When discussing predictive analytics, there’s no better place to start than Tinder. Yes…Tinder. Let me explain.
Tinder was created “simply as a way for people to connect,” says Tinder CEO Sean Rad. Tinder is built on quick and simple match making. It finds your location using GPS, and then finds you potential matches close by. Based on the profile picture of that potential match, you swipe right to ‘like’ them, and if not, you swipe ‘left’ to pass. If that person also likes you, then you’re matched. It’s easy, but probably not the best solution if you’re looking for a long-term relationship. On the other hand, there’s eHarmony. This online-dating site matches individuals based on how users respond to “29 Dimensions of Compatibility” shown to be highly predictive of relationship success. Their goal is long-term matchmaking.
So, what’s the difference between Tinder and eHarmony? What is the common theme here that allows us to progress along the thread of relationship?
The more you know about a person, the better of a relationship you have a chance to build.
It’s actually a similar theory with knowing your online customer. When we move beyond traditional segmentation strategies, and use the information provided to us by our online customer, we can successfully predict the behavior of that customer, thus giving us an opportunity to market with relevant, meaningful content and maximize the possibility of conversion.
We can accomplish this with predictive analytics. Prediction is one of the most mature forms of analysis. According to a study on the evolution of business intelligence solutions, predictive analytics provides the most business value, but is also the most complex to implement.
The premise of predictive analytics begins with grouping like customers together, looking at their behavior in aggregate, and anticipating future behavior based on group similarities. Traditionally, this has been done through basic segmentation. The logic of basic segmentation is two-dimensional, like a decision tree. Does this customer meet a certain basic criteria? Yes means that it’s ok to head down this path and into this bucket. However, this segmentation technique, while valuable, is also limited. We’re defining segments, and marketing to those segments based on yes-or-no type of criteria (is this customer female or male).
So those variables that we’re trying to predict customer behavior on are missing that multi-dimensional element. Yes, customer A might be female, but what if she is also making purchases for a male family member or child? Predicting that customer A is going to behave in a certain way based on limited two dimensional information leaves a lot of room for error.
A better way to group like-customers together is through advanced clustering. In this method, we are replacing those two-dimensional variables with three-dimensional ones. As an example, instead of looking solely at customer A’s gender to predict how she is going to purchase, now we’re looking at past purchasing behavior. Yes, we see that this customer is a female, but she’s also made two purchases of male products in the past year and started purchasing infant products two months ago. By layering in this additional dimension, we can market to this customer more accurately to where she is in that moment. We can confidently predict that the infant she was shopping for two months ago has grown, and is going to need toddler-sized clothing soon.
We take this idea and make it a macro-level concept by grouping or clustering customers based on these types of variables. Therefore, we can generate marketing strategies by clusters that cater to the variables that are grouping these individuals together.
The actual process of advanced clustering entails collecting as much customer data as you can: purchase behavior data, engagement data, and data out of marketing tools like an email service provider, and mapping all of that together. The more information that you can analyze, the better. Your data has to be at the customer level, so you need some kind of identifier like a customer ID to tie the data from those different sources together. You then select a set of the most complete data fields as your cluster-defining variables. Make sure that the variables you select have a nice, even distribution across your customers. Good examples would be metrics like total lifetime revenue/days since first purchase, lifetime email engagement, most recent purchase, etc.
Another method that has been used to predict customer behavior is RFM analysis – a scoring technique based on recency, frequency, and monetary values. This process entails sorting previous customers into independent lists within each of these criteria, and then ranking them into fixed groups where each customer is given a score for each list based upon which group they fall. Again, while RFM can be valuable, its method is limited in predicting future behavior. Rather than using statistical analysis to find the natural breakpoints amongst those customers for each variable, customers are forced into fixed groupings. Furthermore, this method fails to indicate how customers have or are going to respond to your marketing efforts. It only tells you about their past purchasing behavior.
As a marketer, ask yourself: are you identifying your top customers? Why are you doing that? Most likely, your answer to these questions has something to do with understanding where to concentrate your marketing dollars. You’re attempting to analyze how your good, better, and best customers are going to respond to those marketing efforts.
Therefore, rather than using RFM, businesses should be leveraging statistical modeling techniques that develop customized scoring analyses built on global attributes. These are attributes that include both purchase and engagement-related data that are unique to your business and can predict the behavior of a person as a whole under various marketing conditions. For example, with variables like age, gender, marital status, place of employment, geographic location, and more (if available) combined with transactional data out of Magento and external data like seasonality, much more robust models can be created that score or rank customers based on their probability to respond to your marketing efforts.
However, doing the analysis is only the first step. How do you use that information? The next thing you have an opportunity to do is to take what you now know about your customer and find new customers that look just like them.
Here’s an example: After clustering or scoring your customers, you can take those groups and identify the results of the variables you used to segment off of, and tailor a media spend based on that criteria. Again, geography, previous types of purchases, and more come to mind. A few ways to use that data in the media spend are custom audiences, look likes in Twitter, buying search terms, etc.
So. You’ve acquired that quality traffic, traffic that has the possibility of acting just like your best customers. Now, you have to convert them.
Most marketers have a (large) portion of their budget dedicated to acquiring new traffic. The question is, how much are you spending on converting that new traffic into business? Interestingly enough, only 39 percent of marketers consider conversion rate optimization a priority. So, even just lightly investing in CRO puts you ahead of most other marketers. Neil Patel has a great piece on the importance of conversion rate optimization. Conversion or Experience Optimization is the practice of testing for the best possible digital experience that will cause a customer to convert. This will often include running tests that segment the online experience based on the characteristics you defined in your initial customer analysis.
There are a handful of affordable optimization products out there that you can leverage to make this testing/segmentation strategy happen. Optimizely is one of these, and it is whom we partner with to offer these services at DEG. This process then becomes cyclical because you can further inform that initial customer analysis by segmenting the behavior patterns from your test. Were there any other factors, beyond industry-type, that were driving those customers to convert? You can achieve deeper insight into performance for different sub-sets of individuals. Add that data to your customer analysis model and wash, rinse, and repeat.
After those individuals convert, use what you know about them to keep them coming back…all the while collecting more information on them to continue to personalize that experience to the customer.
Use your predictive analytics skills to acquire new customers, engage with them in the right way, and retain them by being relevant. In other words, optimize for what they actually want in that moment.
Customer data is crucial. It’s how you use what you know about your customers to get more of them, to get them to come back, and to make sure you know they convert. You can do this with predictive analytics. Step out of traditional marketing analysis and into the future of statistical modeling to take your marketing efforts to the next level.