By Stephen H. Yu 07/24/2017
One of the weirdest experiences in modern-day shopping is product recommendations soon after the purchase. I am talking about those relentless messages that follow you around weeks at a time. If they are relevant, it can be useful for the buyers. For example, offers like “Do you want a bottle of yoga mat cleaner along with the yoga mat that you just purchased?” can be beneficial. But it gets really annoying if they keep offering the same items repeatedly, where a buyer would wonder “How many more yoga mats do I have to buy before you leave me alone?”
The sad part is that such ridiculous marketing is being conducted under the banner of “Personalization.” That is more like “personally annoying people,” and definitely not personalization.
At the risk of stating the obvious, let me point out that personalization is about the person. It should never be about product, brand, channel, or marketer. Further, marketers should never be abusive simply because they obtained some tidbits about their customers.
Clearly, “personalization” is the buzzword of the day. We can see that trend at conferences, marketing meetings, industry papers and blogs like this. And unlike other buzzwords that came and went in the marketing industry, I am predicting that this “personalization” is here to stay for a foreseeable time.
Why? Because consumers demand it, they feel that they are entitled to it, and marketers finally have the technology and data at their command to do it. But alas, only if they do it right. Unfortunately, I see a lot of marketers – even so-called leading online marketers – just annoying their customers every chance they get. As a result, it is very difficult to find good success stories about personalization.
So, I ask marketers this question: Are you really committed to doing the right thing, or are you just saying that word simply because it is a new thing to do now? There has to be organizational commitment because doing it properly requires a lot more than just purchasing a personalized engine off the shelf.
Leaving a commercial recommendation engine in the default mode with raw data is a lot like thinking that coffee magically comes out of an espresso machine. I believe that such a myopic view is the main cause for all of those rudimentary and ineffective personalization efforts. For you to enjoy that cup of coffee ever so conveniently, someone had to cultivate coffee plants, harvest the beans, process them, transport them, do all the paperwork to go through customs, domestically distribute them, roast them to various degrees and package them for espresso machines. Likewise, for personalization engines to function properly, incoming data must go through some serious refinement processes.
Without a doubt, proper personalization starts with a personalized data view, which is neglected all too often. Some may use terms like “360-degree customer view,” “single customer view” or my favorite, “customer-centric portrait.” No matter. All the transactional, behavioral, demographic and environmental data must be realigned around “each” customer or prospect. Some may say that they already have customer ID system that connects all those data points (many don’t). Great, but that is just a good beginning. We still need to convert such “event”-level data into “descriptors” of individuals.
Transaction-level data may tell you what happened on a certain date, for how much money and for what product. Descriptors of individuals display buyers’ personal spending patterns by product categories, channels, discount buckets, time periods, etc. That is quite different from stacks of a transaction or event-level data sitting in some Big Data platforms designed for mass storage and rapid retrieval.
When we line up information around people, we often find out that we really do not know much about our customers. All those fancy variables created around the target individuals have many holes in them, for various reasons.
Maybe they are new customers, or they just browsed a few items but never bought anything yet. Some customers may have shopped only in certain categories, but stayed away from others. Some customers may have been very diligent in deleting their online trails. To do the personalization properly and consistently, we need to fill in such gaps.
Most of the personalization engines, unfortunately, are currently set to act only on explicit “known” data. When marketers go too far only with what’s known to them, the customers who casually let some parts of their lives known to marketers get bombarded with the same messages until they get completely sick of them.
That is a sad situation as, categorically speaking, people with known behaviors often account for less than – at times far less than – 5 percent of the approachable universe. So, in that scenario, 5 percent get to be stalked relentlessly, while 95 percent are completely ignored. Not ideal at all.
This is where statistical modeling enters the personalization arena. The importance of modeling cannot be overemphasized even in the data-rich environment, simply because we will never know everything about everyone. Statistical modeling systematically converts “unknowns” to “potentials.”
No, we may not know for sure that a particular target is indeed a “gardening enthusiast.” But yes, we can say that she is “very likely to be” a gardening enthusiast, with statistical techniques effectively mining available data — such as what other products she purchased and browsed with varying frequencies and intervals. The results of the models are “scores” by which you can measure the degree of confidence, as in a nine out of a 10 scale. This is much simpler than having to worry about hundreds of variables with more holes than Swiss cheese.
Building a customer-centric view and filling in the gaps with modeling techniques is far more superior to a default setting of a personalization engine that would just ingest unrefined SKU-level data and spit out reactionary product offers. For one, resultant marketing messages become more relevant and less annoying.
Secondly, having “potential” values for certain behavior enables marketers to act on most of the targets, not just fractions of them with explicit data. Moreover, marketers are able to rotate messages with multiple personas assigned to each individual target. Why show only one thing again and again? The target may have scored high in other categories.
The result of modeling work will make the personalization engines run better, too. After all, those software solutions are designed to ingest any type of variables.And the model scores – which are essentially summaries of hundreds of data points – look just like another set of variables to machines. Consider such model scores as really tasty coffee beans that you can put into your shiny espresso machine.
Country store owners in the old days were known to have personal touches because they treated their customers as people. They would not have offered more hammers to you just because you just purchased a hammer. They would have put it in context, and then they would have suggested products that you may benefit from. (As in “Hey, don’t you need protective gloves, too? I know you’re a klutz!”)
Now we have access to enough data, technology, and mathematical skills to do such personalized marketing to millions of people at a time.But it will work only if marketers do commit to the proper steps, preparing the data specifically for personalization efforts and programming personal touches into algorithms.
Technology made things easy for us, but it is equally easy to abuse it. Let’s not forget that we are just personally abusing other human beings when we abuse technology and data. And we’d better not call that personalization.
About the Author:
Stephen H. Yu is the Practice Head, Advanced Analytics & Insights for eClerx. He is a world-class database marketer with a proven track record in comprehensive strategic planning and complete tactical execution, from data modeling to targeting and personalization based on advanced analytics.