May 02, 2017 / by Dan / news / TAGS: AI, banking, ML, traffic

Experian make practical use of machine learning

Credit reference agency Experian hold around 3.6 petabytes of data from people all over the world. This makes them an authority for banks and other financial institutions who want to know whether we represent a good investment, when we come to them asking for money.

Like all financial services, they are being rapidly changed by waves of technological innovation sweeping through industry – none more so than artificial intelligence and machine learning.

Machine learning is essentially teaching computers to teach themselves – much the same way as humans can - by giving them access to huge amounts of data, rather than having to teach them to do everything ourselves.

Experian’s CIO Barry Libenson about how the business – a pioneer in Big Data-driven analytics – is adapting to meet the challenges and reap the rewards offered by the new generation of cognitive, self-teaching technology, and the ever-growing data streams which power them.

Libenson outlined three driving forces behind Experian’s move to be at the vanguard of the “fourth industrial revolution”. The first is new and emerging technology.

“There’s a movement towards open source technology which is less costly to operate and scales very effectively, so essentially you have a lot more horsepower at your disposal and can operate on much larger datasets.

“Just a few years ago when we did analytics on a dataset it was based on a smaller, representative set of information. Today we don’t really reduce the size of the dataset, we do analytics across a terabyte, or petabyte, and that’s something we couldn’t do before.”

Larger datasets obviously give a more accurate picture of whatever they represent, leaving less margin for error. This leads to analytics, simulations and insights which more closely reflect real-world outcomes – such as whether someone will repay a loan.

The second driving force is that it isn’t just the size of data which has increased, but its speed as well. Thanks to sensors, mobile phones and Internet of Things (IoT) technology, data is coming in at greater velocity than ever meaning insights can be based on what is happening now, rather than what happened previously.

“A decision based on information that’s a month old is not nearly as impactful as a decision based on data that’s a day, or hours, old,” Libenson says.

Thirdly, there has been a significant change in the way their customers want to consume the information they provide. The growth of software-as-a-service (SaaS), cloud based platforms and API solutions based on open standards has brought about a seismic shift in the way the data trade does business.

Libenson says “It used to be a very simple model, you would deliver a piece of technology to the customer, and they would install it and run it from their location.

“Now the large financial institutions don’t want you to dictate to them how they consume information, they want to tell you how they want to consume it, and you have to deliver it to them in that way. So they can say ‘hey, Experian, we have a simple question we want to ask and we expect a simple response, and we want it in real-time, and we want to be able to get it 24 hours a day from anywhere in the world.’”

The technological advancements which have changed the way it’s customers expect Experian to service them also provide immense opportunities for Experian themselves. Where machine learning excels as a technology for driving change is its potential for automating complex but often mundane and time-consuming calculations at incredible speed, using information from vast and quickly-changing datasets. These type of calculations make up the bulk of what Experian does in its role as a credit reference agency, with data coming from transactional records, marketing databases and public information such as court records.

One potential use which is causing a lot of excitement and is already on the way to becoming a reality is speeding up the traditionally lengthy and labor-intensive process of applying for mortgages.

“Machine learning is absolutely one of the hottest topics right now and something we are embedding into our products, in terms of better decisions and analytics,” says Libenson. In most of the world the process of applying for a mortgage has changed very little in the last few decades.

Libenson says “I’ve applied for several mortgages over the years and the only thing that’s different today from how it was 30 years ago is that I’m able to complete about 80% of the process using digital signatures instead of having to sign massive stacks of paper.

“At the end of the day though, in order to get that check and get the money deposited I still have to go to a title office and sign 50 documents so the money can be wired to the property owner – customers hate that, it’s a broken process.”

The result is that applying for a mortgage often takes weeks, or even months, due to all of the parties involved – buyer, seller, and financial institutions in the middle – having to collect and often notarize all of the necessary paperwork.

Experian is now using machine learning to look at the data elements that are most frequently needed during the application process, and learn how it can most quickly be located and passed to where it is needed. Through the probability-based process of simulating and modelling requests, information retrieval and distribution, it will become more adept at this over time as it gains more experience of the way all of the parties need the process to be handled.

“It looks at ways to simplify the process, to reduce the amount of paper used and also to get to a decision much faster … with this new technology we should be able to get to a decision in a few days, rather than weeks, and potentially much faster than that.”

The machine learning algorithms will come to learn what data is or is not important- “Over time we may find out we don’t need to care about five years of tax returns – but what we need is five years of credit payments.

“So we reduce the workload on one dataset and increase it on another,” says Libenson. “My guess is that we will be ready to roll this out in 2018 or 2019, and by 2021 or 2022 we will find the datasets we are using will be quite different from the ones we initially used.”

Experian is well-positioned to pioneer technology in this field, due to the fact that banks, insurers and public sector bodies all routinely come to it for its data. Packaging these lines of communication as another service is a great example of a business diversifying its service portfolio by mixing cutting-edge analytics with ever-growing volumes of data.

Libenson is clear that there is nothing to be gained by avoiding talking about the potential challenges to this wave of data-driven decision making – security being the most obvious.

“When you start talking about the changes we can bring about through technology like this, the one thing that’s critical is security being robust, and privacy being ensured.

“We’ve never seen such vigorous assaults on security and privacy – from nation states, hackers and organized crime which wants to monetize people’s information in its own way.

“The trick here is to not let the technology get ahead of the security that exists. We have to spend as much time and energy focused on making sure the ecosystem is secure as we do delivering the services – that everyone asking for information is entitled to it, and isn’t going to use it in an inappropriate way. Doing that on a global basis, around the world – because every country is different – adds another layer of complexity to it.”

It is definitely still very early days in terms of the change that AI and machine learning will bring to the way financial institutions do business. But certainly, everyone stands to benefit – consumers’ lives will be simplified and financial institutions will make decisions based on better and more relevant information.