We’ve been hearing about marketing personalization for a while now and are even told that customers demand and expect a level of personalization that few retailers can deliver.
But what exactly does “marketing personalization” mean?
I asked Joelle Kaufman, Head of Marketing and Partnerships at BloomReach about this. She explained that, for most people, personalization really just means “targeting by segment” and primarily takes the form of sending different emails to people belonging to different segments.
“The dirty little secret,” she added, “is that companies have, on average, only three segments.”
As an example, she referred to a traditional retailer whose three segments were “women,” “home,” and “return user.” To target that last segment, she said, the retailer would do things such as “personalize” the hero image on their website based on a user’s past behavior.
Related to this problem is the simple fact that people just don’t visit most websites repeatedly. Visits to specific websites are usually driven by specific needs and if that need is for something like a bed or an appliance or a car, said visits may be years apart.
Finally, even if a retailer can track your behavior over an extended period of time, there is no guarantee that something you did on the site six months ago will actually be relevant to what you want to do there today. (More on this later.)
In other words, she said, there’s no way to do any kind of precise or meaningful targeting if you are working with such coarse segments.
Are Your Segments “Coarse”? Or “Refined”?
To solve this coarseness problem, companies can try to develop smaller, more refined segments. Unfortunately, in order to get the statistically significant sample sizes for such an undertaking, you need massive amounts of traffic, and, ideally, repeat traffic that, as we said, doesn’t actually happen.
Of course, if people ARE making purchases on your site, AND they login when they are shopping (or you have a usable cookie trail), you can “personalize” the experience by making recommendations based on past purchases or by doing retargeting/remarketing based on items they have viewed.
The problem on the recommendation front is, as Joelle puts it, “Products you previously bought are not particularly predictive of what you want now.”
On the retargeting front, there are different issues. First, Joelle noted, retargeting through email and social display ends up feeling like a “numbing bombardment” that can turn the consumer off.
What’s more, this kind of personalization really has nothing to do with user intent. It can be difficult to infer what people are actually looking for based on items they’ve viewed and, frankly, if multiple people use the same account (say, parents as well as their kids), then the data your are collecting is pretty much random.
Finally, Joelle said, if you are recommending products based on “what others have purchased,” then you will inevitably recommend your best selling items, which means that your efforts at personalization won’t help move more products or, which is even worse, actually cater to the particular needs of your specific, individual customers.
What is a better approach to personalization?
One answer, according to Joelle, is session-based personalization built around search (which, I should mention, is what BloomReach facilitates).
She walked me through an example demonstrating how this works.
We went to an ecommerce site that sells furniture and she typed “folding chairs” into the search box. The site, which used session-based personalization, served up a dynamic page showing a variety of folding chair options.
Joelle then showed me how the results displayed became more refined as she clicked on items and used different search terms.
“Let’s say, “ she said, “that you are looking for folding chairs because what you would really like is a compact dining room set for an apartment. As I start clicking on items returned in the original search, subsequent results are influenced by and organized around the items I’ve selected.”
The Personalization Engine
The personalization engine behind the site does all this thanks to several innovations. First of all, it is relying on a sophisticated machine learning technology that evolves the results provided to the web visitor based on actions they are taking.
This technology also avails itself of natural language processing. This means it not only learns from the specific terms the visitor uses to search, but also looks for commonalities in the product descriptions of the items the visitor clicks on.
Finally, the technology can rely on a ton of behavioral data – 75 million consumers per month and 150 million pages a day – collected by BloomReach.
The result, Joelle said, is that “the entire experience becomes personalized, sending the consumer the message that ‘this brand is organized around me.’”
In an ideal world, if the personalization engine is good enough, the experience for the consumer unfolds seamlessly and naturally.
“Really good personalization,” Joelle said, “doesn’t feel like it’s happening.”
How Do You Get There?
Seamless personalization may be the dream, but not everyone is going to be able to live it. Nevertheless, if you want to go for it, here are some of Joelle’s recommendations.
First, you need to know what your data is really telling you. For example, if you have past purchase data, you need to understand precisely how it can help you in the present (if at all).
You also need to be able to extract attributes of customer intent from search queries. This means relying on a platform that can understand what’s most relevant to a given query and capable of extrapolating from that whatever is most relevant to the visitor.
At the same time, the platform also needs to understand what is profitable to you, the merchant. After all, the point of personalization is not solely to make the customer happy; it is also to get the customer to make a purchase!
What Can Get in the Way?
As magical as all this sounds, it is not the case that every online merchant will be able to take advantage of session-based personalization.
For starters, any platform relying on machine learning needs a lot of data. This means that, if you don’t have a ton of traffic to your site, your options are limited. Similarly, if you don’t have a lot of products (at least 100 to 150), then the platform won’t have enough meaningful choices to offer the customer.
Content also plays a role here. If your content is not distinctive – if your product descriptions are too generic or brief – then, again, the platform won’t have a lot to work with when trying to figure out why the visitor clicked on this instead of that.
On the flip-side, however, you also have to watch out when using language for your brand that is too idiosyncratic. If customers don’t really understand how you talk about your own products, then they will find search frustrating.
The good news is that, according to Joelle, a robust natural language processing system can help visitors, even though they express themselves generically, if it is capable of understanding context and “synonymizing” – i.e., taking their search terms and orienting them to the right idiosyncratic descriptions.
“It’s an interesting way of teaching your visitors your brand language,” she says.
How Personalized is Your Personalization?
Is personalization at your company especially “coarse,” and not really personalized at all?
What would your organization have to change in order to create a truly personalized experience for your customers?
Let us know in the comments below!