If you want to get a lot of retailers nodding their heads, ask them this: “Are you pursuing a personalization strategy?” You’ll get a pretty resounding “yes!”. In fact, personalization already drives a lot of activity on eCommerce sites. It often influences the products or offers featured on the home page, the order of products you see on a category or search results page, and makes product recommendations both on the site and in the digital marketing (email, retargeting) that follow. If you’ve ever been browsing a site and suddenly had the order of products change around on a product listing page, that was probably a decision driven by personalization.
But personalization is only as good as the data about products and the site’s ability to influence a shopper. Shoppers who are logged in and whose behavior can be tied to past purchase history are likely to get the most relevant recommendations. But a lot of guesses can be made based on behavior alone. As long as a personalization engine can piece together similarities between how you shop compared to how other people before you have shopped, and has good access to product data, it can make pretty good guesses as to what other products you might also like to see and how to organize them.
But in some ways, personalization as it exists today is static. Yeah, it can come up with recommendations, but it’s not necessarily categorizing the behavior to identify insights about that behavior beyond predicting the next best action for someone who is already engaged with you. You throw in some Artificial Intelligence or Machine Learning, though, and the dynamic of engaging with shoppers on a site will change. The question is, how much? Here are three big changes on the way. You decide how much they might impact the online shopping experience…
Dynamic filters / fuzzy attributes
Shopping a site that sells product categories where the facets have clearly not been optimized for the product you’re trying to buy is enormously frustrating. Go to any home furnishings site and try to buy a round kitchen table that has an extension in it. You can select “round” and you can select “oval” but you often can’t filter on one that could potentially be both. Amazon is actually one of the worst for providing relevant filters for different product categories. It’s clear right away which categories they have someone paying attention to, with filters that reflect how consumers actually shop for products, versus those products they haven’t figured out yet (hint: athletic vs. casual might be good a place to start in breaking down women’s shorts).
Product filters can make a big difference in finding the right product - if the seller knows what's... [+]
Personalization as it exists today can only do so much here – it fully depends on the data that is already associated with a product. At its least sophisticated, personalization is basically just a difference engine – 72% of people who viewed this product also viewed these other 3 products, so it will show you these other 3 products too.
At its most sophisticated, personalization looks at browsing behavior, makes guesses as to what underlying attributes are most important, and then shows you other products, ordered by their attribute relevancy to what it’s guessing you’re looking for. The more you interact, the more information it can use to make guesses, testing and discarding or revising according to how you respond to what it offers up. This is a basic version of machine learning.
The next level requires a bit more sophistication. We’ve already seen that AI is capable of developing its own internal reference systems to help make connections from things it learns. For eCommerce that means instead of relying on the attributes already in use to describe products, AI could create its own attributes as a way of understanding what makes products similar to each other for a shopper.
An easy example of this is a shopper who first views a zebra print purse and then leopard print shoes. If the attributes for these products are not detailed enough, for example, they might both be tagged as “patterned” products. Personalization might recommend a plaid purse next and miss the mark. Over time, a learning personalization engine might, through trial and error, narrow down around the commonality of “animal print” and know that people who viewed those two products would most likely want to also see the giraffe print belt next instead. If the attributes are too detailed – down to the level of zebra or leopard – the same trial and error process may eventually lead to a good result, but again, that connection of “animal print” is never explicitly identified.
And what happens when next season comes along and you introduce snakeskin and tiger stripe products? There’s no internal framework for personalization to immediately identify anything common about these products with each other or with the existing zebra, leopard, or giraffe. The personalization engine will eventually learn that these go together and make recommendations accordingly. But will anyone else know? Will the online merchandiser be able to identify that there is a hardcore customer segment that buys all things animal print? (Can you imagine that outfit?) Will they be able to find these customers and market directly to them with animal print offers? Not unless your AI is smart enough to create its own attributes and apply them to products – and then surface the insight to a business user.
It may be that the business user has to be the one to look at this dynamic group of products and see that they should be called “animal print”, and take it from there. But detecting these kinds of patterns is exactly what AI is best primed for, and once enough of these AI-driven “fuzzy” attributes have been named, it’s not that far of a step before they get offered up to customers as ways to filter their product search results on the site. Given enough learning, AI could even tell you the order you need to present those filters, according to which ones are ultimately most important to conversion rate.
Dynamic product categories
Dynamic product categories takes the idea of fuzzy attributes up a level. The best example I’ve seen of this in the wild comes from a demo I recently saw of Oracle’s Commerce Cloud. A shopper on a tea seller’s website exhibits enough behavior for an AI in the background to determine that the descriptor “zen” is something that appeals to her. And while shopping for tea, the AI dynamically assembles a whole new product category and offers it up to the shopper as a recommended category: zen teapots. No merchandiser went through and tagged a bunch of products as “zen”. But AI could easily detect a pattern that relates products to each other, deduce a potential name for what these products all have in common, and put together a category now populated with zen qualities. Will it be right all the time? It will certainly be more right more often over time.
[Oracle screenshot]
The important thing for retailers is to have the flexibility organizationally to take advantage of an insight like this. If the zen category works and works with a lot of shoppers, that is going to be crucially important for marketers and merchandisers to know so they can proactively act on that insight.
Dynamic product bundles
You can also take fuzzy attributes down a level, to dynamic product bundles. Amazon does this crudely today, by basically listing 1-2 more products that are “often” bought together along with the one you’re looking at and offering a way for you to add all 3 to your cart at once. There’s almost never any price break in those bundles. It’s just the convenience of the one click to add to them all to the cart that is the benefit.
But with AI, these bundles can get smarter – both in terms of making better guesses as to what to bundle together, and even to make a better guess as to whether to offer a discount on that bundle. Again, it comes down to the level of detail the AI is capable of driving to in order to determine patterns. If all you’ve got are fingers, all you can make is a finger painting.
More important is for the AI to surface highly successful bundles to online merchandisers and marketers so they can capitalize on the patterns the AI has found, whether patterns in how the pricing is received, or if there is a kind of commonality to the bundles that helps identify an otherwise invisible customer segment.
The Bottom Line
For consumers, personalization over all is a good thing. In fact, perfectly deployed, a consumer will notice nothing other than delight at how easy it is to find products they like. At its best, personalization encourages deeper engagement on the site which leads to bigger baskets and higher conversion rates.
For retailers, AI-driven personalization is an opportunity to gain access to relationships between products and customers that were previously hidden within the wash of clickstream data. But making those relationships explicit is key. And having an organizational structure (and associated processes and systems) in place will be key to being able to actually take advantage of these insights, to move them from reactive to proactive.
I’m sure that connecting AI-driven personalization to CRM data and your digital marketing platform can potentially lead to a future where you don’t need to get into the middle of identifying a customer segment and a group of products or offers, and proactively finding more customers like them and making them offers. You could pretty easily automate those links.
But at what point will the merchandiser in that scenario lose touch with the shopper? Or even with the products? By all means, make your eCommerce experience more personalized and dynamically tailored to the shopper. But make sure you always understand the “why” behind the recommendations.
