Personalization is the heart of digital. As a direct channel, digital rewards personalization in almost every form and the sites and companies that have created distinct competitive advantage using digital have almost always found unique new ways to deliver a highly personalized experience. In my last post, I mapped high-level types of personalization to specific strategies. In this post, I want to join that exercise to my earlier listing of analytics for personalization to create a full mapping for personalization – from strategy to tactics to analysis.
As with most of the blog series that I write, it should be fairly apparent that some elements have emerged and changed as I do the writing. In particular, when I laid out the Personalization Strategies I decided that they were too detailed and too numerous to easily dissect so I went up a level and introduced the concept of Personalization types. The language here has been bothering me, and I’m going to re-cast the Personalization types as Strategies and what I originally laid out as Strategies I’m going to rename as Tactics. These changes making the whole, I think, cleaner and less confusing. Here’s a graphic that summarizes the discussion from the last post with strategies in yellow and tactics in grey:
Now let’s expand the view by matching up the analytics techniques I originally described with the tactics they support.
As you can, NBO techniques span the broadest range of analytics methods. I put scoring first because scoring works particularly well in environments where you have to rank alternatives. But decision-trees, clustering, and rule-based systems all make sense in this context. I could have put rule-based systems everywhere, but it seemed to me that Triggers and Thresholds embody the rule in the placement and so it felt redundant to add it as an analysis type.
Here’s the tree for personalizing a full experience:
Content History methods tend to be fairly complex – since you have to understand large amounts of high-cardinality browsing behavior. So while, again, I could have put rule-based systems here, I think it’s better to focus on the techniques I’ve listed. I will say that rule-based techniques are often used hand-in-hand with both clustering and scoring to implement personalization. For VoC-based personalization, you’re typically going to use rule-based systems, but I thought clustering was also appropriate since VoC data often clusters very well.
Let’s take a look at the full tree for suggesting products/content:
At the content-level, market-basket analysis has long been a core technique. Traditionally used as a way to group carts (dating back to grocery), it’s quite effective as a way to link views as well. Ecommerce companies frequently use product views in addition to actual baskets to link product affiliations. Part of the reason that’s important is that it gets you to analytic views of competing products – something that only looking at checkouts won’t accomplish.
Finally, here’s the tree for offer tuning:
There is, obviously, a whole discipline related to offer tuning that might better be described as price optimization than personalization. I’ve tried to focus on offer-tuning as a personalization option, which necessarily involves some form of segmentation.
Put all this together and you have a nice picture of enterprise-wide personalization strategies and the tactics and analytics that support them:
As I said in my last post, this isn’t mean to be one of those “don’t drink red wine with fish” exercises. Most types of personalization might be supported by almost any analytic method in at least some circumstances; I’m more interested in providing likely ways to start. In the spirit of which, I’m next going to tackle some thoughts on the technology stack for personalization and how to think about putting the right pieces together.
Should have the formal X Change announcement this week – hang in there!