According to a 2017 survey of merchandising professionals and category managers, ⅔ of of them consider personalization as a top priority to implement within the next year. And this is with good reason: brands can collect dozens or even hundreds of behavioral measures online to make personalization a reality. But brands with an eye for the future should focus personalization on customer intent -- what a customer wants to achieve and what they plan to do next. Many talk about behavioral marketing campaigns that are responsive to customer interactions as they are happening in real-time, but just as important are the behavioral marketing campaigns that are prescriptive to individuals based on their historical interactions. We call this - predictive marketing.
I am often asked by marketers to explain predictive marketing in their terms. To boil this down, predictive marketing leverages the behavioral data you already store and combine it with predictive modeling to power customer messaging based on predictions of their future shopper behavior, their propensity to convert, likelihood to churn, or potential future spend. Predictive marketing opens up a whole world of possibilities for marketers as it takes the guessing game out of customer intent and allows you to truly provide the right personalized message to the right person at the right time.
Predictive Marketing Is Based on Behavioral Patterns & Trends
The follow-up question I then get asked is - how does predictive marketing work? The mathematical models involved have been called many different names over the past few decades. You’ve likely heard them all and crossed them off your list as simply the buzzword of the day - pattern recognition, statistical learning, machine learning, data mining, predictive analytics, data science, and now artificial intelligence. While these terms are not fully interchangeable, they share in common a set of algorithms that pick up behavioral patterns that can be used to build models that form the basis of predictive marketing. I’ll use the term machine learning here to describe the mathematical models that power predictive marketing, but one can really swap any of those other terms in its place without losing the thrust of the argument.
In a recent consumer survey of millennials and their shopping habits, we asked them to classify their shopping habits and how they like to be communicated to by brands. From bargain shoppers to researchers to seasonal shoppers, their shopping habits varied wildly. Some preferred receiving messages about all sales a brand has and some preferred receiving sale messages regarding only items or categories they’re specifically interested in.
But here’s the thing. You don’t need to survey your customers to find out what type of shopper or customer they would classify themselves as. Machine learning takes out the guesswork and picks up on the behavioral patterns of each individual customer to identify if an individual typically takes their time to purchase a product, or if another individual makes buying decisions impulsively.
Marketers may choose to hold off on discounts for customers that are likelihood to purchase to preserve margin, at least until the time-window of the prediction passes by. Or if a customer has never before transacted, a high propensity score along with an offer may help to close that first transaction, which then increases the likelihood the customer will transact again. Likewise, churn models also provide tremendously useful information. Obviously, a customer who is likely to churn will need additional nurture, especially if that customer was a high-valued, highly engaged customer in the past. But rather than viewing churn as a point-in-time measure, monitoring churn risk continually, and noting when churn likelihoods are increasing signals the need for a pre-emptive action. In both of these models, the models aren’t a “decision” in of themselves, but rather form a more accurate picture of shopper intent so that our messaging strategy is timely.
Machine Learning Is Necessary for Scalable Predictive Marketing
So is machine learning all that necessary? Aren’t there other ways we can identify these behavioral patterns? The quick answer is “no”, there are no other ways to find the best ways to predict behavior because the number of behaviors and, more importantly, the combinations of behaviors is so vast, no one can assess all of them manually. For example, if we measure 10 behaviors and we want to find which combination of these 10 are most important in predicting an outcome, we have to assess the 45 2-way combinations or the 120 3-way combinations. We might do this. But if we are honest with ourselves, we are really collecting 100+ behavioral measures which brings the number of 2-way combinations to 4,950 and the number of 3-way combinations to 161,700! It’s not scalable - nor a good use of time - for an individual to analyze all of those combinations. And that’s exactly why machine learning exists: to leverage predictive mathematical models to automatically and quickly assess the vast number of possible solutions efficiently so that we can find the best way to predict an outcome.
Amazon and Walmart Lead the Way - and It’s Working
Companies like Amazon and Walmart have made headlines for their predictive marketing methods. Amazon has teams of data scientists that build custom mathematical models to predict customer intent that power messaging. And it works, 55% of their sales are attributed to machine learning recommendations. Walmart’s teams of data scientists continually optimize their supply chain, identifying ways to better forecast product demand and improve efficiency of delivery.
- Predictive Marketing Results:
- On average, our clients open rates of 40%, 13% click through rates, and 9.3% conversion with our predictive marketing campaigns.
- A well known hardware store saw 33.7% conversion on the predictive marketing campaigns targeting high value customers that were disengaging from their brand.
- An upscale clothing brand saw a 16% increase in email click through rates and 8% increase in conversion rates running predictive marketing campaigns targeting highly engaged non-purchasers.
But most organizations don’t have the resources or data scientist teams at their disposal like an Amazon or Walmart do. And they shouldn’t have to honestly. It goes back to the age old question of build vs. buy. For companies that desire to scale and remain flexible, investing in a platform that is obsessed with providing their customers the best in predictive marketing techniques allows them to do just that.