Figuring out what the next big thing will be before everyone else does is the pursuit of any business in wireless. You might be able to predict what phone a customer is likely to upgrade to based on their previous purchase, but what about when and where? Get into the details and suddenly it becomes much trickier.
Unless, that is, you harness the power of machine learning to do it for you – and can interpret the results. Brightstar’s data science team in Australia have been doing just that, helping determine trends from vast quantities of data that are changing how retailers manage and distribute their inventory.
“Data science is a new term for something that’s been around for a couple of hundred years,” explains Dr Greg Hill, Head of Business Analytics at Brightstar. “It’s the idea that businesses can be improved by thinking in a clear way about what the data is telling them, then looking for opportunities to make improvements, while borrowing from the toolkit of natural scientists.”
What is new, however, is the scale on which we can collect and understand data. Data science has exploded in the last decade due to the proliferation of data sets and the technology enabling humans – us biological bottlenecks – to comprehend the bigger picture.
Teaching a computer to teach us
The wireless industry stands to benefit from its advances as much as any industry. Some trends are clear. There are different kinds of phones sold in different kinds of stores (“We see that wealth effect where affluent areas buy top of the range phones,” Greg explains). But others are less so. Phones that perform better from a network coverage perspective do better in the metropolitan fringe areas where coverage may not be as good as in the central areas – and as Greg points out, those areas are constantly changing.
“In the past you had to have large teams of people with lots of specialized hardware to be able to do this sort of thing,” says Greg. “But now the growth in the power of computing platforms means that now a 25-year-old with a Masters degree and a $3,000 computer is able to undertake analysis that might have been a NASA level project in the past.”
That means a team of PhDs – Greg’s doctorate is in information economics – at a multinational company need to think bigger, and create computer programs to do some of that for them.
“We develop algorithms to understand the choices people making when buying a handset,” Greg explains. There are many of more these than you might consider at a glance, and with each comes a new branch in the decision making process.
“It’s a very complicated purchase when you go and sign up to get your new phone. You choose your operator, pre-paid or post-paid plan, you make a choice about what sort of store or retailer experience you want to have, online or through a call centre. We use algorithms to train statistical models to make predictions about what an individual customer is likely to do.”
Each of those factors will vary, by customer profile and location, so the retailers who can understand and navigate all these nuances can reap the dividends.
“We look for how quickly particular stores can sell through. We look at making a prediction about how quickly an individual phone will get sold through an individual store based on the previous history of that store.”
These decisions can be made on a micro level, not macro – making their impact felt that same day. “Based on all that, the algorithm makes a decision about how many in what colour to send to the store later that morning.”
Predicting the future of data science
For Greg, the rise of machine learning has been on trend that’s been easy to discern – he even switched careers in anticipation of it. “Around the end of the Dot Com Boom in the early 2000s I realised decisions were going to be led by algorithms in the future,” he says. “Hence the switch to academia and information economics.”
Greg sees another trend coming that he believes his team will be poised to help partners with. “Telcos are becoming increasingly customer centric,” he explains. “They are looking more and more at segments within their customer base and trying to understand how they can best meet those segments. We want to help our clients create those different customer experiences.”
That means more data and more powerful computers to crunch them. In the future, Greg foresees a time when realtime algorithms will be able to make live decisions directly for customers, without the need for analysts to check and interpret the findings, and generally keep AI on a leash. “A lot of the heavy lifting will be automated and done by machines – we want to get to that point,” says Greg.
In the meantime, he’s hiring, and not just in Australia, or for those with doctorates. “I go for people with a high degree of curiosity,” he reveals.
A McKinsey report (Big data: The next frontier for innovation, competition, and productivity, 2011) has warned of a shortage of data scientists by 2020. The Harvard Business Review (Data Scientist: The Sexiest Job of the 21st Century, 2012) has declared data scientist to be the “sexiest job” of the 21st century. It doesn’t take a supercomputer to spot the opportunity there.