‹ back to all posts
Breaking Down Promotional Email, Part 3

Breaking Down Promotional Email, Part 3

Going Beyond Personalization to Predictions

Hello readers!  It's time for this week's promotional email breakdown - our third and final installment in this series.  Last week, we discussed the pros and cons of segment-based recommendations.  Before diving into this week's topic, I wanted to clarify a small point.

In last week's blog, I pointed out that since I'm a dude, recommending women's slippers is irrelevant, but several readers asked if I had possibly purchased women's shoes by accident: here's a screenshot of the actual purchase in question next to the hero image of the first email I received.


BOOM!  Unisex slippers.  That should mean that "men's" and "women's" signals are evenly distributed, and that the retailer wouldn't have enough information to assign my customer profile to a gender.  As far as I could tell, there's nothing else in my purchase history that suggests I'm a woman, but even now I'm still receiving "women's slippers" recommendations. To be clear: I don't find this offensive or anything, in fact I don't really care that these recommendations are inaccurate - but that's actually worse: because these emails don't relate to me, my preferences, or something I personally care about, I simply ignore them - which means I will never purchase based on anything shown.

Today, we'll be discussing two emails that represent examples of content powered by Jetlore's AI-driven prediction platform.  Our method of content creation accurately predicts content that customers care about by utilizing series adaptation, artificial intelligence, and a deep understanding of each customer's preference in semantic product attributes.

A Quick Re-Cap on AI-driven Predicted Content:

Jetlore's AI-powered prediction platform utilizes artificial intelligence to track every user impression (not just purchases: clicks, views, email opens, ratings - literally everything), and creates a unique profile for each user based on product attribute preferences.  Profiles incorporate real time updates as customers interact, promoting the attributes that customers signify as relevant, and decaying the attributes they signify are not.  With that rich customer profile, our learning-to-rank technology dynamically ranks every item within an inventory for each customer, enabling the retailer to automatically surface highly relevant predictive content.

This method gives B2C enterprises the ability to predict and surface the best content for each customer at any given time, and automatically adapt experiences as time goes on.

Email Example 1/2:

These emails were sent several days apart from a Jetlore customer whose mission is to enable people to easily, quickly and organically discover new products.  

They sell a wide variety of items that range from the world's smallest camera drone (which I am struggling not to purchase), all the way to e-learning courses.  Recently, I purchased a "tech-skills" bundle from their site to brush up on javascript, but I've also bought small gadgets and novelty items from them in the past.  Take some time to compare the emails below - what do you notice? (You can click on them to see the entire email)


First and foremost, both feature discounted products, but do not feature the same collection of products.  It's important to note that I opened the first email, but did not interact with any of the items displayed - in last week's example I did the same thing, but received an almost identical email a few days later. 

My user profile with this company already shows a preference for tech and gadgets - but because I didn't engage with any of the content this time, I sent signals that the collection of items represented didn't resonate with me. The Artificial Intelligence took that information and incorporated those signals into the following experience.  

You'll also notice a few items placed in both examples, which demonstrates Jetlore's ability to adapt and incorporate every impression.  A customer clicking on something doesn't necessarily mean they're going to purchase, and a customer ignoring an item doesn't mean they aren't interested at all. It just means that they weren't interested in that moment (one more reason why single data-point segmentation is ineffective).

While I didn't engage with the items in the first email, I have looked at both earbuds and the LED light in the past - those weren't eliminated completely, but were shown in a different order lower in the email.  Jetlore's AI-driven technology combined previously known information with my recent response to their content, to automatically test and develop an understanding of what activates my unique purchase intent.

The most important takeaway from these examples:

All of the content from both emails come from the exact same layout.  Jetlore Layouts allows retailers to create dynamic wireframe emails, made up of blank content placeholders, which populate the appropriate content for that user at the time of email open. In these examples, the retailer created a template that featured a large product block at the top of the email, and 20 product blocks below with a static banner at the bottom of the screen.  

The retailer can send this template an infinite number of times to their entire email list - no two customers will see the same collection of goods, and no customer will ever see the same collection of goods twice.  I haven't received another email from this retailer yet - but when I do, I know it's going to be a collection of goods that relates to my demonstrated interests, that's distinctly different than the two emails I previously received (and didn't interact).

In comparison to the other examples I've provided in the past three weeks - Segment based and "one-size-fits-all" content, what's your initial reaction to Jetlore's AI-powered prediction content?  Which of the examples would you actually respond to?  

Join the conversation on Twitter @jetlore, and on LinkedIn - www.linkedin.com/company/jetlore.