When it comes to digital shopping experiences, search personalization is more important today than ever before. If in 2019 about 60% of shoppers expected personalization, that number increased to 80% a year later.
The general consensus is that in 2021 delivering personalized recommendations will become a necessity in digital commerce. A website that doesn’t offer personalized search will no longer last in the marketplace. Today, it’s no longer acceptable to respond to a specific query with a hundred irrelevant results. Users expect relevant results only.
Most eCommerce companies out there are sold on the importance of personalized search. But – as there’s often a but – they believe it’s ‘out-of-their league’.
The myth: You need big data to deliver personalized search results aligned to user intent.
Big players like Google and Amazon leverage all the data they have access to ultimately increase revenues and customer experience. An example of the types of data some of these big companies have at their disposal: google search history, Facebook likes, social media interactions, and even phone conversations.
You’re right in your assumption that some machine learning programs need data to function optimally. But as you’re about to find out, there are many different types and approaches to personalized search – and some, don’t need big data.
Long story short – even without big data – you can deliver relevant, personalized results perfectly aligned to user intent. Let’s tell you how then.
How Does Seach Personalization Work?
It’s important to distinguish between two types of search personalization: history-based and in-session.
1. History-based personalization is when the platform knows the user-either because the user is logged in or because they peruse the site often. In this case, the platform collects information about user intent over time, building its knowledge base, getting to know them and enriching the user profile with every new interaction. Often, users are added to engagement groups on the guidance or recommendation of the personalization algorithm.
While this can be a fantastic way to create a relevant digital experience, the reality for most eCommerce businesses is that this isn’t feasible. Most users, even if they create a profile, don’t return often enough to build a rich user profile. But that doesn’t mean smaller players can’t achieve personalization. AI-powered eCommerce platforms have another way to offer users personalized search results.
2. In-session personalization seeks to offer relevant experiences from the moment a consumer enters your site, without access to big data. How? By using a smart personalization algorithm that takes into account every action the user performs on the site to learn about their individual search experience. The idea is that your shoppers reveal preferences and interests when moving from one product to another, all within a single session, revealing purchase intent. The goal is to quickly figure out the user’s intent and use that information to point them towards the most relevant product recommendations.
Think about it this way: If history-based personalization is comparable to shopping in a neighborhood store where they’ve known you and your family for years, in-session personalization is like going to a welcoming new store on the block and interacting with a very intuitive and attentive seller who makes every effort to learn about your needs and preferences at that point in time.
Both guarantee a rewarding customer experience that makes visitors feel valued and respected, potentially increasing customer engagement with a brand.
How Do You Determine User Intent?
In a nutshell, session-based personalization means gaining insight from every action your user takes on your site to learn more about what they’re looking for. As users make choices, their clicks and queries are used to improve their personalized experience, nudging them to the right outcome faster.
There are several ways to gain insight into the user’s intent without any background information.
Time And Location
Right from the start, the system can start personalizing the user’s session with intent data based on the time of their visit and their location.
For example, if it’s December and the user logs into a sporting goods store from Australia, they might be shown summer sports gear. But if another user visits from Vermont at the same time of year, obviously it’s less likely they’re looking for a surfing board. Instead, ski gear is more relevant.
Time of day is also taken into account to guess the user’s intent: for example, if you log into a health store during the evening, you might be shown relaxing, sleep-inducing teas, herbs, and pillows, while in the morning you’d see energizing drinks and yoga gear.
That doesn’t mean, of course, that the user’s shopping path is set in stone from that moment on. After all, it’s quite possible for people to look for relaxing teas in the morning after a sleepless night!
With every new user behavior and web search, the algorithm enriches the user’s profile with more information, increasing the relevancy of their search by serving up personalized recommendations.
Query Auto-completion And Auto-correct
An intent-driven personalization platform should include a good auto-complete system. User’s queries are often cryptic or full of typos, so it is essential that the system knows how to read between the lines to decode intent.
For example, consider a user typing “sw” into the search box. The search engine immediately considers two scenarios: a) the user is actually looking for something that starts with “sw” (for example, “sweater”), or b) “sw” is a typo and the user is actually looking for something that starts with a similar combination of letters (for example, “sh” for “shoes”).
