Natural-Born Data Scientist: Unleash the Beast Within

by Kevin Lindsay

posted on 06-20-2016

We aren’t all data scientists — but, that doesn’t mean we shouldn’t roll up our sleeves and acquire the full potential of our Big Data and data-driven marketing initiatives. My colleague, John Kucera — a data scientist himself — explained in his recent Adobe Summit session that each of us are born with a natural curiosity for science. However, in the words of Carl Sagan, “Every kid starts out as a natural-born scientist, and then we beat it out of them.” That means it’s well within our reach to think, act, and behave like data scientists. As marketers, we just need to commit to reclaiming our scientific edge — to acting, thinking, and testing like scientists: data scientists.

Curious and Curiouser
It all starts with regaining our innate curiosity and skepticism: two keys that separate scientists’ — and data scientists’ — processes from the rest of the pack’s. Passionate curiosity should manifest itself as a relentless quest for the truth in each and every one of your marketing campaigns. Suppose you change your banner for winter coats, causing conversions to go through the roof — amazing! But, that doesn’t mean you should always have winter coats in that hero spot.

It’s the curiosity — what was it about this creative in this placement that drove that success? And, it’s also skepticism — was it simply your brilliant thinking that made this conversion magic happen? You must don your scientist hat, fire up those Bunsen burners, and work to uncover what contributed to this win, so that you can replicate the experience again and again.

Data science sits at an interesting intersection — ultimately, where marketing and science meet — an area in which we’re all razor focused on solving problems. That’s what these two fields have in common, and that’s where marketers can dive in and put their know-how, analytical thinking, and machine-learning sidekicks to great use.

Getting Started
Still feeling stuck about how to proceed? According to John, it starts with harnessing your curiosity, asking questions, exploring data, and developing hypotheses. Depending on your business and industry, some good questions to get you started might be:

The insights you find through exploring — usually done through a platform like Adobe Analytics — lead you to ideas for creating hypotheses. Need some examples?

Again, these are hypotheses based on what you’ve observed through your exploration of the data. There seem to be meaningful trends, but you must better analyze and interpret the data to see if that’s truly the case. To do that, we need to determine whether it’s causal or correlated. Because correlation is not causation. Commit that to memory — maybe even jot it down and pin it to your wall.

John shared an amazing example of this during his Summit session. Examining historical data, it was noted that, as temperatures around the world increased, the number of pirates sailing the seven seas decreased dramatically. Clearly, this dip in the global pirate population is causing global warming. Let me reiterate: correlation is not causation.

Acting Like a Scientist
Correlation is not causality. Healthy skepticism — and, yes, common sense too — is critical to the data-science process. But, marketers must also tap in to other key traits to be successful on the data-science front — even just as simple casual observers and fans.

1. Awareness of Precision
Every measurement has an amount of error associated with it. Your goal? Avoid overcommitting when your potential upside is overshadowed by the error tied to it. So, if committing lots of resources and budget will yield a two-percent improvement that comes with an error margin of five percent — don’t do it. That two-percent jump is just noise. But, if the error measurement is .05 percent, and your upside is two percent, shout it from the rooftops and get going!

With a bit of skepticism, of course, on scenarios that are so overwhelmingly positive.

So, how does one obtain those error measurements? One example is Adobe Target’s sample-size calculator, which lets you input data about visits to your site, conversion rates, and how long you want to run a test. Then, it spits back a rundown of how many items you can test and what level of error you’ll get as a result. Easy as that.

2. Establish Your Control
To excel in this arena, you need the mental discipline of a scientist — or, something close enough. Back to the winter coat example: let’s say you ran that promotion in the late fall when people really needed jackets. You had a prominent banner on the homepage touting an amazing brand of coats, and you saw enormous conversion rates.

But, you are a scientist. And, you are a skeptic.

Without a control, it’s hard to say how meaningful that promotion was. But, with a control — maybe one group who is not shown the coat promotions — you’ll be able to start analyzing what’s really going on. Maybe the promo did make all the difference, and people who weren’t exposed didn’t convert anywhere near as much. Or, perhaps, people just bought coats — it was November, after all.

3. And (again) — SKEPTICISM
Ironically, if you are skeptical enough of the data and how it was collected, you gain a deeper trust of the data as it passes your tests. This allows you to trust counterintuitive results and use them to your advantage. Once I see that this promotion drove these results — thanks to having a control, establishing an error margin, and testing other possibilities — I trust that this campaign moved the needle, and I can start rolling out additional elements confidently. Because when you do — ultimately — trust the data, you can reap benefits in a big way.

Let Technology Be Your Data-Science Surrogate (But Take the Credit Yourself)
Again, you don’t have to be a data scientist — after all, you’re a marketer, aren’t you? But, as we’ve discussed, that doesn’t mean you can’t use data science to be a better marketer. Adobe Marketing Cloud offers built-in data science, including predictive analytics, audience discovery, and of course, Adobe Target’s statistics-driven decision engine that powers website personalization such as recommendations. Together and individually, these solutions take the heavy lifting and much of the time-consuming pieces out of testing your hypotheses, looking at thousands of correlations instantly, based on historical and real-time inputs. By automatically personalizing content to visitors, these systems can improve conversions or revenue with machine learning that discovers correlations between visitor properties and responses.

And, above all, trust your autopilot. Yes, skepticism is important, but it’s also essential to let your machine partner optimize and personalize — without your backseat driving. That might be much easier said than done, depending on how you’ve tackled personalization in the past, but worth it.

Your next step? Focus on reinvigorating your natural curiosities so you’re not only asking why something’s happening, but also unleashing your inner scientist to explore, hypothesize, test, and optimize. Be smart and be scientific in your approach — an approach powered by precision, an emphasis on the control, and of course, healthy skepticism. Automation can take some of the heavy lifting and self-doubt out of the process, provided you can trust your sidekick.

When the pieces come together, it’s powerful — and you don’t even need to be a data scientist to realize some serious end results. Just roll up your sleeves and tap into your inner scientist — he’s definitely kicking around in there, waiting to be unleashed on your next marketing initiative.

Topics: Analytics