Big data and analytics can transform how businesses operate. So what exactly is the reality of having a data scientist on board, or even outsourcing the work?
We interviewed Anthony Antoniou, our Head of Data Science, for his point of view on the day to day reality of what a data scientist is expected to deliver, and what actually interests him most.
Q: Why in general do clients come to you for outsourced data science services?
A: Clients come to us when they have a certain level of knowledge, but don’t have the skills or resources to undertake the analysis. Some are very good at phrasing the question they want to have answered, where others give us a much more open brief.
Q: So what is an example of an open brief?
A: A client who has a lot of data, and just wants to be shown what they can get out of it. A recent example has been a client with a large number of retail properties, who had several different sources of visitor data. We were asked to pull them together into a single short meaningful report for their board, which showed the trends in terms of the volumes and types of visitor, and where they were actually going to.
Q: And by contrast what would be an example of a very directed brief?
A: Someone who was trying to solve a very specific problem. We have a financial services client who was concerned about lapse rates, and wanted us to examine all the factors that impacted lapse. So we looked at how they were recruited, through which channel, what kind of people were being recruited, what product they were sold etc. Having looked at all the factors individually, we then looked at them multi-variately in order to find the combinations of factors that delivered customers with high and low lapse rates. The results helped the client a great deal in understanding where they needed to recruit.
Q: Overall what kinds of project are given to outsourcing data scientists?
A: Well you can break them down into reporting and modelling. The clients who want reporting tend not to have any internal data manipulation skills, so that they have to rely on external help. However, with more and more self-service tools being made available like Tableau for instance, we expect that the amount of outsourced reporting is going to drop. But having said that, there could always be a role for a data-scientist in setting up a dashboard so that reporting becomes almost continuous. Dashboards can in themselves be quite simple or complex like for instance the customer journeys taken on a website and how they are linked to conversion.
Q: So what about the ones who want a model?
A: Generally clients who want a model tend to be the ones who want to optimise their marketing spend. The focus is on getting better value for money. For instance because emailing out catalogues is so expensive, home shopping companies get very particular about who they send them to. This leads to us building response and value propensity models for them and challenging their traditional recency, frequency and monetary value models, although these can be pretty effective in themselves. To optimise marketing spend one needs to stick to areas where everything is measurable. This often favours direct channels over broadcast like TV, although there are techniques we use for getting to understand the impact of broadcast channels.
Q: So do you get to look at the effectiveness of social media?
A: We are just starting to do this. We have some results in from looking at the connection between buzz on social media and sales for a very large consumer electronics company.
Q: But what about paid for social media?
A: Well the interesting question here is what longer term value it delivers i.e. what sorts of customers, and how they play out over time. Our own customer data platform UniFida is helping us with understanding where web visitors have come from, and what value they deliver over time.
Q: How would you define ‘data scientist’?
A: Well it’s a much over used term but I see it as someone who goes through an end to end process involving all the stages of gathering, manipulating, analysing and interpreting data.
Q: And does that incorporate AI and Machine Learning?
A: Yes it can. The difference with AI (with the application of Machine Learning) is that it should be capable of unsupervised learning, where a model or process is in place that improves itself as more data becomes available. A classic example in the marketing space is where retailers make product recommendations based on changing consumer preferences. The potential for AI in marketing is where there is a continuous ongoing activity like the behaviour of visitors on a website.
Q: What interests you most as a data scientist?
A: Essentially where I can add real value. This can be very visible in certain areas like home shopping. Very often what one finds out has huge financial significance for the client company. I also like a challenge. We were recently asked by a radio music broadcasting company if we could predict which of their stars were rising and which falling so that they know whom to play. Fortunately, they had an app where listeners where asked to record what they thought about specific tracks. We followed the direction of changing approval or disapproval over time in order to get an early indicator of public opinion.
Q: Do you have problems getting the right data to analyse?
A: We often do, although some clients keep their data in a very orderly fashion. Also, we have been developing our own cloud-based customer data platform called Unifida, with the specific objective of pulling together in one place everything we know about an individual, joining on-line and off-line sources. This provides excellent structured data for analysis, allowing me to spend less time preparing data and more time deriving value from it.
Q: What advice would you give a graduate data scientist who was just starting out?
A: The most important thing is to work somewhere where their skills are appreciated, and if possible where they are part of a team so that they can learn from others. And yes, a good variety of work helps a lot too.