Finding the Balance Between Storytelling and Data

It's a game of correlation versus causation

a blank-faced teacher in front of a chalkboard
Some marketers focus on the stories while others focus on the machines that help analyze these stories. Getty Images
Headshot of Shelly Palmer

There are two basic camps in the world of marketing: numbers camp and pretty pictures camp. I live in both camps, and you should, too. If you live in the numbers camp, I’m going to confirm what you already know. And if you live in the pretty pictures camp, buckle your seatbelt because this is going to be a bumpy ride.

Marketers are storytellers

Yes, we are. Marketers tell stories. We deal in the meta-realm of emotion and the unspoken languages of music and color. I get all that. And although you will come to doubt my sincerity, I truly believe in world-class storytelling. But stories aren’t everything, and today they are nowhere near enough.

Many marketers I know spend enormous amounts of time and energy trying to craft narratives to explain why a customer might buy their product. The search for causation is, by definition, a grail quest. It is impossibly hard, and there may not actually be a holy grail.

If you walk into an average marketing department at an average mega-corporation, you will see mood boards and design target posters with names and descriptions. “Sally is a 24-year-old postgraduate student. She lives in XYZ college town. She has a pit bull named Brad. She is a nester and wants to own a home someday. She goes to hot yoga three days a week.” There’s the customer journey as imagined by the pretty pictures marketing department. It’s all very old-school and (I agree) has its place, but not for most of today’s marketing.

Here’s why.

Correlation beats causation every time

If you spend some time analyzing sales data, patterns emerge. They are not always obvious. This is why you lean hard on your data science departments for help. If you find patterns, they are often actionable. The actions are testable. The results are optimizable. And value is always created. Always.

A case study

One of my favorite case studies is about a mortgage brokerage firm that wanted to create a mortgage calculator to incentivize people to apply for their mortgages. They built a simple, interactive mortgage calculator that fit inside IAB-standard ad units. They ran a digital ad campaign against their design target (Sally, from above, as a first home buyer. See? I told you I like design targets).

Applications started to come in at a brisk pace. So brisk, in fact, that the human mortgage underwriters could not handle the volume. They needed an algorithmic way to analyze inbounds and score them for their potential to become “good” loans.

They decided to correlate the online interaction data with the banking data to figure it out. Sadly, none of the math done with the 80-ish data points they collected seemed to be any better at predicting the quality of a mortgage customer than the bank’s human underwriters supplemented by the bank’s existing black-box credit score information.

But in the world of correlation, “not everything that can be counted counts, and not everything that counts can be counted.” The 80-ish online data points told part of the story. It was awesome for understanding abandoned loan applications and optimizing the copy and colors used in the ad to attract more interaction with Sally lookalikes, but it just wasn’t enough to solve the “bad-applications-volume-and-velocity” problem.

Machines see things that humans don’t

The data science team at the mortgage brokerage firm created an automated toolset to inbound the online loan applications. They input all of the available data from interactions with every mortgage calculator and built profiles that matched digital IDs with applications about underwritten mortgages.

The humans didn’t see anything particularly out of the ordinary with any of this data. But machines see things that humans don’t. With the use of various algorithms to analyze the complete data set, certain patterns emerged, one of which was extraordinary.

An algorithm could predict with 86% confidence that people who played with one specific slider on the mortgage calculator for more than 3.8 seconds had a 79% chance of being good credit risks and being approved for a loan. The math worked in two directions. If a person played with the calculator for less than 3.8 seconds, there was an 81% confidence level that there was an 84% chance the person would not be approved.

This correlation of time spent with the interactive ad cut down wasted human underwriter time by over 40% and increased the ROAS efficacy by over 11%.

You don’t need to know why people who spend more than 3.8 seconds with one of the sliders on the mortgage calculator have a 79% chance of being approved for a loan, you just need to know that the math is correct.

By the way, when the human marketers tried optimizing the calculator to increase the use of this specific slider, the efficacy of the applications process diminished. There was no causality between specific slider use and the desired outcome. The humans were desperate to tell a story about why and how; the computer just did the math. What worked was getting more people who need mortgages to engage with the calculator. And that is what the ad is optimized to do to this day.

Printing money

If you found a slot machine that paid out 5.1 cents for every nickel you put into it, you would hock your house and leverage yourself to the hilt to find nickels to feed it. This is the difference between searching for causation (a narrative) and finding a mathematical correlation that predicts an action with high confidence.

It’s not sexy; it’s not emotional; it’s not even that much fun to do. But today, all of the tools to find and exploit correlations in your marketing data are available. And while data may be more powerful in the presence of great content, a solid correlation can create immense value all by itself.

@shellypalmer Shelly Palmer is CEO of The Palmer Group, a strategic advisory, technology solutions and business development practice focused at the nexus of media and marketing with a special emphasis on machine learning and data-driven decision-making.
Publish date: October 22, 2019 © 2020 Adweek, LLC. - All Rights Reserved and NOT FOR REPRINT