This year’s conference programme for DataIQ Future was rich in insight about how organisations are adopting the new techniques of artificial intelligence (AI), machine learning and data science. Presentations across the day proved that these are no longer just academic ideas, but value-driving (and cost-saving) techniques for business.
What emerged strongly from the day was that there are common issues and opportunities presenting themselves from the changing data and technology landscape. Here are five that stood out.
Adoption by business and by customers is not the same thing
Artificial intelligence and machine learning were a thread across most of the presentations, as organisations look at the most insightful and efficient ways to explore huge data sets. This desire to adopt powerful techniques is important, but it can put the organisation a long way ahead of what its customers (or society more generally) understand and are willing to accept.
Scott Gallacher, interim head of AI at Department of Work and Pensions, described how one of DWP’s experiments had surfaced this dichotomy. “We recently did a piece of work on blockchain as a way of delivering benefits. That got perceived as a conspiracy to stop people spending their benefit on other things, like alcohol and cigarettes. We took a hit for experimenting,” he said.
In the public sector, the scale of the task means it has to look for better working methods, but equally faces much greater sensitivity and visibility than in the commercial realm. “For us, if a system goes down, that lack of availability impacts on benefits. So reliability is paramount.When we are making decisions on people, we need the best level of certainty – not just a yes or no. It is about making it as beneficial for them as possible and trust is paramount,” explained Gallacher.
Forecasting which of the current data and technology propositions will gain widespread adoption and be sustainable is difficult. As he noted: “We are planning for 2020, but who knows how things will be by then? We believe if we have certain things, like big data that is accurate and accessible, we have a decent chance of being successful whatever the situation.”
Maths keeps moving
Alice Jacques, senior data scientist at Channel 4, made a simple, but telling point: “Academic research doesn’t stop at the point you leave university.” Unless you make a real effort to stay informed, there is a risk that what you know about a discipline gets fixed when you graduate. But research and innovation is the very lifeblood of academia and new ideas, techniques and technologies are emerging all the time.
“Data scientists need to read academic papers and go to academic conferences, which are like speed dating for scientists. If you come away with zero ideas, it was the wrong event – you should pick up five or six,” she said.
This constant pace of development was clearly outlined by Clive Humby, chief data scientist at Starcount, in the way data sets have evolved and the techniques needed to understand them have similarly expanded. “Geodemographics dominated for 15 years, but there has been a move towards actual data, then social and open data, and now the internet of things,” he said.
Each of these requires specific analytical techniques which mathematicians have been busy developing over the past decades. Actual data assumes that the past is a predictor of the future, while descriptive techniques allow groups to be identified and clustered, explained Humby. As data volumes have grown, however, the number of gaps in records has also increased, leading to the need for regression techniques to model what those variables could be. In the most recent development, graph theory identifies links between data and is powerful when interpreting social networks or relationships within messages like email.
AI and machine learning have been in use for academic research for far longer than their more-recent application to commercial challenges. That is why knowledge transfer between academia and business has recently been accelerating. But Gallacher sounded a note of warning: “A lot of academics in the machine learning are used to getting relative uncomplicated and complete data. That is not the case in the real world.”
You need to build an academy
Embedding the skills of data science into the organisation is not simply a matter or hiring the right person. For one thing, the whole organisations needs to be capable of explaining its needs in the right way. For another, it has to learn how to absorb the ouputs. Plus, that data science team needs to keep its knowledge fresh. The whole organisation needs a workout.
Unilever has recognised both sides of this problem. “People were coming to the insight and analytics team and saying, ‘we need analytics, AI, cognitive computing’,” said Kjersten Moody, VP of insight and analytics. “We started thinking about how to support them, the tools that would require within our team and also for the recipients of our outputs.” The function also wanted to change the language in the business from that of business intellligence to that of analytics.
Moody applied product design thinking to the creaton of a learning and development programme for the business. “That was very new inside the organisation,” she notes. Over the course of eight months, she put together a self-learning programme to support both the analysts and their customers. “We have a global group of 60 who self-organised themselves into ‘The A-Team’ to promote the new culture within information and analytics and beyond,” says Moody. Her tip for success with an academy is, “create the culture and let it surprise you.”
The Office for National Statistics has embarked on a major project called The Data Science Campus which has a similar goal. “ONS had a skills gap,” explained interim head of data science, David Johnson. “An external review by LSE said our skills wouldn’t meet the needs of the organisation in five years time and that we needed to ramp up our data science.”
The idea for the campus is to ensure both that its analysts keep abreast of skils, techniques and applications via a Masters in data science for public statisticians, as well as to transfer understanding of this area into the ONS and Government. “It is about using data science as a toolkit,” said Johnson.
Channel 4 is staying on top of changing techniques via a partnership with University College, London. As a result, it has already recruited two PhDs and four data data scientists. “One of the benefits of the relationship with UCL is that we pick up ideas which are out of our comfort zone,” explained Jacques. By ensuring a flow of ideas, the broadcaster hopes to avoid intellectual lag and other risks. “We don’t want to inheret technical debt,” she noted. It also takes steps to keep its data scientists intellectually engaged by running counter-culture competitions. “It gives them the feeling they are noticed and they are getting their name out there.”
Invest to make a difference
“It takes between one and three years to launch a new service. But for a market leader, how long does it take to get disrupted? If you blink, you’ll miss it,” warned Amit Pau, managing director of Ariadne Capital. He pointed out that, in 2007, Nokia had a 70% share of the market for mobile phones, but Apple had just launched its iPhone. “In the tech market, winner takes all if they get it right.”
Pau also made the point that, “everybody wants to be an entrepreneur now”, not least because of the sums investors are looking to place into technology start-ups. Softbank has a $100 billion technology fund, for example, and recently bought ARM for $21 billion. Corporates are also looking to acquire useful new data technologies for strategic reasons, rather than financial returns.
This is having a trickle-down effect as smaller investors recognise the heat in the market and follow it. Ros Singleton, chief operating officer at broadband provider Relish, described her own experience as an individual investor who joined Angel Acadame and placed funds into the data science hub Pivigo. For her, one reason for the decision was gender politics. “Only 10% of VC funding goes to women and only 7% of the senior investment partners in VC firms are women,” she noted. “You need both sides of the coin, not least because balanced boards achieve a better return on investment.”
Beware of zombies!
Humby had a warning about becoming complacent or even trapped in one way of operating: “Organsations can become trapped by the business model they have put in place because their data focuses on a single thing.” He cited customer churn as an example of a metric that is given high importance. But in one organisation he worked with, it emerged that customer service agents were struggling to transfer customers from one service to another, so they simply closed an existing account for the individual and opened a new one. What looked like churn was actually a broken process.
It is not just ideas and technology that can become zombies, it is also customers. John Belchamber, global business intelligence lead for Telefónica, noted that, “Pokemon Go has turned people into ‘phone zombies’. It has changed the way people walk around cities – and also the traffic on our network. The game now has 30% higher phone usage than Facebook.”