You’ve been pretty busy since we first met you. What have you been up to?The biggest achievement is that we’ve launched our second-generation platform. It’s much more scalable. We can handle a lot more brands and a lot more campaigns. It’s been a coordinated effort between the engineering side and the product side. Very exciting!
What’s been most exciting?
We launched very sophisticated optimization. It was a lot of work for the data science team, which was just two people. We did everything by ourselves — from deciding what data to collect to how to process the data to working with the engineers to coordinate data processing to developing the algorithm to testing out different algorithm performances to actually writing the production code. We’re quite proud of ourselves. [Laughs]
Let’s talk about that term "algorithm" for a minute. That gets tossed around a lot today. Can you tell us a little more about exactly what an algorithm is and how you go about creating one?
We log a lot of data, a lot of attributes, but it doesn’t come in a readily accessible format — the data has to be extracted. I work closely with the engineers on data extraction to make sure these are the data features that we can use for optimization.
What do you do with the data once you’ve extracted it?
Based on what we want to optimize, we have to formulate the problem. Is it a classification problem? A regression problem? Or something else? If it’s a classification problem, for instance, then there’s the question of how many classes you are trying to detect. And then you have to decide which of the many possible algorithms you want to use to determine the classification. Each algorithm has advantages and disadvantages. There are different approaches to the same outcome, so we iterate a lot in the R&D phase. And then, after all that, we will actually write our production system.
To use a terrible analogy, when you’re first extracting the data and deciding which way to go, is it sort of like “Jeopardy!” in that you have to figure out the answer you want first and then figure out how to ask the question?
Yes, it’s actually quite like that. That’s what we mean when we talk about how to formulate the problem. You have to figure out how to express the problem in a mathematical formulation.
That seems like it could potentially paralyze you with too many choices.
The choices are actually what’s interesting to me. PebblePost created an entirely new market with Programmatic Direct Mail®. There are a lot of things for us to sort out. For example: What’s the best way to do attribution? Data science is involved in a lot of these initiatives. That means I end up going to a lot of meetings. [Laughs] But it’s very interesting to work across many teams. Also, I’ve been involved in many kinds of reporting projects from the engineering side. That’s a new experience for me — to actually manage all the reporting-related projects.
Has that changed your perspective?
It has helped me understand the business needs a lot better. For these reports requested by the business intelligence team, I have to understand the query myself in order to understand what type of information they’re trying to extract and then try to explain the problem to the developers.
With all you’ve had going on lately, have you had time for any exotic trips like the African safari you told us about last time?
In May, my partner and I traveled to the Cinque Terre in Italy. That was really nice. We would take a hike in the morning and stop at a village for a nice lunch. Then we would hike to the next village in the afternoon and have gelato. And then we would take the train back and have a nice dinner with wine. I usually don’t drink much but over there I was drinking wine with every meal. [Laughs]
Where are you off to next?
Peru. We’re hiking the Inca Trail to Machu Picchu.
Machu Picchu sounds like the perfect destination for a data scientist who enjoys hiking. A whole different kind of algorithm went into creating that.
Yes. I read the book, “Turn Right at Machu Picchu,” to learn a little about the culture. And I’m spending my evenings climbing stairs to prepare.