Get to know our team at UpStream. This week we talk with Statistician Tess Nesbitt, PhD, about building statistical models at UpStream.
How do you see retailers using UpStream to change their marketing?
Because we build statistical models at the customer level, we are empowering marketers to customize their strategies at this very same granular level. Many of today’s existing solutions have the tendency to aggregate out individual effects and lose the granularity that customer-level data embodies, which restricts marketers to strategize at the macro level. Our methodology facilitates both the micro and macro level analyses in tandem for the most comprehensive insights that a marketer can extract. With rapidly advancing technology, marketers are already starting to move toward tailored, customer-specific marketing treatments (for example, retargeting) and Upstream enables marketers to trek down this dynamic marketing path with more than just a bird’s eye view.
What success are you most proud of?
I think one of our proudest accomplishments was designing methodology and writing code to build sophisticated, multi-level statistical models with big data in the cloud at breakneck speed (comparatively). When I was able to build a model on a few GB of data in less than 10 minutes, I was in nerd heaven (everyone loves the law of large numbers right?). Back when I was in grad school, most of the available statistical computing packages would huff and puff while turning the cranks for hours before failing to spit out a model on datasets of such epic proportion. Modern computing has been given a major dose of steroids that opens up so many doors for an analyst. The more data we can use, the more powerful and stable our models can be, the more data we have to test our models with, and the more we can ensure generalizability.
I think we have also been very successful at translating the business problem into the language of statistics in the front end and then translating the statistics back into the language of business at the tail end. Statistics is not a universal language, but we take pride in this translation. In my opinion, which is completely unbiased of course, our attribution work is a delicious and innovative blend of theoretical statistics and decision theory that is applied in a new way to marketing analytics.
Describe UpStream’s company culture and the role that plays in development.
While we definitely know how to put our noses to the grindstone, let’s just say that it is not uncommon for our boss to surprise/treat us with an occasional Britney spears tune on the loud speaker (it’s amazing how she can get your cerebral juices flowing at 5pm). Although sometimes this doesn’t mix with the ethnic tunes emanating from the German brew pub on the first floor of our building. All joking aside, Upstream’s diverse mix of statisticians, developers, strategists, managers, and hybrids of all such sorts allow us to cover every element of the analytical food pyramid. Our work is extremely collaborative and allows every participant to flex his/her strongest muscles.
What is the single thing you love most about the work you do?
Well, nobody becomes a statistician for the celebrity status. We love the challenge of making the unexplainable explained. In a world where we are tracking (and being tracked) and learning more and more about consumer behavior, the terabytes of data will continue to accumulate at our fingertips. The challenge lies in developing methods that can effectively and efficiently scrub, manage, model, extract invaluable insights, and weed the noise out of giant databases. Most importantly, the statistics must yield actionable and interpretable results. Achieving this right balance between sophistication and simplicity is always a challenge. At Upstream, not only do we get to the wear the hat of a statistician, but we also get to wear the hat of the decision theorist who must search the grid of the most profitable implementation of our modeling work. Combinatorics is a beautiful beast. :)
I know this question said "single" thing, but I’m sharing a couple "single" things. One of the most intriguing aspects of my work is the people that I get to interact with. I have had the privilege of learning from and in turn imparting and/or expanding knowledge with some of the biggest players in the retail marketing game. As a statistician, my training in graduate school was very general, and being to apply it to a niche that can use it as a vehicle to capitalize on future prospects, extract more value form the current customers, and more efficiently allocate budget across marketing channels and campaigns is exciting and rewarding.
What makes UpStream different than other attribution solutions?
At the risk of sounding like a broken record, Upstream’s attribution has the ability to put each customer under the microscope so that interesting effects are not aggregated out or lost, as is often the case with macro level analysis. We are not only building models to explain the world in which marketers live, but we are also striving to use these models as the first step in a multi-step process to guide marketers down the path of prospective optimization. It’s easier (comparatively) to analyze retrospectively, but looking forward and determining a customized path to the greatest return in light of dynamic marketing is somewhat unchartered territory.
