“In order to tackle fraud you must be one step ahead”: Martina King, CEO, Featurespace

Martina King is the CEO of Featurespace, a UK-based provider of adaptive behavioural analytic technology and services. The business works with a multitude of leading businesses in the UK to analyse complex fraud behaviour.

PaymentEye and Martina discussed how Featurespace was created, the importance of tackling fraud in the UK, and the most common cases of fraudulent activity.

How was Featurespace formed?

Our machine learning software system was developed at Cambridge University by the late Professor Bill Fitzgerald and his student Dave Excell. Bill was the head of Payment Processing in Applied Statistics at Cambridge. Bill was associated with machine learning techniques, such as separating signal from the noise and change point detection, and he is one of the most respected authorities in the world when it comes to applying new techniques and new methodologies to data. Dave Excell, who worked with Bill to develop Featurespace, is CTO of the company.

Featurespace has developed into an extremely creative company, looking for an alternative to existing rules and systems, and a way for businesses to be able to not only block, but also review the number of genuine customers that would have been turned away i.e. reducing false positives.

How does Featurespace’s technology identify the false positives as well as the real fraud and fraudulent customers?

We model absolutely everything that’s occurring in the dataset at an individual level, such as separating the signal from the noise point. We are able to understand what “normal” behaviour looks like at an individual consumer level, and also at an event level, and then we can anomaly test.

This enables us to see if someone is acting completely out of what would be expected to be the ‘normal’ behaviour, and we determine whether it’s a new fraud attack or previously unseen fraud attack. Both are revolutionary in the reduction of fraud, but simultaneously we can understand what a normal pattern of behaviour will be for a “normal” customer, therefore we can let transactions go through that previously systems would have blocked. Typically, this will reduce the volume of fraud by around 70%.

The more defined you are in decision making, the greater the accuracy of the results. It was quite surprising to us in many respects when we were given access to the software that we could provide for the customers we were working with.

How important is it to explore possibilities around stopping fraudulent activity happening in real time?

Somebody asked me what would be the point of having an advanced system if you couldn’t identify the fraud attack that’s occurring. The whole idea with fraud is that you need to try and stay ahead of the fraudster. If a business is operating one step behind all the time, then you are not protecting your business as well as the technology you’re operating on.

What advice would you give to a business that’s looking to reduce the fraudulent activity it is exposed to?

It’s important to have a guide on the front line in the organisation that is aware of modern methodologies. I have been very impressed by the quality of the people who work on the frontline solving these problems day in, day out.

It’s important to be inquisitive about new methodologies and techniques, and quite often businesses need the internal support of their organisation.

As CEO of Featurespace, what does a typical day look like?

Even though we are a machine learning business, we still deal with real businesses and people, to ensure that we do the best we possibly can for our customers. I make sure that we hire and retain the best possible talent, and day in, day out, and I make sure the business is running smoothly and manage our teams.

Within Featurespace there are platforms that can stop fraud happening with algorithms. Are these monitored automatically or is there human intervention?

It’s primarily down to automation, but there is a human element. We produce a full anchoring solution; all the user interface and we are deploying our software into our customer’s organisation either on premise or in the cloud, and then they are able to manage and see alerts, identify where the fraud detects are, and rely on machine automation to take care of the rest.

Typically, our customers are relying on our systems to be able to identify these types of fraud attacks and to ensure that they are teaching the system to learn the characteristics of a fraudulent account, which protects the business from suffering more of them.

Is there one kind of breach that’s more common than others? Are there any big differences that you see within the data?

Fraudsters try all sorts of methods, but a typical problem that we solve frequently is application fraud. This is when people pretend to be somebody that they’re not to get through the system.

It’s a very typical type of fraudulent activity. If somebody is creating a fake account to wire money then this is also something that our system would be able to spot.

 

Related reading