Editor’s note: Industry consultant Shelly Palmer is taking his popular newsletter and turning it into an Adweek article once per week in an ongoing column titled “Think About This.”
At a recent dinner party, an accomplished executive told a story about how he and his wife were certain that their devices were listening to their conversations.
“I was talking to my wife about a pair of designer shoes that she wanted to purchase, and not 10 minutes later while she was doing some online research for work, she saw an ad for that exact pair of shoes,” he said. “She hadn’t searched for the shoes; the ad just appeared. Clearly, our computers or our phones are listening.”
I listened politely for a few minutes, as other guests shared their own versions of “surveillance state” anecdotes. Then, I spoke up.
“Which do you think is more likely? There is secret software that breaks about 20 different local, state,and federal surveillance and privacy laws, that neither I nor any of my clients know about but that are being secretly used by me, my clients and other advertisers to put the right message in front of you at the right time in the right place? Or, thanks to your online behaviors (and the privacy policies, terms and conditions you have agreed to), we have access to enriched data sets and our predictive models and machine learning tools have evolved so quickly that we have an uncanny ability to understand your behaviors well enough to put the right message in front of you at the right time in the right place?”
No one is listening to your conversations with any tool that will be used to put advertising messages or content in front of you.
This begets more questions from the party. “What is an enriched data set?” “What is online behavior?” “What is a predictive model?” “What kind of machine learning are you talking about?” “Is that AI?” And my favorite: “How do you know what I’ve been talking about with my friends?”
Enriched data sets
Data is more powerful in the presence of other data, and much of our data is public: address (phone book), cars (warrantee lists), owning versus renting (public records), work (location data from your phone, LinkedIn or other public websites), how many people you are responsible for (inferred from their purchasing data), what you ordered for dinner (via social media posts), where you had dinner (credit card info, which is legal to obtain if the company has a business relationship with you), how much debt you carry (a credit report), your credit score (credit reporting organizations), etc.
The more data you have, the more accurate your predictions can be. But now let’s add in online behaviors.
When you click on something, you exhibit an online behavior, stopping while scrolling a social media site to look at a meme or message, swiping left or right, tapping an icon on your smartphone, picking up your smartphone (accelerometer), walking or running using a health app (GPS), talking to Siri, Alexa, Google, Cortana or Bixby, or playing a game of any kind on any device. All of these behaviors are captured, logged and used to enrich your profile.
However, your enriched profile is not in one place, and every company that wants to send you a targeted message does everything it can to create a single view of the customer. This includes cobbling together the most robust, enriched data profile possible. The better the profile, the better the predictions.
Most predictive models fall into two general categories: classification and regression.
Classification algorithms identify new data as belonging to a specific class or category. There are binary classifications (two possible outcomes such as male/female) and there are multi-class classifications (data that may belong to multiple classes or categories). This is roughly analogous to a person asking, “What is this?” then thinking about it and declaring, “Oh, it’s a cup. Let me put it in the cupboard.”