Why Businesses Fail: A Data Expert Explains It All
Data science is used in everything from medical research to mortgage applications, yet it’s not widely used when it comes to predicting outcomes for budding businesses. Thomas Thurston, the founder of Growth Science in Portland, Ore., has spent the last seven years building databases and algorithms aimed at finding commonalities between businesses that thrive and fail. The idea started while he was working at Intel and gained traction when the esteemed Harvard Business School professor Clayton Christensen invited Thurston to spend a year honing his research as a Harvard fellow.
Today, Thurston, 36, splits his time between working with large- and mid-size companies on new business development, and applying his research to investment decisions at venture capital firm Ironstone Group.
Entrepreneur’s Sarah Max chatted with Thurston about why entrepreneurs and investors should pay as much attention to numbers as they do their instinct.
ENTREPRENEUR: When you meet someone at a cocktail party, how do you describe what you do?
Thurston: I usually just say I predict if businesses will survive or fail. That’s a good shortcut, but usually leads to more questions. A data scientist, in my mind, is somebody who looks at data to try to find patterns. Have you seen the movie Moneyball? We’re that kid from Yale.
ENTREPRENEUR: What inspired you to study odds in business?
Thurston: I was at Intel about seven years ago and part of a new business that Intel killed for no apparent reason. I remember being frustrated and thinking that surely we could look at all the investments Intel has ever made to see patterns. As we expanded the data to companies outside of Intel we found these patterns persisted. That was when I spent a year at Harvard with Clayton Christensen. A lot of what we found dovetailed with the work he was doing.
ENTREPRENEUR: When did you strike out on your own?
Thurston: I came back from Harvard in 2008 and the group at Intel I was working with was getting reorganized. I left Intel on good terms and a couple of emails later I had a startup on my hands.
ENTREPRENEUR: And did you run your own startup, Growth Science, through your algorithm?
Thurston: I did, and in the first three years I did not like the answer. We were doomed. Like a lot of companies we work with, I kind of ignored it for a while because the business was growing and I was doing well. Then around year three, all the stuff I warn people about [drawing the ire of a large competitor] hit us smack between the eyes. We had to rethink the business model, and now, knock on wood, it seems to be working.
ENTREPRENEUR: Tell us a little about the key factors that go into your models.
Thurston: There are multiple tools, but they all triangulate onto an answer that is very clear. We tend to look at the strategy of the business, and that’s where we find the most predictive variables. What are some things about the strategy that offer predictions? If it’s a brand-new startup that has the best widget on the market, we find that those businesses fail around 90 percent of the time. Their odds are much worse than most companies. On average about 70 percent to 80 percent of businesses fail in 10 years. Yet, most startups have that strategy – doing something better than everybody else.
ENTREPRENEUR: This is, obviously, pretty counterintuitive. Do explain.
Thurston: If they actually are better, they start taking profitable customers away from their huge competitors. It doesn’t take long for their competitors to realize this and squash them like a bug. If you take the best customers of big companies, they have to respond.
ENTREPRENEUR: What strategies have better odds?
Thurston: There are many, but my favorite is a good counterpoint. It turns out, if you go to market with the worst product but it’s the cheapest, you can have anywhere from six to eight times greater likelihood of survival. Think of Walmart, McDonald’s, Southwest Airlines and, in the beginning, even Intel. That’s how they all got started. It’s a good place to start because the big companies won’t be as quick to respond if you’re not taking their best customers.
ENTREPRENEUR: Where do you get your data?
Thurston: We have tools that pull in data for us from the web. Sometimes we buy data, but that’s extremely rare. A lot of it we’ve hand built over time with thousands of companies.
ENTREPRENEUR: You teamed up with prominent investment banker Bill Hambrecht to launch a venture capital fund based on your findings. Google Ventures has followed a similar strategy since 2009, but the approach isn’t exactly mainstream. What do most traditional venture capitalists say about what you do?
Thurston: Most venture capitalists push back because their whole thing is their intuition and experience as human beings. Their data set, if you will, might cover a couple hundred businesses they’ve experienced firsthand, and they can only draw from two or three of them at a time. With data science you can go into the thousands. They’re going on their gut, and for some of them it works but for most of them it doesn’t.
Here’s another example: Teams are the number one thing VCs talk about. Yeah, entrepreneurs who’ve been successful in their first company tend to do better at their next one, but the difference between the best and the worst is only about 12 percent. In other words, the seasoned entrepreneurs are only about 12 percent more likely to survive. If half of your decision is based on the team you’ve wildly over-weighted that variable.
ENTREPRENEUR: What books do you recommend entrepreneurs read to get a better grasp on data science?
Thurston: Nate Silver just came out with The Signal and the Noise. If data scientists had baseball cards he’d be like a Babe Ruth. One of my favorite books of all time is Daniel Kahneman’s Thinking, Fast and Slow. He doesn’t mention data science at all, but the book is about cognitive bias.
The human mind is very, very good at some things, but it has limitations. The things we tend to be bad at, computers tend to be good at, and vice versa. I’m not saying only use algorithms to make business decisions. An algorithm can’t tell you the CEO is a total jerk and you shouldn’t work with him. But if you can put those two things together, you have much stronger predictions.