This page is intended for users in Singapore. Go to the page for users in United States.

The Fight Against Fraud: DataVisor on Tackling Fraud with Unsupervised Machine Learning

When I think about machine learning as a millennial consumer, Netflix and shopping first come to mind; recommendations would pop up every week on the next show I should binge-watch or a new pair of jeans to add to my perpetually bursting wardrobe.

Then there’s Bus Uncle, arguably Singapore’s favourite chatbot, to fill me in on bus arrival timings (in perfect Singlish!) the next morning as I rush to work, regretting my late-night movie binge and impulse purchase after combing through 12 pairs of jeans. Thanks, Artificial Intelligence!

On the flip side, one of machine learning’s most important uses for businesses is in fraud detection. Taking this technology a step further with unsupervised machine learning is DataVisor, the industry leader in fraud detection solutions.

I sat down with Matthew Harland, DataVisor’s Technical Account Manager, after their recent meetup to dive deeper into their cutting-edge Unsupervised Machine Learning technology, and how it helps to solve a problem that costs businesses millions of dollars every year.

Although we hear about fraud all the time across consumer industries, Matthew believes it is an issue that deserves more appreciation.

Fraud Detection, Unsupervised

Online fraud is rampant and can be split into three main verticals: social e-commerce fraud, financial fraud and advertisement fraud. Working directly with the world’s largest financial and internet properties like Western Union, Pinterest, Yelp and Wechat, DataVisor protects such businesses from a wide array of attacks including fraudulent transactions, promotion abuse and even money laundering. To date, DataVisor has protected 4.1 billion accounts globally.

While Artificial Intelligence (AI) is a scalable approach towards tackling fraud, widely-known limitations include extensive training and historically flagged labels that are first required to furnish traditional machine learning models.

This is where DataVisor steps in, with its proprietary technology based on Unsupervised Machine Learning (UML) models. By processing a company’s historical data through advanced UML algorithms, without requiring pre-determined labels of normal or fraudulent behaviour, DataVisor proactively detects and solves fraud or suspicious behaviours within a company’s ecosystem, and helps translate that into actionable insights for the company to undertake.

DataVisor’s UML solution helps businesses capture attackers without the need to train data beforehand, and often before they can do damage.

“Most of the clients we work with are technology-first companies, and they have always tried to develop something in-house to tackle fraud. While it is effective to a certain extent, fraud detection is not their core business. When it comes down to specialisation, this is where DataVisor comes in. We offer a service that provides companies with an uplift to what they’re existingly seeing. So when we put our results in front of our clients, a lot of them are shocked at what goes on unknowingly behind the scenes,” Matthew explained.

Figuring Out The Fraudsters

Historically, fraudsters are known to be masters of innovation and disguise; developing a wide range of schemes to commit and conceal fraud, and finding new ways to do so every time. So how do we outsmart the fraudsters?

When looking at a fraudster’s behaviour and their evolution throughout the years, there are two ways to detect fraudulent activities: point of attack and upstream detection.

To be able to attack a company’s ecosystem, a fraudster needs to already be part of the ecosystem and want to incubate himself within that and in scale: Scripting mass registrations or abusing credit card information bought off the Dark Web, among others. When these actions eventually get flagged as fraudulent behaviour and by the time companies act on securing their website, fraudsters have already moved on to more sophisticated forms of attack.

It’s a never-ending and ever-evolving cycle. Fraudsters tend to test the limits of a company’s system and see how far their actions can stretch without getting caught. This is why unsupervised machine learning is more effective in detecting fraud than the ‘trained and labelled’ models; it helps companies detect new and unknown attacks before damage is done.

From The Top

What makes DataVisor stand out is first and foremost the technology that they employ and how the company truly values innovation in a way that has practical applications.

“UML is a branch of AI that has been in theoretical existence for quite some time. But in terms of practical application, DataVisor is one of the very first companies to be able to demonstrate tangible results that arrive—or are derived—from this field of study that has been long debated from an educational or theoretical point of view,” Matthew shares.

He added, “Technology is after all developed by people, and I think the quality of hires in our core development teams is testament toward the quality that we output as a service for our customers.”.

This focus on quality is a reflection of DataVisor’s two co-founders, Yinglian Xie and Fang Yu, both of whom are PhD holders from Carnegie Mellon and UC Berkeley respectively, and have personally spent years in Microsoft Research together.

“As much business acumen as they do have, they also have the technical experience to really understand what it means from a tech application perspective, and problem-solving into real work scenarios. This dedication, once embodied at the top, translates into hiring decisions; developing the necessary skill sets—and very modern skill sets for that matter—to tackle a problem that is also very modern,” Matthew explained.

Matthew’s journey with DataVisor stemmed from his background in advertising technology, where he learned just how rampantly fraud thrived in the industry. “For me, realising the extent that fraud permeates at every level within this industry, and to be on the upper side right now; to be able to put some of my knowledge in practice and application; to see my ideas amount to something—that means something for me. From a job satisfaction perspective, it ranks in quite high up there,” he beamed.

Matthew also shared that the trust and autonomy placed in him to help the business and DataVisor’s customers in the region is what motivates him. “Above and beyond everything, the ability to grow and be involved in an industry that is niche and at the forefront of a lot of concerns (motivates me). Just being in this environment enables me to grow personally, and I think that all these factors combined has allowed me to be happy where I’m at, and be quite proud as well, ” he added.

Watch Out Fraudsters

Earlier this year, DataVisor was enrolled into the Infocomm Media Development Authority (IMDA) SG:D Spark programme, empowering the company to advance its growth in Singapore and the region by gaining access to leading industry partners, government tools and grants. With their excellence in fraud prevention technology, DataVisor will be a key player in supporting the digital transformation of businesses in Singapore, and propelling our vision towards a digital economy.

Wantedly, Inc. (Singapore)'s job postings
2 Likes
2 Likes

Weekly ranking

Show other rankings
If this story triggered your interest, go ahead and visit them to learn more