labeled either legitimate or fraudulent. Each order also had a great deal of details included (shipping address, billing address, IP address, email used when placing an order, etc.).

The algorithms then built an initial model of what factors were associated with fraudulent orders and which ones were more associated with legitimate ones. The system then used this initial model to comb through the training data again to see if those associations resulted in the same decision (fraud or legitimate order) as the orders were initially labeled with. Any errors in the algorithm’s prediction were then used by the system itself to tweak its model by adjusting the statistical weight given to each detail about the order. Then the algorithm checks the predictions from the model against the training data set again, and so on until the algorithm’s predictions are very close to the labels given to each order in the training dataset.

One of the benefits of this approach is that not only does it do a good job of accurately telling the fraudulent orders from the legit ones, but the algorithm also tells you the statistical weight of each factor.

Another advantage to this method is that the learning continues even after real paying customers begin using it. Once Riskified used its large training data set to generate a good working model, it then deployed the fraud prevention system to paying customers. The customer’s own real data is then fed back into the system, continually refining it. More accurately, the data from all of Riskified’s customers are used to continually update and improve the algorithm, whose benefits are then enjoyed by each individual customer.

Constant adaptation is key

Solutions based on machine learning are able to stay on top of fraudsters’ ever-evolving tactics, something which other fraud prevention tools are unable to do. On top of that, these solutions are able to make highly accurate decisions within milliseconds, much faster than manual order review. By allowing companies to earn more revenue, and decrease losses from chargebacks, machine learning is paying dividends for e-commerce companies by improving the accuracy and speed of fraud prevention.

About the author

Andy Robert

Andy Robert provides helpful information and assets for those looking for high quality tech services. Their mission is to provide the authentic and determined information so you can make an informed decision on cloud.

Add Comment

Click here to post a comment