Prattle CTO Bill MacMillan on Building 21st Century Data

  • Wednesday, November 30, 2016


  • Tell us about your background.

I feel like my background is–in a sense–pretty straight ahead. I got my undergrad at the University of Wisconsin-Eau Claire, my master’s at Iowa, where I received a lot of my training in terms of statistics and math, and then finished my Ph.D. at Michigan. After Michigan, I took a postdoc position at Washington University in St. Louis. I transitioned out of academia after a year at WashU and began my stint as a corporate data scientist at Monsanto and, later, Express Scripts. It was during this time that I also began working with Evan on Prattle and found the project so interesting I decided to take it on full time.  

The less straight ahead parts were that I have a strong background in construction, furniture repair, and manual labor generally. I was an upholsterer during undergrad and did stints working in home construction. My education was, largely, self funded off the back of my own labor.

  • What attracted you to Prattle?

The project was nicely aligned with the skills I had developed in academia and with what I had done professionally. As we’ve continued to grow as a company, I’ve evolved too.

My graduate work involved statistically modeling the actions of government agencies, and my first major project after finishing my Ph.D. involved natural language processing of government regulations. When Evan and I started talking about the idea behind Prattle, I found the central challenge at once familiar and fascinating: could I create a metric that scored a central banker’s language? I had done everything involved in that task up to that point, except that one exact thing. It was the logical next step in my own work…and a really cool business opportunity to boot.

In the beginning, Evan and I were simply colleagues who had presented at conferences on the same panel once or twice. It was just a great coincidence that we work well together as business partners. After I created the first set of scores for the Fed, the question was really “what now?” From there we started building a real business.

  • How has Prattle’s tech evolved since you began working on it, and where is it going?  

Initially, I treated it like a research project. It was an idea to play with and see if I could make something happen. At the time I hadn’t developed the skills to produce a full application that does something consequential. Instead, my approach at the outset was to build a method and see if it works.

We’ve gone through a couple of waves of innovation with the technology. At first the system literally took all night to run. Now the system can process and score a document within a few milliseconds. Getting there mostly involved really understanding the method and the limitations of the programming languages we were working with.

The next hurdles were systematic challenges: how do we grab a specific piece of information off a website (especially if we’re not sure about the skill of the website’s admin)? How do we do that quickly? How do we build a system that accomplishes that?

Moving forward, we’re now building systems that allow us to do the kind of work required to build novel, challenging data systems in a clean and efficient way. At this point, it’s much less about any one product or implementation, and more about how we can make a scalable, workable solution for rapidly developing new products.

  • How relatively complex is Prattle’s data production process compared to other text analysis companies?

A straightforward comparison is difficult because we do things differently than most of our peers. For one, the math behind the algorithm is novel and dialed in for a specific purpose: to analyze complex, nuanced, and veiled language. In addition, our approach to machine learning is structurally unique. Our methods have quite a few analytic solutions in the math (versus computational), cutting down on the support systems needed to produce the data. By increasing the complexity of the actual math used, we’ve managed to make a big data problem a much smaller problem in terms of scale–allowing us to have lightning fast solutions.

Also, other companies don’t try to tackle the gathering of information in the way that we do. We go for primary source material, gathered as quickly as possible, with little human oversight. Most companies that gather primary material that isn’t already clean (e.g. Twitter data) tend to need much more human involvement and rarely attempt something that is close to real time.

In short, what Prattle does is both more complicated and less complicated than other companies.

Ultimately, we’re just very different.

  • How can a central banking/finance novice use Prattle data to make expert insights?

Our scores represent a profound amount of information and it can be used in a number of different ways, but, simply put, you can just look at the data and, almost instantly, get an idea of where the economy is going.

This is great for novices. They don’t need to look at the FURBUS or have their own inflation model. Arm them with a chromebook and a Prattle subscription, and they can beginning spotting market-moving trends.

  • What excites you about the space you work in?

Despite its sophistication and the legions of brilliant minds it employs, the finance industry still has much room for improvement. It’s that opportunity that I find really exciting.

One of biggest challenges facing financial professionals is keeping track of the factors that shape the market. Frankly, it’s impossible. By collapsing the most important economic trends into a single signal, Prattle is doing its part to make that impossible task possible.

The Signal Team

Click here to Request a Trial