By Martina King
Having spent the last two years immersed in the uses of predictive analytics and machine learning, I find myself questioning the future of mundane tasks completed by humans today. So naturally, on a recent flight, as the pilot calmly announced important information to his cabin, I sat wondering how long it will be before his reassuring voice will be replaced by the dulcet tones of a computer, albeit one that sounds just like him.
The role of pilots could change dramatically as machine learning transforms the workplace. Quite how much is uncertain, but a recent study from Oxford University suggests 47% of today’s jobs will be automated in the next two decades – that’s a revolution!
Prof Bill Fitzgerald, Head of Research in the Signal Processing Laboratory and a fellow of Christ’s College Cambridge, along with his Ph.D. student Dave Excell, spearheaded the task of automating the world through the application of Bayesian statistics to Adaptive Behavioural Analytics and machine learning problems. An introduction to their approach was recently posted on Uber’s blog, an excellent way to see just how incredibly complex maths can be turned into a commercial product, which simplifies and improves customer experience.
Bill and Dave developed and perfected an approach to anomaly detection: spotting subtle, real-time changes in potentially infinite streams of data to teach machines how to react, respond and learn. In short, they built an engine that sucks in data and churns out answers, dramatically reducing human intervention.
Many organisations have beaten a path to Bill and Dave’s lab with exciting challenges. For instance, they’ve used their software to attack commercial and social problems, the solutions of which would have been cost prohibitive, if not for their automated analytical approach.
It leads me to ask the following questions: can sensors costing £1m each (planes use loads of them) be replaced by clever maths (inexpensive replicable software)? Is it possible to establish the markers of harm, via data, to identify and help addicted gamblers? Can skin lesions be monitored for subtle changes, providing an early melanoma warning system? Can international retailers understand the behaviours of their individual customers, and provide the attention to detail that customers expect in a small community corner shop? Can sophisticated fraud attacks be spotted in real-time and blocked, even when there is no idea whatsoever of the type of attack that could happen and no rule exists to block it? Can you simply tell me what is going on in my data and find the value in it for me?
Establishing in a lab with commercial data that the academic approach is robust was stage one, the second part of the challenge was turning the academic approach to interpreting huge data sets – the algorithms and integrals – into a deployable commercial system. This is how Featurespace ARICTM was born. This engine is a robust and scalable platform capable of processing millions of events per second and that’s just the start.
The ARICTM engine has found commercial progress when applied to identify and block fraudulent attacks in the Gaming and Financial Services industries. It spots new types of fraud in real time, whilst improving detection of known types of fraud and reducing the number of customers inconvenienced by the blanket adoption of rules. Therefore, by having a real-time understanding of normal or expected behaviours, Featurespace enables businesses to spot unexpected transactions.
The engine is used in the same vein to promote Responsible Gambling, which identifies erratic or unusual behaviours indicative of potential gaming or gambling problems, and provides insight into the player health of all users. Featurespace profiles behaviours at an individual level, understands what constitutes ‘normal’ gaming activity for each specific player, and reacts accordingly to send the right message at the right time.
Now the know-how to make these advances exists and the system to deploy the ‘clever maths’ does, too. So, what can we expect over the next few years? The appetite for smarter analytics systems has gone from emerging to ravenous, and luckily there are several sectors whose data is perfectly suited for smarter analytics systems.
Take the retail sector, which employs some of the most talented Technology and Information Officers – always ready to try out the next technological advancement, from QR codes and Augmented Reality to new payment methods, a single customer view and RFID delivery solutions. The retail sector is ferociously competitive and tech leaders still bear the scars and crowns for how well they were prepared for the consumer shift to internet and mobile purchasing.
Finding a way to reduce cost and improve performance will always be high on the agenda for an industry where margins are fragile; hence, the overall desire to embrace technologies at least as fast as the consumer does, and to find new methods of automation.
Similarly, the Gaming sector—historically an early adopter of technology—is well placed to leverage the insights gained from this new approach to maths and machine learning. From high-street betting shops to major online trading platforms, the Gaming industry is using analytics to prevent fraud, identify risky behaviour, and market to customers as human beings on a personal, individualised level.
Lastly, the Financial Services industry will rapidly become the gold standard for the best implementation of data analytics. The Financial Conduct Authority’s mandate to put customers at the heart of company practices combined with banks immediate requirement to monitor what their subsidiaries and employees are doing, requires a cleverer approach than just employing more humans in compliance and data science.
Senior Executives tell us they would love to remove the need to worry about what is going on inside their data. They want a system monitoring all data, all of the time that highlights, with a priority score, what they should investigate for revenue growth and protection purposes. That’s precisely what Bill and Dave invented.
Adaptive Behavioural Analytics can identify fraud before it happens, never inconvenience a legitimate customer, and avoid costly and frustrating reacquisition efforts by cleverly retaining customers with a tailored approach to communication and they are just the solutions built for progressive executives looking for a more effective approach.
Pilot-less flight is possible today but are we prepared to trust the machine more than the human pilot? Possibly not yet but applying the machine to understanding our data is a comfortable next step. The revolution is being driven by organisations wanting to find clever ways to generate more revenues from their existing customers and manage their cost bases. What better way to achieve this than with fit-for-purpose algorithms that are downright futuristic – fasten your seatbelts!
