Everything you thought you knew about analytical modeling has changed
Editor's Note: The following is a guest post from David Baker, COO and co-founder of Cordial.
The news of the last year or so has been tough for people in the business of building analytical models. A few blown calls on the world stage will have that effect. But those who believe it's possible to predict a specific outcome should pay a visit to the gambling tables of Las Vegas. On the way, they would do well to consider the words of Peter Drucker, who said, "Trying to predict the future is like trying to drive down a country road at night with no lights while looking out the back window."
Drucker was right. It's folly to believe that a model can predict the future — a lesson I first learned in the Dark Ages of the 1990s, when the process took weeks and nobody thought it was cool to be a quant. Of course, in those days we had relatively few data points, and so our picture of a typical consumer bore a striking, if one-dimensional, resemblance to the model builder. But a few years from now, by 2020, according to one report, every single person on the planet will generate 1.7 megabytes of new information per second. To put that in perspective, Audible estimates that an hour of audio represents about 28 megabytes of storage space, which means the average person in 2020 will generate the equivalent of one hour of high-quality audio in about 16.5 seconds.
What does it mean to have that much consumer information? Arguably, the flow of information is about to go from a drip to a mighty ocean, but even that analogy minimizes just how transformative this moment is. We aren't talking about new insights about the same old consumer. We're looking at the arrival of an entirely new consumer, one that marketers must engage with via machine learning. Which can be challenging, because machine learning changes everything we already know — and a whole lot more we have yet to learn — about the practice of analytical modeling.
Democratization of decisioning
Many marketers think of the marketing stack as an unwieldy data warehouse in need of coordination. But the stack is also a way of thinking about the democratization of decisioning. The more information marketers ingest, the more questions they can pose. With more questions, marketers can game out more scenarios. As they game out more scenarios, marketers democratize incremental decisioning. That doesn’t mean everyone gets a decision, it simply means there are levels in the company that own decision rights for different decisions, and more importantly, those levels can be expanded and coordinated as the need for more decisioning arises.
A recent interview with eBay's director of marketing technology offers a glimpse of how machine learning democratizes decisioning at the moment. For eBay, the challenge is to personalize a marketplace with billions of items for sale and 167 million active buyers. The company's solution was to empower eBay's marketers to create deals as they saw fit, but to then place those deals into specific buckets so that a machine learning model could leverage the user's browsing history, along with other data presumably, to predict which deal a customer would find most enticing.
Is eBay's marketing team predicting the future? Absolutely not. What they're doing with machine learning is giving themselves an exponentially greater number of bites at the proverbial apple. Think of it this way: The old adage is to fail fast, presumably because you’ll find success that much sooner. But a better way of stating that old adage going forward is: "Fail smarter." The deals eBay marketers create are still hit and miss — they always will be because prediction is not perfection — but the overall enterprise at eBay is hitting and missing a lot faster than it used to, which means it’s also learning at an exponentially faster rate.
But remember, democratization is a two-way street
Customers also have access to machine learning tools. We already trust machine learning to organize our inboxes, suggest the movies we watch and manage our homes. Soon, machine learning will write our responses, and eventually, the consumer side of many engagements will be automated too, or at the very least, supported by machine learning. So, what happens when a marketer's machine learning tools lock horns with those employed by consumers? The short answer is we don't know. But if you want context for understanding the dynamism of democratization, look at your navigation app.
The mapping functions on our phones are more than mere route-makers. Today, we expect our phones to tell us the fastest route given prevailing traffic conditions. Navigation apps are adaptive by nature, which is necessary as traffic patterns are fluid and environmental conditions evolve (e.g weather impact on routes and traffic).
Of course, there are a finite number of routes because a given city has a finite amount of space. The engagements marketers are concerned with are exponentially more complex. That complexity will grow as democratized decisioning becomes further baked into consumer behavior. Which means you can build an enterprise to fail smarter, but you must remember that whoever is on the other side of that engagement is also failing smarter. The deals from eBay, for example, see enormous benefits in relevance, thanks to machine learning. But on the customer side a similar machine learning tool might filter out impulse buys because that model has learned the following about an individual consumer: the regret of impulse buys.
But when will the model, articulated via machine learning, predict the future?
Never. Analytical modeling isn't about the big, predictive reveal. It's a discipline built for incremental decisioning, risk management and creating scenarios to react to market conditions. Machine learning opens up new capabilities in analytical modeling, but marketers should remember Drucker's warning about predicting the future. Machine learning won’t tell you what’s around the corner on that dark road, but it will give you exponentially more options for reacting to whatever the future holds when you get there. Most importantly, machine learning isn't something to be feared, in fact it only accelerates what marketers are already good at — creative thinking and delivering great experiences to their customers, and that’s really the end goal we should keep our eyes on.