The Genesis of Gradient Metrics


At Gradient, we’re still inventing our story day by day, but a short chapter has already been written. We often discuss internally about our purpose and vision for the company, and we decided it became time to write down a few things about how we got to where we are today. Our history is core to our purpose, our values, and ultimately our operating principles.


Personal background

The short version of the story is that I’ve always had an interest in math — I was a math major at Pomona, got a master’s degree from the London School of Economics, and studied statistics at Wharton while getting my MBA. And during my MBA, I decided that I really wanted to “do” cutting-edge statistics full time, and that to do it the way I wanted to, I had no choice but to start a company and get clients on my own. 


Before Wharton I had a number of jobs — all related to climate change, which has been my pet issue for as long as I comprehended the challenge. Right before Wharton, a friend and I started a company to finance LED lighting retrofits in small commercial spaces; the company didn’t take off, but I loved the process of starting something new and wanted to get more training and resources in the private sector space. That’s why I applied to business schools, and I was lucky to get into Wharton. 


I came to Wharton without a background in marketing; I’ll be bold and say that it wasn’t until my intro marketing class with Patti Williams that I truly understood what marketing was all about and why it was so interesting (covering pricing, product design, and distribution — critical strategy areas for any organization, public or private). I was lucky to get assigned to Patti’s section (in fact, the luck keeps increasing, as she’s been an adviser to Gradient since even before day 1). 


The next semester I was lucky to take Eric Bradlow’s class (the second half of Wharton’s introductory marketing classes). Eric is a case in point of Wharton’s Marketing Department: he has a PhD in statistics, not marketing, and his research is as much about statistical methods as it about anything else, as his “research interests” keywords suggest: Bayesian computation, latent variable models, marketing research methods, missing data problems, analytics, psychometrics. 


Although I had to catch up in marketing, I was lucky to come with a background in statistics and the ability to write code. I had in the past written some Ruby (on Rails) and Clojure and at one of my prior jobs I had picked up R. (For those R nerds reading this, I worked with R before the tidyverse and the pipe operator %>%. Dark days, indeed). This gave me all the opening I needed to lunge headlong into the world where marketing research and statistics intersect. 


The beginning

My summer internship was at Venmo, where I was able to put a lot of ideas and techniques into practice. The projects I did there (factor analysis, segmentation, customer lifetime value) were fun. I wanted to keep doing that kind of work. I knew that every organization needed this type of analysis, and the ability to connect the business challenge to the technical solution was a major differentiator. Most data scientists don’t have a lot of exposure to the “business side” and for people that handle P&L’s, the reverse is even more true — most product/brand managers aren’t equipped to translate their hypotheses into an analysis or experiment. (Although that’s changing, fast). 


In fact, it’s such a rift that most organizations aren’t set up to have people that do both well. For example, the Big Three consulting firms (McKinsey, BCG, and Bain) and big analytics firms employ MBAs to be “translators” between their clients and the back office data science teams. And most data scientists at tech-enabled firms aren’t the ones wrestling with the core strategic questions for the company. I wanted to do both, and do both well — this meant building it myself. 


Before getting back on campus, I had resolved to really commit to getting clients and building my book of business. Academically, I started taking all the advanced statistics courses I could fit into my schedule. I started pitching myself to well-connected professors and anyone that would listen about the kinds of projects I could run for them, and within a few months I had some engagements (with Estée Lauder Companies and an early stage startup that would become Care/of). By the time winter break rolled around, I had enough evidence that I could make this work and decided to make it official. 


What’s in a name?

Of course, one thing you have to do when you start a new company is name it — so why “Gradient Metrics”? Well, the domain was available! (Just kidding, kind of…)


From the start I knew that the mission was to build a company that would be more (much more) than just me, so I didn’t want to include my own name. And having little budget to buy a single-word URL, it was clear that I was going to have a two-word name. “Metrics” “Insights” and “Analytics” were all contenders. To me, “insights” was too broad. I wanted to own the fact that we were going to be quant-focused. “Analytics” brought to mind dashboards of numbers, but not complex modeling. And that’s why I settled on “metrics” as the complement. 


And when I found that was available, I knew that it was a perfect fit. Beyond sounding appropriately “cool” for a firm that would work with consumer clients, the name itself had a meaning that aligned perfectly with our purpose.


So, what is a gradient? You probably know the definition that comes from physics:

“an increase or decrease in the magnitude of a property (e.g., temperature, pressure, or concentration) observed in passing from one point or moment to another.” (e.g., a color gradient). 

Or the definition from everyday life:  

“the degree of a slope.” (e.g. the gradient of a road)

But it also has a technical definition in mathematics:

“the vector formed by the operator ∇ acting on a scalar function at a given point in a scalar field”

Umm… what?


Proceed with Caution!  Math below.

Basically — if you remember your calculus, you’ll recall that a derivative (the rate of change) of a function (a relationship between an input and and output) is the same as the slope of a curve (that function plotted on a graph). Simply put, a derivative is what we call it if your function has a one dimensional input (e.g. the change in y with respect to x), and a gradient is what we call it if your function as a multi-dimensional input (e.g. the change in y with respect to w and x). 


In machine learning and statistics, a (very) common situation is to  model something of interest (e.g., will a customer churn? What is this a picture of?, etc.) with data (e.g. what the customer has purchased in the past, or the values of each pixel in an image). Your model is never going to be perfectly accurate; there will always be some amount of error. We quantify that error using different metrics (with names like the “sum of squared errors”, or “logistic loss”, and so on). As the computer updates the model, it often uses the gradient of the error metric with respect to its inputs to know which way to update all of the model parameters. Gradient descent, along with its variations and analogs, is an absolute linchpin of statistics and machine learning. It’s one way we find models that can use data to predict and explain the things that we see in the real world.


So the name “Gradient” fit all the criteria:

  • It had the right amount of consumer sex appeal to fit right into the marketing and branding world
  • It was a shibboleth to our other audience — technical data scientists — that understand the “other” meaning behind the name
  • A reasonable domain was available to purchase

A global company was born

I got the advice from one of my Wharton professors to wait until the New Year to incorporate the company so I wouldn’t have to deal with any tax-related issues for just a few months of the year, so on January 2nd, 2016 (I needed a day to recover from New Year’s Eve), I filed the paperwork to become a NY LLC, and a few weeks later, all the logistics were handled.


The mission was (and is) simple, but twofold: 1) help organizations of all sizes make better decisions using data and advanced analytics, and 2) to be a place where the smartest technical minds can do their best work. 


A few years later, I’ve convinced a few people to work with me, and we’re working with some of the most prestigious organizations around the world, doing cutting edge research using advanced statistical and machine learning techniques. I couldn’t be happier with where we are today or where we’re headed.