Customer experience decisions can no longer be left to gut instincts. In the age of data science, companies are increasingly turning to data-driven insights to drive more efficiency and boost profits. A wide range of industries already use analytics to improve customer experiences and more will continue to do so in the next decade. At the same time, extracting value from data is expected to become more intricate and demanding.
It is therefore no surprise that data scientists are described as the "unicorns" of the business world. In fact, the U.S. will experience a shortage of 190,000 data scientists capable of harvesting actionable insights from Big Data by 2018, according to a report from McKinsey Global Institute. Here, we speak with four data scientists about their roles and how businesses should be thinking of data science as it evolves.
In this discussion, we feature insights from:
- Jared Bowden, data scientist at Adlucent
- Nicole Dvorak, data scientist at Forrester
- Michael Ferrari, head of data science at The Weather Company, an IBM business
- Mark Schwarz, vice president of data science at Square Root
1to1 Media: What do your roles and responsibilities include? What aspect of your job do you find most challenging and exciting?
Mark Schwarz: Generally, people have a misperception of data scientists -- most people think we spend all day running novel algorithms or looking at charts. In fact, it can be very different. Most of my work centers on clearly defining problems and using data to solve them. Data science goes beyond analytics; driving change requires compelling explanations and insightful iterations on the results. Being a data scientist requires me to effectively communicate information, and then turn data into actionable steps.
One of the biggest challenges in data science is determining how much change comes from data. Data has the power to change perspectives, but in order to prove effective, it needs to change behavior. While measuring how much a specific data set has changed behavior can be difficult and challenging, it's one of the most interesting aspects of my work. Typically, I spend about 80 percent of my time cleaning and preparing large data sets and only about 20 percent of my time using them to drive business forward. Even though it's not the most glamorous part of the job, it's where I can be creative and adventurous.
At Forrester, we were interested in evolving our data methodology beyond our bread and butter surveys and to employ a Big Data source. That required a whole new set of skills in terms of being able to clean, manipulate, and store that sort of data as well as analyze it. That's what spurred me coming on. I've been here two years and right now I'm the only data scientist but we're looking to add more people to the team.
What's interesting is how data science can be expanded to a lot of different applications. For example it can be applied to business analytics or streamline operations, make marketing tactics more effective, and inform what's working and what's not.
Michael Ferrari: I haven't held this role for too long yet [Ferrari joined IBM in March] so I can't provide many details but everyone knows the Weather Company has access to a tremendous volume of weather data in terms of geographical reach and types of variables. Where this gets interesting is blending the weather data with location data.
When we start to look at how certain times of day and weather might be related to foot traffic, we can start to build analogs or lookalike models to understand what someone who matches this profile of a user might do, such as where a shopper might go next. We can then ask detailed questions about how does that profile based on that user interact with vendor A as well as vendors B, C, and D?
We're fortunate to have access to the data that we have, which will let us do some really innovative work. But data is messy. Making sure that the data is in a format that's accurate and usable is a lot of work. We have a small team that's fantastic at doing that but I'm also looking for more people with the right skills to do this and more.
Jared Bowden: I use statistics and machine learning to make digital advertising more effective. In more specific terms, my role is focused on identifying patterns within very large datasets, and developing models that are capable of capitalizing on these patterns in ways that make our strategies more informed, and profitable. In terms of challenges, the process of building an ambitious solution can be a journey: Computers can be frustratingly unforgiving machines, but there's something intellectually satisfying about seeing your ideas incrementally realized through code.
Being a data scientist has been described as part analyst and part artist, why is that?
JB: Data science is a field of practical creativity-hard skills are necessary, but it's the innovative application of these skills that make unique solutions possible. I think it's fair to say that there's an art to the process of creative thinking and development, and I think many data scientists thrive when they are given freedom and space to test new ideas.
ND: A company could set up a model and keep running that model but it's like someone drew the picture and the company is just coloring it in. Data scientists are required to actually draw the picture. And that means you have to know what sort of metrics you're looking for to measure and what kind of research approach you are going to take to get the answers your boss is asking of you.
That requires some creativity. There are so many different data sources and ways you can combine and analyze data that the approach is the creative part. Of course you need to have the tools and the skills and the statistical competence to be able to do it. But also you need to step back and think about whether your approach is the best one.
I'm responsible for communicating in terms our clients can understand. Part of the consideration is what will make sense to the client, where will they derive the most value, and what's the data point that will stick? How do I help them turn an insight into action? All of that involves creativity.
