MailChimp’s mission has always been to empower small businesses. Our Automation, reporting, and Multivariate Testing features are designed to be both easy to use and powerful, because even the smallest business should have a few power tools in their shed.
One such tool that’s usually reserved for larger e-commerce businesses is automatic product recommendations based on customers’ previous purchases. You probably interact with these every day. Amazon puts them all over, from your homepage to your shipping confirmation, and Netflix knows if you’re more likely to enjoy Master of None or Fuller House. But Amazon and Netflix are giants. Small businesses can’t do product recommendations, right? It requires data, math, and infrastructure, and a whole bunch of other boring things that most people don’t want to think about.
That’s where MailChimp’s data science team comes in. We’ve built Product Recommendations right into MailChimp. Now, small businesses can send emails personalized with the items each of their subscribers is most likely to buy.
It takes the guesswork out of e-commerce in just 3 steps:
- First, you connect your store to MailChimp.
- Then, we analyze your sales data and do the math for you.
- Finally, you drop a Product Recommendations block into your campaign and hit Send.
That’s it. We’ll generate personalized recommendations for products that your customers are likely to buy, saving you all sorts of time and effort and enabling you to do cool things like:
- Include product recommendations at the bottom of your next regular campaign
- Set up an Automation workflow to send product recommendations a couple of weeks after a new customer buys something
- Send product recommendations to customers who bought something a while ago and have been inactive since
Here’s how Product Recommendations work…
Let’s consider an example store that sells MailChimp swag. Meet Tonee! Tonee has great taste, likes our store, and just bought a shirt. Nice.
But that’s all we know. We can segment campaigns and trigger Automation workflows with this data, but we’d still have to figure out what to say. It’d be great if we could use this order to guess what Tonee’s next likely purchase would be. Then we could send an email about that product.
OK, cool. I bet the data will tell us what product to recommend. Sounds easy, right? Let’s visualize all of the customers and orders from our fake store:
Maybe that’s not so easy to understand. There’s a lot going on, and it takes time and effort to make sense of all of this data. Fortunately, we can do some math to identify important trends.
It turns out that people who buy the shirt often buy the socks and, to a lesser extent, the hat. They’re also not into the cards, action figure, or books. Just like that, we’ve turned our purchase data into actionable information. Now, we can recommend the socks and hat to Tonee in our next campaign.
But wait, what if a subscriber hasn’t bought anything? In that case, they’ll receive a list of items that have been selling well lately, and nobody will receive a recommendation for something that they’ve already bought or that was out of stock when the recommendation was generated.
Product Recommendations are based on this general idea, but they’re a bit more complicated than that in practice. They’re even more complicated if you have thousands of products and customers, but fear not—if your store is connected, we’ll take care of the hard part.
…and here’s where it gets technical
We know our product recommendations model works because we’ve been testing it behind the scenes. We tested using the data of millions of customers, products, and orders across our e-commerce customers and made product recommendations based on that data. Then, we compared our product recommendations as well as random product selections to actual purchases using a metric called Discounted Cumulative Gain (DCG). Then, we plotted the median recommendation-to-random DCG ratio for each store. A value above 1 indicates that we learned something about customer preferences and successfully predicted future purchases.
Custom recommendations were able to predict future purchases more than 98% of the time. We also used this data to set criteria that a store must meet before they get recommendations. Here’s what your store needs in order to use Product Recommendations:
- More than 50 different customers in the past year
- More than 10 available products
- More than 500 orders in the past year
But! For newer stores or stores that don’t otherwise meet this criteria, you can still send top sellers instead of custom recommendations.
To get started, you’ll need to upload e-commerce data that has product images, links, and inventory amounts. Our Shopify and BigCommerce integrations will do this for you, but you can also do it yourself with the MailChimp API. Our Magento integration, available for download on June 7, 2016, will also offer these features.
Once you’ve connected your store to MailChimp, the Product Recommendations block will be available in the drag-and-drop editor.
We hope Product Recommendations will help you break into the world of data-driven marketing. Above all else, we think your data should help you make smarter decisions, not overwhelm you. We hope that our new tool will save you time and increase your revenue by effortlessly generating content that simply makes sense. Let us know how it works once you’ve tried it out.