Since about halfway through my MBA, my career has been devoted to applied statistics. Since as long as I can remember, I’ve been super into math. It was my college major, and had I been smarter, I would have gone to a PhD program in math (either pure or applied); alas, my other passion in college was climate change, so my first seven years after college were spent trying to find a career working on climate problems. (And I burned out, hence the MBA).1
It turns out, though, that marketing science is one of the many engines of statistics (along with ecology, biology, finance, the social sciences, and more), and Wharton has some pretty serious players in this field. This is where I got exposed to the intersection of marketing and statistics; since the first part of my career was “industry”-focused and I had gotten unhappy, and since I realized it was a perfectly acceptable business career to write code and work on math problems, I decided to live at this intersection and make the next phase of my career functionally-focused (e.g., a career that would let me write code and let me work on math problems).
This is how I started Gradient. Originally it was just me, but over time it’s grown into a real, albeit still small, company. Revenue-wise, every year since I started we’ve grown somewhere between 35-50%, and overall the CAGR is 46% if you use the first full year (2017) as a base (and 66% if you use my truncated 2016 graduation year as a base). Most of what Gradient does is classic quantitative market research. It’s a mature, large, existing market, but it’s highly fragmented, and we win by simply being better than our competition. Since most of our clients are using our outputs as an intermediate product between the raw data and some strategic decision, “better” is defined as being fast, easy to work with, and generating confidence in our process and results.
We do better than our competition by doing three things:
Relentlessly automating the parts of the value chain that are automatable. A lot of this is recycling our code for analyses into encapsulated packages (in R), but there are lots of other areas we automate things, like using workflow to route issues through the correct stages and assignees, launching templated surveys, etc.
Using modern management practices, like agile workflows, retrospectives (kaizen), poke yoke. We use a mishmash of software engineering and Toyota Production System practices
Working backwards from the problem. I wrote a blog post on LinkedIn about how we think about this, which I call “model driven consulting”. The contrast here is with “data driven” consulting or “process driven” consulting. We certainly follow processes and collect a ton of data, but that’s not what’s front-and-center for us. We build the model first, before we collect the data, because the model determines what data we will need to collect. We build statistical models that model that data that we collect that are designed to produce specific outputs, like a rank-ordering of which features to build next, or the distribution of willingness-to-pay for a product across the US. Many of our competitors collect data and then see what they can do with it.
Who are our customers? We’re about a 50/50 split between working with end-clients directly and through strategy agencies. Strategy agencies are companies hired by (typically) larger companies to help them put together a coherent approach across product, messaging, engagement, and advertising for new and existing products. The kinds of outputs they produce are along the lines of “we’re going to target XYZ customer to meet ABC emotional and functional needs with product features LKJ &c, &c.” Our projects tend to culminate (although often after we roll off) with choices about which product features to build, how to price them, where to market them, and with what types of advertising.
Which type of customer do we prefer working with? We like both. Typically when we work directly with the end client they are growth-stage startups, (although some earlier stage as well) and so we get much more visibility into our impact on their business. But working with intermediaries is, overall, a better business for two reasons. The first is that it’s an effective sales channel for us (they typically have deeper relationships with larger companies than we do, and can provide substantial repeat business); the second is that this isn’t their first, or even 10th, rodeo. They’ve done projects like this before, they know what they need from us, and it’s typically substantially less effort for us to pull off.
Gradient has also been a great platform for my career, and I’m exploring ways to make it a great platform for every one of my colleagues as well. I asked on Twitter recently for help on how to think about creating an equity plan for Gradient, with the proviso that our business is much closer to a traditional partnership (e.g. law or consulting) than it is to a startup, so the format of our equity-sharing will be quite different than what most entrepreneurs are taught and think about. I have a rough idea in mind but it’s not ready to share internally or externally quite yet.
Part of the reason it’s been a great platform for me is that it’s given me a great window into observing and thinking about “model verticals”, or common business challenges that can be addressed with a statistical model. So far I’ve found and explored a few. One of them is Recast, which tells you how effective each of your marketing channels is through a highly complex and specialized model. Another is a brand we’re calling Appraise (which is, at least for now, under the Gradient umbrella) which is a model designed to audit the health B2C companies for investors and operators through the lens of customer profitability. In the next post, I’ll share how I think about these models as a business (what can make them good or bad) and some other verticals I’d be interested in exploring.
I’m now dipping my toes back in that water, developing a somewhat unhealthy obsession with tree planting (it’s natural carbon capture and storage!). One Tree Planted is the best nonprofit I’ve found if you want to support their efforts.