MF: I absolutely believe there's an art to this. Just having the technical skills doesn't mean you'll be a great data scientist. At the end of the day, what you need in this space is curiosity. If you take 10 people and get 10 different approaches, that's a good thing. We want to have people who are looking at problems in different ways and looking at how to organize the data in different ways.
MS: There's definitely an artistic element to data science. It requires technical skills, creativity, and insight. The mix between analyses and artistry comes when interpreting heavy volumes of information and large data sets, and then structuring problems to create actionable next steps. For instance, our store relationship (SRM) platform, CoEFFICIENT, collects feedback into behavior changes, and with its insights metrics, we can measure how and which changes have been helpful for our buyers. The entire process is certainly part artistry and part analytical.
What are the types of data that you work with and can you give an example of a predictive model you've made that influenced the customer experience?
MS: One example is a predictive model for a corporate business unit in auto. The CoEFFICIENT platform captures detailed, specific data about how corporate programs are being rolled out in the field. The analysis helped confirm that the group's annual goal was attainable. More specifically, it used data in CoEFFICIENT to explain which field actions were the most important predictors of growth.
The model gave the corporate team detailed collateral to use in internal and field presentations. We helped them take a directionally correct gut call and turn it into structured metrics to watch with their field staff. Most importantly, we were able to close the action loop by showing how the plans were progressing during the year.Results were available on-demand in revenue, predictive metric, and qualitative form.
ND: Forrester is using data in a less traditional sense since we're doing more ad hoc analyses. We are still operating within the sphere of computer programming and statistics and Big Data but it's not so much optimizing marketing campaigns as looking at what sorts of attitudes and preferences drive more engagement in banking apps. We do broader research topics like that.
And so in addition to survey data, we're experimenting with data capture methodologies including behavioral data for research purposes. We're also using online communities as a type of qualitative data to help prove the "why" behind the data.
JB: Adlucent's Data Science Team is currently focused on developing products that will make advertising a more personalized experience. Intuitively speaking, we understand that people are more receptive to information when it's relevant to their interests. Within the context of causal everyday conversation, relevancy can be thought of as the abstract ability to "read the room." When you want to maintain this relevance at scale, and serve attention-worthy information to millions of people, this uniquely human ability requires a programmatic analogue.
The products we've developed to meet this need use a range of different machine learning techniques to identify the demographic and psychographic characteristics that motivate our clients' most valuable audiences. Once these patterns have been established, we can activate this information by building secondary models to retarget high-performing audiences with relevant content, or expand our scope to target new audiences that share a sufficiently similar set of interests.
MF: [See answer above]
How do you see the role of a data scientist evolving as more companies use data to optimize services and products?
MS: As more companies require new and more advanced technology, demand for data-based research and analyses will continue to grow. Data science requires deep knowledge, scientific rigor and understanding to drive action and change behavior. In the future, as data preparation becomes less messy and more routine work, the role may break into a more specialized force. I think the data prep space is slowly getting more structured and more routine. Over time, say 10 years, data science teams (engineers included) will spend much less than 80 percent of their time prepping and cleaning data.
JB: Data-based decision-making is the new normal for business. The ways that we use this data will continue to evolve, as will the perception of data practitioners. In general, I think we've found that the field is too dynamic to pin down, and the skill set is too diverse to delineate. And that's fine; I tend to believe that "data science" is a term that was appropriately coined. Scientists are people who have the background required to answer unique, typically quite specialized questions. We acknowledge that there are different branches of science, and there are different ways to answer scientific questions. It's mostly semantics, but like to think that "science" will continue to be a valid way to describe what data scientists do.
MF: I hope the term data science goes away in a few years. It's industry parlance, but I don't think it's representative of what we do. Data science is so broadthat there's no such thing as this unicorn data scientist who can do everything.
It's a fast moving field and there are so many different components ranging from hardcore engineering to the softer sciences which still have to have that quantitative frame of reference. So I'm hoping it evolves into sub-disciplines that can all talk together but don't necessarily have to be the same role. I used to try to build a team of people who could do as much as possible, but I've learned it's more important to build an interdisciplinary team with people who have certain skill sets and the ability to think creatively.
ND: Harvard recently launched a data science program and MIT Sloan is starting one in September. Universities in other words are realizing this is an important skill set to have. And as more programs become available, I think data science will become more ubiquitous and may evolve into distinct jobs. You might be a data scientist specifically for marketing science or for operations. Then again, a lot of the skills and tools are the same, so there will still be a lot of overlap. It'll be interesting to see what happens.