You’ve no doubt heard the buzz about personalization:
Customers expect personalized experiences on your website.
Customers are more likely to leave and less likely to buy without a personalized experience.
Your competitors are already embracing data-driven ways to deliver more personalized experiences.
The gap between the performance of businesses that personalize and those that do not is widening. Companies that haven’t already invested in personalization are at risk of being left behind.
Take a deep breath.
You’re just experiencing a standard case of “FOMO”, which Oxford Dictionary defines as “Anxiety that an exciting or interesting event may currently be happening elsewhere.”
I hope to convince you that, regardless of your starting point, you’re still maybe not as far behind as the hype cycle would have you believe. For instance, in our recent benchmarking survey, only 1 in 5 respondents rated their company as having a ‘world-class’ data-driven culture. Additionally, a number of improvements in the technology behind personalization means that now, not five years ago, may be the best time to get started. Bottom line: embrace a sense of urgency in your decision to adopt personalization, but not a sense of panic.
In the rest of this post, we’ll break down the current lay of the land in 2017, how that’s changed and expanded since Ryan’s original post, give you some ideas on how and where to get started and talk about the fundamental principles that will remain, regardless of how the technology landscape changed.
So, what is personalization, anyway?
In his previous post, Ryan defined it as follows: Personalization = the targeting of an offer, content or experience based on either an action or actions a customer takes (behavioral), or based on something you know about them (demographic).
That still holds. Since that post, there have been a number of improvements in the technologies and the techniques yet the fundamental principles remain.
Visitor & Data Collection
As access to data improves, our ability to tailor experiences for customers on the site does as well. In early days, this was constrained by information that was available to the customer’s web browser (New/returning, time of day, day of week, minimal prior visit clickstream behavior). The holy grail was to get the visitor to log in, at which point our ability to store and retrieve data about them with which to tailor experiences increases dramatically.
Visitor & Device Stitching
Have you ever been presented with a recommendation on Amazon that you really liked, only to switch devices later and not be able to find that item again? Maybe not recently, but this used to be a huge hurdle for retailers. Until you logged in, the data stored about you was local to your browser or device; those sessions were all disconnected.
Visitor stitching solves for this by associating all anonymous activity on a device with the logged in data once the visitor logs in. It captures all of the otherwise-orphaned data and associates it with the user so it’s not lost. From a business perspective, it can mean learning a lot from customer behavior during those anonymous sessions that we would have previously thrown away. Google Analytics’ blog post describes this process in more detail.
Enrichment with Third-Party Data
Alongside this stitching technology, anonymous visitors to your site are no longer completely anonymous. Partner with one of the many customer marketing platforms in the space and you’ll be able to associate customers with their (anonymized and rolled up) off-site browsing activity via ad networks. Doing so will give you visibility into where else customers have been on the web off your site without a sign in. For instance, customers who are interested in sports may have previously been on sites such as NFL and NBA or ESPN. Knowing this small amount of information about them may help you craft a more relevant experience for them on your own site by providing clues as to their intent or segment. Anil’s post provides additional definitions as to the differences between first- second- and third-party data.
Logging in is still the holy grail in being able to deliver personalized experiences, but these techniques have closed the gap somewhat by providing a rich set of information about previously anonymous visitors.
Data Layer/Real-time Access to Data
If you’re familiar with Tag Management, you’ve no doubt heard of the concept of a data layer, or data hub. Closely linked to the other two concepts above, it’s the idea that you can store & retrieve data about your customers in a centralized way while they’re on your site. This might allow you to target visitors who meet certain CRM criteria (opened your recent newsletter), or who have abandoned a cart in a prior visit with a targeted message.
Improvements in the fleshing out of the data layer and the real-time access thereof mean fewer limitations on how we can target and fewer development hacks to get there (we’ve all at some point relied too heavily on cookies to persist behavioral information, am I right?).
The technology behind product recommendations has changed and improved in the last few years as well. Many product recommendation tools used to rely solely on product transaction data to determine both–which products were similar as well as what types of products were typically purchased together (think Amazon’s You Might Also Like recommendations).
Today’s product recommendation tools use not only sales transaction data to come up with relevant product recommendations, but also look at the product metadata itself to make that determination. For instance, using a technique called computer vision, the tool may be able to learn that two handbags within your product catalog look similar and may therefore recommend that a visitor viewing one may also like the other, aesthetically similar, handbag.
Another machine learning technique looks at the product descriptions themselves to build recommendations, i.e. if a particular pair of running shoes was described in your catalog as being most beneficial for someone with “mild pronation” the tool would link this product to other products in the catalog that were described using similar language. Intuitively, this passes the sniff test; if products are described in a similar way in your catalog, they’re probably similar enough to appeal to the same customer on the site. This technique could also save a lot of time otherwise spent creating complex and difficult-to-maintain product hierarchies and attributes .
All of these add additional arrows to the quiver in order to provide more relevant recommendations to customers on your site and don’t require you to build up a long history of transactions in order to start benefiting from recommendations right away.
Regardless of the improvements in technology, we believe that the fundamentals of personalization haven’t really changed.
Tools Only Get You So Far
As David covered in his recent post, people, process, and culture are fundamental keystones of a sustainable personalization program.
Effective Personalization Always Starts with Strategy
It means thinking about your business and your customers. At Clearhead, we like to frame this in terms of Problems. Through PSM and audience discovery, we determine:
- What problems or hurdles do customers within the segments or customers with these intents experience as they encounter your web property?
- How big are those customer segments?
- And, what type of behavior might signal their inclusion in the segment?
From there, we can create hypotheses specific to those customer segments or intents in an effort to ease their path. An example of a segment-specific hypothesis might be:
I believe that presenting customers who have previously abandoned their carts with a prominent message on the first page of their visit will allow them to more efficiently recover that lost cart. If I’m right, we’ll see an increase in order conversion rate for those “cart abandoners.”
Hypothesis Testing is Still the Core
This is perhaps the most fundamental principle of personalization that has not changed. Every time you use personalization to solve a problem for a segment of customers, you should be thinking about this as a hypothesis. In other words, you should be testing that new, differentiated experience targeting against the control in a randomized experiment so that you can measure the effectiveness of that new, personalized experience. We believe that the audience will respond in a positive way to our new experience, but we’re never sure. We’re trying to beat the control with our personalized recommendation or personalized experience; if we cannot then the targeted effort was all for naught.
In sum: don’t panic, but start soon. Leverage new tools and technologies to make personalization easier and lay a strong foundation of processes. Solve problems for segments of customers and remember that everything is an experiment.