RetentionGrid is now AVARI. Read more about how we’ve grown in Our Story: From RetentionGrid to AVARI.
Have you ever wondered how Netflix seems to know exactly what kind of movies you love? The company uses predictive analytics to accurately recommend films you are likely to enjoy based on your previous selections and on films enjoyed by users with similar tastes.
But predictive analytics aren’t just for those who watch movies and TV shows online; RetentionGrid offers this kind of technology in email campaigns for e-commerce businesses. It’s part of an overall technology called a Dynamic Content Engine.
Using predictive analytics, we create tailored product recommendations based on each customer’s individual buying history, as well as the buying history of similar customers.
Let’s break down predictive analytics, learn how RetentionGrid brings that power to your shop, and discover what it means for your bottom line.
Predictive Analytics: Associative Analysis
When it comes to RetentionGrid’s predictive analytics, two kinds of methods are employed. The first is called Associative Analysis.
When you view a product on Amazon, you’re shown a recommendation, e.g. people who bought a camera also bought a camera charger. Imagine that you have two products: the camera (Product A), and a tablet (Product B). Let’s say that, on average, 10% of all orders are for product A and 10% are for product B. The probability that a customer is going to order A and B, if they are completely independent of each other, is (.1 X .1 = .01).
The goal is to identify which items are more likely to be bought together than any two random items. This analysis works great if you look at correlated items like accessories (the camera and the charger, for instance) but it is not enough, because it does not take into account your customers’ past purchases or preferences. To deal with that information, we also employ a method called Collaborative Filtering.
Predictive Analytics: Collaborative Filtering
To predict what your customers are likely to buy based on personal preferences, we start with an analysis called Collaborative Filtering. This is the model Netflix uses. Netflix analyzes your preferences, then finds people who are similar to you in order to predict what movies you are most likely to enjoy.
We use a similarly advanced concept at RetentionGrid. We look at a variety of data to match each customer with similar customers, and to match the products they’ve purchased with similar products. In particular, when comparing customers we look at overlapping purchases, sensitivity to pricing and discounts, frequency of purchase, time between purchases, and specificity, which is how well items and other criteria are matching.
We then calculate overlap-to-specificity ratios to see what products are most popular together. Our algorithm is weighted, which means that the stronger the overlap, the more credit goes to products that co-occur with the product we are interested in.
Our Super Secret Sauce Revealed
The fact that we also take specificity into consideration is part of the RetentionGrid secret sauce; paying attention to how many extra items were purchased is not commonly analyzed.
We compare the overlap and specificity data to all frequent, repeat buyers. For instance, let’s say a new shopper bought two items from a store. We compare that shopper with all repeat shoppers, so we can recommend products the new customer is likely to buy. Since we can tell what a specific customer is likely to buy, we can include these items in your email marketing campaigns to dramatically increase conversions-to-order and skyrocket revenue.
We’ve also created a solution that produces interesting, meaningful results for all of our shop customers, no matter what size. To ensure that our young shops without a lot of data could successfully use our product recommendations for their customers, we created an algorithm that performs the collaborative filtering analysis described above, but that also uses product metadata such as categories and tags, in addition to individual products.
Let’s say you are a shop with thousands of items and your orders are spread thinly, so there are not many customers who purchase overlapping products. Maybe you sell 30,000 different wines and only one customer has bought a bottle of 2011 Cakebread Cellars Chardonnay Reserve. Since no other customer has purchased that specific bottle, we can instead recommend wines also purchased by buyers who have bought wines in the California region or Chardonnay category.
If there are matches for both the products and categories, we do a weighted sum. An optimization algorithm is used to discover whether the categories or products themselves will produce the best outcome.
Machine Learning Feedback Loop
RetentionGrid doesn’t stop with injecting products into your campaigns. The performance of each recommendation is the most important part. RetentionGrid monitors what products are resulting in higher conversion rates for which types of customers, and continuously feeds that information back into the data model. The recommendation algorithm gets smarter and smarter over time, which means the longer you work with RetentionGrid, the better your campaigns will perform. It literally learns and improves by itself. It’s an independent, intelligent, self-optimizing machine.
Predictive Analytics and Your Bottom Line
Are you excited to see how RetentionGrid could help your business grow? Our Customer Success team makes it effortless to get started, and we work seamlessly with any email service provider. You can be up and running in just three weeks with fully custom-created campaigns. We do it all. Design, writing, coding, automation setup, performance monitoring. When you work with RetentionGrid, you get a whole team with it. Let’s chat about how we can collaborate!
