Quantifind co-founder John Stockton leads the development of methodologies that explain how business decisions translate into financial results. His research on extracting collective information from unstructured data guided Quantifind’s early technical development, and his ongoing work to bridge machine learning and signal extraction continues to shape the company’s roadmap. He earned his BS in Physics from Stanford University and his PhD in Atomic Physics from Caltech, where he was a Hertz fellow.
We sat down with John to explore how Quantifind’s methodology works and why explanatory analytics helps marketers drive revenue with the right approach to data.
How did you go from physics to a startup doing strategy for brands?
First, I blame Ari Tuchman. As cofounders, we tend to jokingly blame each other for a lot of things, but he’s the one who convinced me to join him in leaving physics for the startup world. I also blame the National Science Foundation, since they enabled us to explore our collective data mining ideas that eventually turned into Quantifind. They called our bluff when we applied for funding back in our whiteboard days.
Seriously though, we’re glad they did. It’s been an honor to have had the chances we’ve had, and if you look at a lot of Silicon Valley success stories, it’s a fairly common path for scientists to turn small technical grants into products that have huge impact on the business world, which is what we’re trying to do at Quantifind.
Also, physicists love a good revolution. We get excited when old models break under new situations and open up the potential for something new. This is obviously the case in the world of brand analytics, with the boom in available data, machine learning interest, and general receptivity towards data-driven strategy. Old models aren’t cutting it. It’s time for something new. That’s exciting.
What about having a physics background prepares you for something like marketing analytics?
Whether you’re in physics or strategy, it’s all about finding a signal to direct your future actions, and not getting distracted by noise. In the quantum physics realm, where Ari and I did research, there are some funky things about how you estimate a system that are obviously quite different than the funky things you find in consumer analytics. But signal extraction is the core idea they share. It turns out many of the same approaches apply quite nicely, and those approaches are opening up new possibilities in marketing decision-making.
The main difference is that in brands, it’s culturally less common to use technology to help drive decision-making. Marketers face difficult decisions every day, decisions that are best guided by a lot of factors, some they can control and some they can’t. The question is, how do you navigate through bad weather? Do you stick your finger in the air and trust instinct? Do you learn to read new instrumentation and trust in technology to guide you? Or do you find a happy balance among instincts, intuition, and instrumentation?
Also, as much hubris as physicists sometimes have to run roughshod into other fields, one of the things I like about the team we’ve assembled at Quantifind is that we’re all pretty pragmatic and humble when it comes to understanding the limits of technology. We have a great team, not only from a data science standpoint, but also in design and product, and we all understand that our job is to know how far data can take us, and when to augment human intelligence as opposed to replacing it.
That’s where explanatory analytics come in. If the humans in the loop don’t understand the recommendations coming out of the machine, then you’re dead in the water. If they do, then you have a chance to make a difference for their strategy.
What does “explanatory analytics” mean, and why is it valuable?
Explanatory analytics allow you to understand how and why business decisions translate into financial outcomes. This accomplishes two things: You can validate the success of past actions by discerning which decisions drove revenue, and more importantly, you can use this information to guide future actions.
For example, if you’re a marketer, you can understand which factors are driving revenue among certain demographics. A beverage company accustomed to seeing sales peaks in the summer might know that sales spikes haven’t been as high over the last few years. Why is there a decline in the average spike? Which messages or demographic groups should the company target in the future in order to grow revenue? An explanatory approach to analytics can answer these sorts of questions.
Explanatory analytics also mean marketers can explore various correlations and test their human intuition against the data. It’s more than just tracking data; they can also start to discover new opportunities. And because they understand why things are correlated, they have more confidence in actually acting on the data.
In short, explanatory analytics allow a user to understand the stories hidden in their data and to take confident actions, all without resorting to technical jargon or black boxes. Businesses need to make strategic bets based on data, but there is no way a business will make such a move without the trust that comes from understanding how a data insight translates into strategy.
Predictive analytics is generating a lot of buzz among marketers. How does explanatory analytics compare?
In overly simplistic terms, predictive analytics tell you what’s probably going to happen if circumstances don’t change. Explanatory analytics tell you why things are likely to happen so you can change the outcome.
Predictive analytics can be lots of things, but in the context of improving marketing strategy, it’s often good but not sufficient. If you’re not careful, predictive analytics leads to passivity. People say, “This is what’s going to happen” and don’t always feel empowered to try to change it. But the whole point of analytics for marketers is to help them make better decisions. That means we need to set up an active feedback loop where human instincts and intuition play a role, where things can be understood and explored, and where outcomes can be changed as a result.
If you’re in a plane that’s crashing, you don’t want to accept the outcome— you want to figure out what’s wrong, fix it, and pull up. Likewise, if you pull off a miraculous landing that saves hundreds of lives, you probably want to know how that happened so you can do it again. Explanatory analytics is similar. We often say that we want our earliest prediction to be wrong, not because the model was bad, but because our insights inspired an action that helped our customer improve their fate.
