Why Recommendations Are Not Predictions (And Why Marketers Need to Care)
This blog series is based on The Definitive Guide to Predictive Marketing, available for download now. Check out Part 1 and Part 2, and stay tuned for the installments to come!
Everyone is familiar with online product recommendations. Amazon was one of the first big web sites to use them; now the phrase, “You might also like….” is pretty much ubiquitous across the web. So, basically, recommendations aren’t anything new.
What is new? Some marketing technology companies today are referring to their recommendations as predictions. This causes a hefty amount of confusion, which is the last thing marketers need amid a sea of buzzwords and ever-changing Google algorithms, to name a few challenges.
The bigger issue is that recommendations and predictions are very different things, with very different roles to play. Before you decide to incorporate predictive intelligence into your marketing mix, it’s crucial to know what you’re getting into – and what the possibilities are.
Let’s start by thinking about how a recommendation is developed.
Generally, a recommendation is drawn from previous behavioral data in a customer’s history. That data is then used to place the individual into one or more segments. Then a recommendation software uses click-stream data and purchase data (or abandonment) to recommend products for that buyer, based on look-a-like modeling. The eventual “we recommend” message is based on what customers with other similar behaviors have gone on to click or purchase.
So, recommendations rely on a basic business logic — “if customer bought X, recommend Y.”
Smart marketers know enough to go beyond this, and they also use personalization, which lets them make recommendations based on the individual customer’s explicit behaviors and implied interests. That’s how these marketers are able to produce meaningful lift in response metrics and in revenue.
Predictions are fundamentally different.
Predictions make an explicit claim for the relevance of a behavior or product using a complex data model. Accurate predictions just can’t be derived from simple logic statements that incorporate only a handful of data points.
At the technical level, it takes an entirely different tech stack to manipulate data and build the models necessary to produce predictive intelligence. At the functional level, predictions go far beyond recommendations in both application and flexibility. By leveraging predictive intelligence, a marketer can figure out which cadence, channel, discount, product price range, content, and messaging is best-suited to each customer.
The real power lies in combining predictions and recommendations. That’s why it’s so important to understand the differences between the two, and to verify that you are getting, and leveraging, both.
To illustrate, here’s an example:
Recommendations drive lift in response rates by serving products based on the user segment. The next step – personalized recommendations – is far more advanced, and helps optimize conversion by serving content that appeals to a specific individual. Recommendations used by themselves will demonstrate some lift over randomized content. But if the recommendation is not combined with a specific prediction, marketers are almost guaranteed to be leaving money on the table.
Take “Steve,” who’s next purchase is predicted to be worth $123. By serving a recommendation for a sweater priced at $80, the marketer loses potential revenue. Similarly for “Andrea”, if she is predicted to spend $55, serving her a $150 blouse may dissuade her completely from making the purchase. (Mind you: This is only one example of how predictions can better drive recommendations to increase revenue. To learn more, download our free guide to predictive marketing.)
Ultimately, by combining recommendations with predictions, a marketer can optimize the content that is served to individual users, impacting both conversion and revenue. This is the only approach that optimizes customer lifetime value.
Predictions have the power to deliver significant lift—which is why it’s important to make sure they’re not merely recommendations in disguise. How can you tell if your technology partner is making recommendations or predictions? Simply ask them:
• How, specifically, are your recommendations and predictions different?
• Can recommendations be used without predictions?
• Which technologies power your recommendations, and which ones power your predictions?
• What lift is generally seen when using your recommendations, and how is that lift improved when predictions are then applied?