Reminder to Marketers: Don't Be Creepy

When it comes to personalization, most marketers don’t get it right. They either do too little—e.g., indiscriminately blasting the same message and offer to everyone—or they go way too far, making people feel spied upon. Think about it this way: We call people creepy when they get closer to you than you want them to be. It gets worse when they seem to know a lot about you and proceed to act like a stalker. People can detect manufactured friendliness from miles away. So why do marketers seldom take the right approach? Don’t assume—and don’t be fake We may tolerate some false friendliness when it happens in person. When flight attendants say “buh-bye” to every deplaning passenger, we don’t stop and complain about their fake smiles because there is no real harm in that, and ultimately makes for a pleasant interaction. Personalization Is a Long-term Commitment But things are quite different when it is not a face-to-face situation. When every interaction indicates that the instigating party “knows” something about the target of the sales pitch. Here, the key words to remember are “not in person,” “knowing too much” and “being sales-y.” Imagine a used car salesman. Picture them in a cheap suit with a fake smile and sweaty handshake. They act like your long-lost friend, all to sell a questionable preowned vehicle. Unfortunately, that is the personification of most online personalization today. Let’s break it down further. “Not in person” means the interaction happens remotely, face unseen, raising the level of suspicion. Most people are already afraid that online businesses know way too much. (In this, they may be giving too much credit to marketers, as most don’t know what to do with all that accumulated data.) Regardless, people are very suspicious of online activities by companies. “Being sales-y” may be just an annoyance factor to some, but it definitely doesn’t help the situation. Use data to fulfill real needs I’m not dismissing the fact that sometimes marketers must act on immediate data. For example, if a person lets you know that he is looking for a new streaming device—via typing in a keyword into a search box, for example—by any means, usher them to the right place, online or offline. You’re not being creepy; you’re simply reacting to fulfill their needs. When you are using affinity models or personas, do not act like you know everything about the prospects. Instead, go with a soft touch. Let’s call that reactionary personalization, where the scope of activity is limited to the small universe of known immediate data. It gets trickier when marketers start assuming things to expand their reach into the not-so-sure domain. “Oh, you just bought a sound bar for your TV! Let me assume that you need many more of those! Here are other sound bars you may have missed in case you are building a whole collection of them!” We all see this type of rudimentary personalization. Annoying as hell sometimes, but still no real harm done yet. Predictive modeling does not provide facts It gets more dangerous when marketers employ tidbits derived from all kinds of data. When you do that, even with the best statistical resources, you could still be wrong. I’m sure you've heard some version of this infamous incident instigated by a reputable retail chain a while back: They were so good at dissecting accumulated transaction data that they labeled a segment of female customers as pregnant based on changes in their shopping behaviors. In this case, switching skin lotion to an unscented brand was one of the contributing factors. Trouble ensued when a father found out his teenage daughter was pregnant thanks to a direct mail campaign. Now, I’m not blaming the data players here. It actually is a great achievement to detect such minute changes in purchasing behaviors. The trouble was in the usage. When something is based on predictive modeling, don’t brag that you know everything. “Likely to become a mother” is NOT the same as “definitely pregnant.” Case in point? Try to yell “You must be pregnant!” at a total stranger who may appear pregnant to you and see what happens. Even if you’re right, you’re wrong for doing that. Gentle nudges, not hard sells When you are using affinity models or personas, do not act like you know everything about the prospects. Instead, go with a soft touch, as in, “We believe you may need this product.” Better yet, don’t even say that and just place personalized offers and products in strategic places and times. Nobody likes know-it-alls and obvious sales pitches. So maybe, acting as if you're not 100% sure is the best hidden feature of models, as they are built to predict likelihood, not state facts. Marketers who are aware of that would tread the path more carefully. Humanize the data in forms of probability scores whenever you can. And let every user know that models and personas are not hard facts, but mere likelihoods. That alone will considerably soften the blows, and prevent all kinds of mishaps down the line. Making Science Brings on Six Industry Leaders to Bolster Personalization Offering