Who should be paying attention to and using explanatory analytics?
Anybody who is in a strategic position to make big decisions and who is overwhelmed with data and doesn't know how to map that data into their strategy.
Quantifind works mostly with CMOs and marketers right now because they’re executive enough and operational enough to be in that sweet spot. But we’re not just throwing one tool to one person. People work in teams, and they all need to speak the same language. As we grow, our users will expand to include other executives outside of marketing, up to the CEO. We also partner closely with in-house analytics and social media teams, which is critical because our solutions help reveal the full business value of the data they’re working with every day.
What are some of the other problems that explanatory analytics solves for marketers?
Marketing is all about growth— about big initiatives to channel bigger and bigger groups through the funnel to purchase. To identify new initiatives, we look deep into the purchase behavior of consumers within a vertical and systematically identify high upside creative or targeting opportunities.
Historically, it’s been almost impossible for marketers to know which factors are driving revenue and growth, and how to maximize those factors. Is a marketing campaign connecting with and converting consumers? Is there an untapped opportunity? When consumers post about your product online, are those conversations related to sales? Traditional analytics techniques haven’t been able to reliably answer these questions.
Explanatory analytics changes this by explaining, for example, how certain data sets - whether social media posts, call logs, or some other source - are reflective of larger consumer purchasing behavior. We want to democratize clients’ data. The answer to a mystery can come from anywhere. Terms that consumers use in call log data might be the canaries in the coal mine that alert us to some undetected market change, or some untapped opportunity. Because we tie this back to revenue, the degree of color and insight you can get through explanatory analytics with new feature-rich data sets is unprecedented.
Does a marketer have to be a data expert to use explanatory analytics?
No. The product should speak the language of the customer. The human shouldn’t have to learn the machine’s language. It should be the other way around.
How does the product speak the language of the customer? Does that have to do with the metrics used?
It’s often important not to focus on just a metric. It’s more about the story. Numbers are necessary but not sufficient to form a strategy. The metric needs to fit into a context that includes where the company has been in the past, what the competitive landscape may look like, and what the potential elasticity is for that metric. Language that describes upside and opportunity follows from this.
What have you learned as you’ve worked with customers?
We've learned a lot from the different verticals we've played in. Even going back to the initial days, doing defense and technology analytics taught us how to effectively present "collective intelligence" from a wide array of sources.
When we started working with movie studios, informing highly adaptive creative and targeting strategies, we learned how to be fast, because the clock is ticking prior to the release date.
Working with large brands has driven us to be very selective in which opportunities we present to them. It's actually an amazing search problem: given all of the possible moves that a CMO can make, which one large investment is worth making to drive the most revenue? What is the risk/reward profile for the set of moves? It's like very high dimensional chess.
How do companies integrate explanatory analytics into data-driven cultures?
Carefully. Every company is increasingly willing "in principle" to be more data-driven, but it's complicated. All companies are different. The groups that own data differ, the groups that own certain decisions differ, and sometimes it's chaotic within the culture of a client.
The saving grace of explanatory analytics is that it is like a "translation layer" and lets the different groups start to speak the same language. As a company, we usually start closer to the people who make the larger decisions. But it is just as important to give interactive explanatory tools with more detail to the analyst groups within brands, because they own the trust. We know it's working when the different groups within a company are sharing and discovering stories across multiple levels.
Two last questions to close up. What are you working on now that really excites you? And what’s your vision for Quantifind’s future?
We want to put more fuel on the fire from an intelligence perspective. This means more data that can power more use cases. So we’re working on building a platform that can take on a lot more variety.
The goal is to understand the driving factors for one system such that you can recognize future opportunities to maximize return. The thing is, there’s only one reality, but in practice you don’t have one pristine data set that describes everything you want. Instead, you have a bunch of different data sets from a bunch of different methodologies. You have many distorted windows looking at the same world.
We want to build products that extract signal from each of these windows to build an integrated perspective and strategy. The process is to first recognize the noise and inherent biases in each source, then correct for them, then triangulate between all of the data to build that single view. When you align and expand your data in this way, there is a kind of network effect, where more data sets enable more questions to be answered in a powerful way.
Since you can’t just “merge” data to do this, the challenge is to build well-designed data products that let us bring this power to the client in ways that actually get used.
Designing this is really hard, but when you do, it’s worth it because you democratize the use of data and the trust in it, and enable clients to integrate it into their actual strategic workflow. You don’t need PhDs or black boxes or consultants to tell you what to do, when you can converse with the data in natural, human ways. The data may challenge the human with a surprising insight, or the flow may be reversed, where the human has a hypothesis that can be conveniently tested with the data on hand. It’s this conversation between domain experts and the mass of all data at their disposal that we want to help design and facilitate.
Ultimately, it’s always going to be an uncertain world. Bets will have to be made. That’s the life of any strategist. We just want to manage that uncertainty and maximize the odds of success with every piece of data and human intel available.