From automation to big data
Automation has changed everything: how we pay, how we work, how we live.
50 years ago, a retailer might have operated a very basic till to take payments. The company’s accounts would have been written down in a ledger. The stock take would have been done at the end of every day or week. All the information pertaining to that business would have been disparate, unconnected and hard to analyse.
Today, for many retailers, the payment system is partially automated. It’s operated by a human, but also connected to a digital stock inventory and accounting system. This generates a lot of data because it records how many products are sold daily, what times purchases are made, what products are popular, which items are purchased together. Businesses are now able to generate and record huge amounts of data, it’s known as big data. Making sense of it is the challenge.
Harnessing big data
Online retailers were some of the first to use these large customer-generated datasets. Do you remember your first eBay or Amazon purchases and those ‘customers also bought’ ads? This was one of the first mass applications of big consumer data. For the first time, retailers were able to profile shoppers in real time and predict what other items they might like based on the preferences of other customers with similar profiles. This system was ground-breaking at the time, if not a little rudimentary by today’s standards.
The possibilities of machine learning go far beyond predicting which customers might like a particular item of clothing though. At Barclays, we’re using the technology in a number of different ways, from fraud detection to a better understanding of the needs of our customers. You need two things to build insights: a big set of data and an algorithm. An algorithm is how a machine learning system treats that data, what it learns from it.
At Barclays, we’re able to build a specific profile for each of our credit card customers. This is important when it comes to spotting fraudulent activity on their accounts; if we know how we expect a card to be used, it makes unusual activity easier to spot. Using data for each customer and secure servers, our machine learning systems spot patterns in their spending, whether it’s time of day, location or amount. We then create a unique spending profile for each customer.
Machine learning goes one step further, it also takes into account the time of year, the weather and a whole host of other variables which can affect how people spend. What’s more, as more data becomes available, machine learning algorithms can adapt and adjust to data patterns built over time.
If a customer then makes a purchase which is out of the ordinary, our systems have already learned to a very high degree of accuracy how likely it is that the purchase is fraudulent or genuine. Using machine learning, we have also developed a series of ‘fraud profiles’, which are specific sets of behaviour often displayed by people committing fraud.
Using machine learning in this way benefits customers because it can prevent financial loss, but it also stops ‘off-profile’ payments being needlessly delayed, denied or investigated because so much data is used in building a profile. Similarly, there’s a great benefit for merchants who have the peace-of-mind to know that our systems can detect fraudsters, giving them greater protection against financial loss when taking payments.
Machine learning isn’t just making positive changes for consumers today, it will also help shape our future…
The ability to learn from huge sets of data can help everyone. For example, The Internet of Things used to refer to connecting your TV and lights to your mobile phone. Now we have smart cities where everything is connected, from smart meters monitoring traffic flow to street lights that react automatically to daylight. When you start measuring variables in entire cities, you can optimise traffic flows, reduce energy use, even ensure public transport runs on time. Thousands of sensors act like human senses and machine learning is the brain that makes sense of it all.
Spotting or preventing serious illness could become much easier with the use of machine learning. The Prime Minister Theresa May recently spoke of the opportunities to detect cancer earlier if the vast patient data set held by the NHS can be harnessed in the right way. If it’s done correctly, tell-tale signs, lifestyle choices, age, sex, location and family history can all be used as data points by computers to learn about risk and allow early intervention. The target in the UK is to diagnose 50,000 people a year at an earlier stage of the illness by the 2030s. That’s a huge number of people with a better chance of beating cancer. Imagine how significant the applications of this approach for other illnesses could be.
The world of finance is another with large data sets and a huge number of variables in play at any one time. Recent research by Barclays found that 62% of systematic hedge fund managers are already using machine learning techniques within the investment process. The key advancement in recent years has been how machine learning can produce models with a high degree of accuracy despite imperfect information scenarios.
In financial markets, there are lots of variables like the mind-set of investors, for example. This is something that is hard to quantify though. In these scenarios, it becomes a matter of probability that the algorithm is correct, rather than a binary ‘yes’ or ‘no’ – but the degree of certainty that the algorithm is correct is increasing.
The applications of machine learning are huge and where there is data, there is an opportunity to find new patterns, to learn and to predict future trends. Whether it’s improving customer service, bolstering security, improving diagnosis of disease, or making our cities greener and more efficient, machine learning isn’t the future, it is making a positive difference now.