Fraud and customer churn are issues that most organisations have to deal with at one time or another, but collecting huge volumes of data on customers to try and identify general patterns is not always the best solution.
Instead, organisations need to be working in real time and adapting and learning from data as it flows through the system, in order to establish what normal behaviour looks like for particular customers and identify activity that is out of the ordinary.
One company that has been working on this concept for several years is Featurespace, an enterprise software start-up specialising in intelligent analytics, which spun out of Cambridge University's Department of Applied Statistics and Signal Processing.
The company has developed a behavioural change recognition engine called ARIC (Adaptive, Real-time, Individual, Change-recognition), which uses behavioural change algorithms to build up statistical profiles of individual people and identify and predict change.
Featurespace has built two software solutions on top of this platform. “Snowdon” delivers real-time fraud detection that helps organisations reduce risk, and “Nevis” enables businesses to protect their investment in customers by identifying and preventing churn before it starts.
“Rather than looking at a crowd-based model of what represents a threat, we are able to spot subtle changes in what an individual customer is doing and highlighting that as a threat,” said David Excell, co-founder and director Featurespace, speaking to Techworld.
Most of Featurespace's customers are in the online commerce sector, because these organisations tend to have a richer set of data about how customers interact with their platforms, according to Excell. Online firms also tend to have a much more of a customer-centred approach to marketing and fraud.
“Traditionally, organisations have tackled churn by targeting customers after a certain period of inactivity – say two weeks – and that's applied to everyone. But some customers may only use the service once a month, and just because they haven't come back in two weeks doesn't mean they are churning,” he said.
Excell said that the same goes for fraud rules, which need to be tailored to a person's activity. It is only when that person steps outside of that normal acceptable pattern of behaviour that alarm bells should start ringing.
“If you've got a customer that is always at an extreme end of what is risky, you don't want them to tripped up every time they try to make a purchase. The system is able to learn what is acceptable behaviour for that customer, and only alert when that customer starts to become abnormal.”
Online sports betting site Betfair has been piloting Featurespace's Snowdon solution to help detect and prevent fraud. According to Debbie Lawrence, head of fraud at Betfair, the system allows the company to gain an insight into each customer’s individual behaviours, predicting in real-time likely actions and detecting anomalies.
This provides significant assistance to the detection of malpractice due to the way fraudsters continually change methods to avoid detection.
“Behavioural analysis allows us to profile all online activity, understand how each customer’s behaviour evolves through time and act immediately when any anomalies are detected,” said Lawrence.
“With Featurespace, behavioural models are constantly – and automatically – refined, enabling the system to detect the latest and most sophisticated fraud.”
Find your next job with techworld jobs