From Reactive to Holistic: Exploring the Various Levels of Personalization

I often talk about personalization in the context of modeling and segmentation. Why? Because, no matter how much resource a company allocates in data collection, it will be impossible to know everything about everyone. Therefore, predictive modeling effectively fills in the gaps of unknowns, as in, “Not 100% sure, but very likely to be something.” Conversely, when you get to have explicit information regarding what your customers want and prefer, you are obligated to comply with such demands foremost. You wouldn’t need any model score for that, either. Start with 'reactive personalization' For instance, if a customer walks into a store looking for a pair of workout shoes, the greeting salesperson should simply lead her to where such items are displayed. The customer’s profile, segment, personas and product propensities are all secondary to the immediate demand. Model scores are for cross-sell or up-sell “after” the customer finds what she came in for, or when the customer is not fixated on some specific variety of products. Let’s call “fulfilling customer’s explicit demand”—or “reactive personalization”—the first degree in personalization. Some marketers may consider “calling customers by their first names” as the beginning of personalization, but I wouldn’t even count such basic accommodation as a step. Now, even when a company obtains explicit information about a particular customer, it would take a serious group effort to comply with her demand in a non-invasive and smooth fashion. There may be no need for predictive analytics in such cases, but how would you show her something she would like, no matter where the contact point would be? In a multichannel environment, interactions may occur in stores, on websites, through mobile apps or at call centers. How would a company share information among all relevant personnel and systems in disparate places? For example, if a hotel chain collected specific preferences, such as “high floor,” “far away from elevators” and “king-sized bed,” it must offer corresponding services, whether the interaction happens at a hotel lobby or reservation website. For retailers, it could be even more complicated as the variety of products and channels increases. Expanding coverages by modeling The most obvious trouble with relying solely on explicit information is that data coverage could be seriously limited. Customers often refuse or neglect to log in or share their preferences. It is not uncommon that less than 5-10% of customers leave any details. Log-in ratio may be less than 20-30%, unless there are clear benefits for them to do so. That leads to uncomfortable situations where the few people who left any information would be bombarded with so-called “personalized” offers—where the target would more likely to be personally annoyed, as in, “Enough about this one type of offer already?”—while the majority is stuck with boring generic messages. After all the hard work to create environments where collected information can turn into personalized messages, such an outcome would hardly be ideal. This is why predictive modeling should be employed to fill in the information gaps. Remember, for gentle nudging, “highly likely” is quite good enough, and immeasurably better than nothing. Stepwise approach toward holistic personalization Holistic personalization means most people are exposed to personalized offers, most of the time, through every engagement channel—not sporadically, through a few convenient channels for the marketers, only for the few who happened to have left some trails. If a company can display different offers to different people based on collected data, that is a great beginning. That means the technical aspects of personalization have reached a certain level. Then, the next step is to see if such a system can be driven by segment designations or model scores in addition to explicit information. Computer systems do not really see much difference between explicit and inferred information; to them, all are just bits of data. The hard part would be creating matching messages and creatives for relevant segments and personas. That is why the creative team must be part of personalization discussion from the get-go, never at the last minute before the campaign deployment. I suggest starting with a few segments just to see if all key elements of personalization—data, analytics, creative and technology—would work together seamlessly. Obviously, availability of explicit information negates the need for modeling, so putting an analytics team in the middle may create some hiccups for human participants. Once a company successfully passes that stage, I suggest starting to develop a persona menu based on marketing requirements. What type of personas would be ideal targets for strategic products? Then, commission modeling projects following the preset menu. I’ve seen cases where hundreds of personas are created for various marketing efforts, but such variety would be the result of years of group efforts. I suggest starting small with essentials ones, say 10 personas for ideal buyer propensities. If that sounds burdensome, start with three or four. It is more important to kickstart the process somehow. It is all about gradual improvements, steadily converting “unknowns” to “potentials.” If you keep at it—but only if you keep at it—your organization will wear a black belt in holistic personalization someday soon. What rapper KRS-One teaches us about segmentation