Using Machine Learning to Spot Expensive Data Anomalies

Joel Anderson  |

Anodot was handing out a flyer at the Money 2020 Conference in Las Vegas this October, which you can see pictured above. It features a series of images, a black and white polar bear where every bear is totally identical, each one entirely indistinguishable from the next...except one. One bear has the slightest difference; it’s perfectly obvious once you spot it. But spotting it is next to impossible, even when you’re looking for it specifically, as the banner urges onlookers to “Find the anomaly.” Which is precisely what Anodot does: detect anomalies.

Business Incident Protection in Real Time

In a sea of cartoon polar bears (or, if you prefer, the penguins that appeared on the flip side), one outlier isn’t necessarily of much importance. But what if you’re talking about a sea of payments? Say you run a busy commerce site with thousands of transactions coming in and going out, each one virtually identical, until that one where your site mistakenly charges a customer three times the appropriate amount without telling them. In fact, it’s a bug on your site, and it’s happening repeatedly.

What happens now? Given the immense flow of transactions going through your site, you’re not likely to spot it. In fact, you’re not likely to know it’s happening until you start fielding calls from angry customers who have checked their bills and realized what happened. Now imagine the reverse situation, wherein you’ve undercharged people. Are you going to start contacting customers to let them know you’re charging them more? Or do you just take the hit?

That is what is at the core of Anodot’s anomaly detection system. Anodot tracks the data coming in from a specific platform and learns how to interpret it. Ultimately, it’s able to identify when something out of the ordinary is happening immediately, so that anyone operating a platform can address the issue before it grows.

“We deliver real time business incident protection through anomaly detection,” said Uri Maoz, VP US Sales & Business Development when we met up at Money 2020. “Reliability is important when we’re talking about payment systems. When you’re talking about a payment system that generates millions of transactions, it’s very important that things will work all the time. If there’s a problem, you don’t want to find out a day, a few weeks, or even a few hours after the fact. Our system delivers real time business detection so that when something isn’t working, you’ll know about it in real time and can act on it right away. We learn what normal behavior is based on the number of transactions for this partner and operation system device, and then learn the anomaly behavior over time. Then, when there is a deviation from normal behavior, we will identify that and alert you in real time.”

Anodot both at Money 2020.

Using Machine Learning to Pinpoint Problems

The reality is that a new era of relatively frictionless, online digital payments means two things. On the one hand, speeding up the flow of commerce has made spotting and fixing problems a lot harder in a certain sense. When a purchase is just a click away at any given point in time, it’s easy for a mistake to slip by in a sea of similar payments. However, it also means that the ability to collect and analyze data on those payments is much easier, creating another tool for addressing problems. What’s more, by utilizing machine learning, Anodot has created a system that’s better at spotting outliers than even a team of dedicated technicians could hope to be.

“We have a lot of companies here that are managing payment platforms,” said Maoz. “They have a lot of payments from different partners, different countries, and different payment types. The data is so granular that you cannot build some threshold and know when something is going wrong. This is where machine learning enters the picture, learning the anomaly behavior over time. Say something happens and there is a drop in the number of payments completed for a certain payment system. It might indicate a problem, but you wouldn’t even know about it. Our system, though, finds the anomaly at the same time there’s a drop. You’re told that something is happening in the system, a drop in the payment success rate, a drop in payment success rates, and payment rejection of a certain payment types. So you get a story describing what’s gone wrong.”

And the degree to which Anodot drills down into a company’s data means that it can identify a wide range of potential issues.

“When we look at the data, we consider every irregularity,” said Maoz. “We will look at the number of payments completed, the payment providers, CP, the types of devices used by each of the customers, payment success rates, the number of transactions, the volume of those transactions. Then there’s the technical metrics like APIOs or latency. We learn the anomaly behavior for each of these, and when there’s a problem, we will be able to correlate that.”

Reacting to Data Aberrations in Real Time

The value for a payment company is pretty clear for a service like this. The money you can save is immediate and clear when you can identify and fix issues with your platform instantaneously. However, as Maoz points out, the applications for a technology like this goes well beyond just payments.

“Companies everywhere, not only in the payment space, have a lot of data that they need to react to in real time,” he said. “If you’re an ecommerce site and suddenly there’s a drop in the number of purchases because something on the website is preventing people from completing the transaction, you need to know that as soon as possible.”

And, as Maoz pointed out, there’s a lot more at stake than just financial loss. The sort of anomalies detected by Anodot can be the sort of thing that erodes trust with customers in a profound way.

“It’s not only the money aspect,” said Maoz. “Best Buy (BBY) published a $200 gift card for $15 on their website by mistake. It took them around 7 or 8 hours to discover the problem. During that time, 10,000 people bought the gift card. At the end of the day, Best Buy found the problem, changed the price tag, and cancelled the transaction rather than give people the cards. Now you might say that they didn’t lose any money, but what happened to their brand? When someone wants to make a payment and it doesn’t work, you press a button and something is not working, it’s troubling. Did the transaction go through? Did it not? Now I don’t feel confident about the reliability of your system.”

Fintech Needs to Foster Trust in the Customer Base

The reality is that, as more and more people are moving from more traditional payment systems into the fintech realm, reliability is essential. Learning to entrust something as crucial as entrusting your finances to a new technology takes a lot of trust. The faith that sending hundreds of dollars through the internet with a single click is something real and tangible can quickly disappear when something goes wrong.

As such, ensuring that a company can identify when something goes wrong immediately and then act to rectify the situation is key to continuing to grow people’s faith that this fintech revolution is everything it’s cracked up to be.

“Especially in the financial space, where we think so much of innovation and new ways to execute a financial transaction, the reliability of the system is very important, said Maoz. “This is why we want to be able to identify any problems before the customer even finds out.

DISCLOSURE: The views and opinions expressed in this article are those of the authors, and do not represent the views of Readers should not consider statements made by the author as formal recommendations and should consult their financial advisor before making any investment decisions. To read our full disclosure, please go to:


Symbol Name Price Change % Volume
BBY Best Buy Co. Inc. 60.18 0.61 1.02 1,951,642 Trade



Symbol Last Price Change % Change










World Economic Forum at Davos 2019 - Kitty Parry CEO DeepView

Matt Bird sits down with Kitty Parry, CEO DeepView, at the World Economic Forum at Davos 2019

Emerging Growth

Global Cannabis

Global Cannabis Applications Corp is engaged in the design, acquire data and develop applications for smartphones and tablets. The company offer Citizen Green platform, an end to end infrastructure to…