Personalization in eCommerce is Broken. Let’s Fix It.
All retailers tout the value of personalization, yet only a fraction are actually doing it optimally, and even fewer truly understand the limitations many personalization engines can expose after deployment.
In fact, most retailers that use conventional algorithms to power product recommendations on their site fall far short of their claims to provide individualized, on-point suggestions. Arguably, they provide the opposite — generic recommendations that apply to thousands of shoppers.
Instead of saying, “This is perfect for you,” they say, “Here’s what thousands of other customers who are somewhat similar to you bought.”
Yet, most retailers don’t even recognize this issue, nor what it costs them in lost sales opportunities and suboptimal customer experiences.
It’s true that conventional methods of personalization are still an important factor in creating value for both customers and retailers as well as generating ROI. However, technology and consumer expectations are shifting at an accelerating pace, which means retailers must do more to keep up.
So, what can retailers do to shift their personalization efforts into high gear?
The Fallacy of Personalization
To understand the fallacy of personalization as most retailers use it today, we have to start at the beginning.
Retailers have always known that personalization is fundamental to providing the optimal customer experience, generating more sales, and increasing customer retention. Decades ago, before the internet, the most successful shopkeepers were those who kept lists of the products their best customers loved or sought and called them when they got new stock. This type of personalization kept customer satisfaction high and cemented lifelong loyalty.
As technology evolved and eCommerce was born, artificial intelligence promised to enable a much more sophisticated playbook — not only for providing real-time recommendations, but also for elevating engagement and better communicating with customers.
With AI capabilities, retailers could automatically cater to each individual customer at scale by recommending products that best reflect their specific style and intentions, which would drive up sales and increase revenue. At least that’s what was supposed to happen.
Here’s How Conventional Personalization Actually Works
Conventional personalization engines take certain information about each shopper — such as their age, location, and gender — and run it through statistical algorithms that have been developed based on the data of thousands of other customers. It then runs the data produced by a shopper’s actions (clicking on an item, liking it, or adding it to their cart) through these algorithms to surface items that overlap.
For example, if you’re a 33-year-old woman from London who adds a floral skirt to your basket, you’ll be shown more products liked or purchased by other British women in their thirties who also bought that skirt.
Here’s the problem: sequencing algorithms that base recommendations on the prior actions of thousands of other customers are not personal at all. They are predictions based on statistics, not individual traits.
At best, they offer a good guess.
Moreover, this mechanism creates an “echo-chamber” of products. Because it continues to promote the same popular items, it increases the likelihood that new shoppers will also buy them. In effect, this creates a cycle that reinforces the success of certain items in the algorithm over others simply because they get more exposure.
In the same vein, traditional personalization does not generate new signals, meaning it will not recommend never-before-purchased items. This should be a major point of concern for retailers, as it means they are missing out on sales that could have happened if the shopper received more relevant recommendations rather than just popular ones.
What’s the Opposite of Personalized?
The result of this system is not personalization, but segmentation.
How can retailers claim to provide personalized recommendations if they present the same products over and over to countless shoppers? What was meant to be “personal” is really an impersonal algorithm that spits out generic results.
The consequences of this are difficult to measure until retailers begin using advanced AI to drive real personalization and start understanding their true impact, including:
- Missed opportunities to capture more revenue
- Suboptimal and unmemorable customer experiences
- Weaker retention and customer advocacy
- Lack of competitiveness
What’s more, once you deploy personalization engines you will not be able to measure the impact of any new solution. The minimum expectation of an A/B test is to be sterile, isolated, and repeatable — however, since traditional A/B-testing tools can’t actually create a clean split between data sets: Data driven into group A will impact group B as personalization engines amplify data of new purchases. Recommendations across both groups will start looking the same, flattening the test and making the results unreliable.
How to Achieve True Personalization With Visual AI
Conventional AI is too dependent on statistical algorithms to provide true personalization, but new, advanced capabilities are making it possible.
With visual AI, personalization engines can formulate on-point recommendations based on the individual actions of each shopper combined with sequencing algorithms that rely on data from thousands or millions of others.
The Next Generation of Personalization
The true power of visual-based personalization engines lies in their ability to generate completely new moments of want by increasing user awareness in a way that was never possible before.
Here’s how it works:
When a shopper starts interacting with products on your site (through clicking on them, saving them for later, adding them to their basket, or actually making a purchase) they provide granular information about their unique taste and style, as well as their current inspiration.
The visual AI engine actually understands and learns from the details in the images of products shoppers interact with — not just their written descriptions — which produces a highly sophisticated web of recommendations based on similarities.
Visual-based personalization engines add essential capabilities to the sequencing algorithms that are already working. They understand what’s popular to others, but have the ability to surface items that best match the needs of the individual shopper, regardless of how frequently they’re bought. This is true hyper-personalization: recommending never-before-purchased items because their visual attributes are favorable to a particular shopper.
Personalization engines equipped with visual AI can uncover the intangible aspects of taste that traditional AI can’t decipher, such as a shopper’s unique love for high-waisted trousers, wrap dresses, or retro-inspired furniture. It has an element of human understanding that can only be matched by actual human beings.
Truly effective personalization architecture should also be able to support A/B testing by splitting into two or more separate recommendation engines that prevent the flow of learned data between them. This way, retailers can accurately discern the impact of the tool.
It’s Time to Get Back on Track
Retailers want to deliver true personalization. It’s a vital tool for increasing revenue in the short-term, but more importantly, it creates outstanding customer experiences that lead to lifelong brand loyalty.
The problem is that most of the companies that have deployed conventional AI don’t understand that the algorithms they’ve built have produced the opposite results.
With advanced visual AI tools such as visual-based personalization engines, retailers will gain the power of true personalization, and all of its benefits: greater revenue, a competitive customer experience, and brand stickiness.