If it’s a shoe shop that doesn’t offer sweaters, the system should be able to recognize that this is most likely a typo and autocorrect it, leading the user to the right result without them having to correct themselves.
Predictive Category Suggestions
The AI-driven personalization system uses machine learning to make predictive category suggestions whenever appropriate.
If a user browses a wide range of running gear and then types “shoes” in a site search query box, the system already knows that they are likely looking for running shoes.
It might then prompt them with a drop-down suggestion for “running shoes” to confirm that that’s what they’re currently looking for.
To help the user narrow down their search and get to their desired outcome faster, the system can use “discovery tags.”
Discovery tags are a mobile-friendly approach to faceted search and navigation. The very features that make faceted search so helpful to users (that is, being able to see filters and results at the same time) are difficult to achieve on a small screen.
For example, a user might type in a generic query such as “shoes” in a search box.
If it’s their first query and the system has no search history associated with the user, it will try to offer a recommendation by narrowing the results down. Just like a good salesperson wouldn’t just bring out all the shoes in a store but instead ask questions first, the system will also try to narrow down the user’s intent, presenting them with quick filter tags such as “running,” “casual,” “formal,” etc.
On the other hand, let’s take a user who has browsed a range of golfing products in the store and then types “Nike.” In this case-again, like a good salesperson-the system will try to narrow its search down based on what it already knows about this particular user: that they like golf.
In this case, they will be presented with quick filter tags leading them to golf-related Nike products, such as “shoes” and “gloves” to help them quickly narrow down the range of products.
Dynamic Facets and Filters
To provide frictionless and relevant personalization, taking clues from user behavior is not enough. The way the search interacts with the product data is equally important.
To put it differently, to provide an optimal customer experience by connecting people with the relevant products faster, search needs to locate those products in the catalog quickly.
Enhanced product data in combination with advanced NLP capabilities is what allows the system to find products in the same way people think about them.
For example, if the user’s search query is “square table,” they should not see or have to apply a filter for “shape” as the system should understand their intent and return a list of tables in the desired shape. Similarly, the system understands that users shopping for “pants” and “trousers” in different regions are looking for the same thing.
This is achieved through a fully attributed product catalog that provides relevant facets and filters to help bring users to their goal faster.
Using refined similarity models, the system tries to predict user affinities based on their topics of interest and affiliations that become apparent as they browse. For example, from just a few searches the system might discover that a user is a runner who cares about sustainability and is also a hockey mom. This is possible even for one-time users who are not logged in: in this case, instead of adding the data to a persistent user profile, the system adds it to a temporary user ID and uses it within the session or until the ID expires.
All these attributes will inform what the system shows this user in her next search.
Intent Aware Product Ranking
If two people from the same demographic mold (for example, two 30-year-old urban professional women) type in the same search query (for example, “shoes”), the system will provide very different search results for each.
These results are based on the fine-grained information gathered either in their previous shopping history (for history-based personalization) or based on their behavior and interactions in-session.
If one of the women just browsed professional business attire and the other searched for t-shirts and jeans, the search query “shoes” will lead them to very different results, as the system will take into account everything it has learned about their varying intents in their individual sessions when presenting search results.
The Importance Of In-session Personalization
Just a few years ago, virtually all marketing was persona-based. Today, that approach will lead to lower conversions, because customers feel mistreated and misunderstood.
As shopping becomes an increasingly digital experience, more shoppers want online interactions to feel personal.
Intent-driven, session-based personalization not only allows you to create relevant and personalized shopping experiences with minimal data, but also by definition bypasses persona-based marketing by treating every visitor as an individual.
In Episode #1 of our podcast, The Ecommerce Edge, Ciro Greco, Vice President of AI at Coveo, and podcast host, Diane Burley, Content Head at Coveo deliberate between the personalization options available to small retailers in the absence of “big data”. But that’s not at all, in a little over 40 minutes, they debunk some of the most pressing contemporary myths concerning digital transformation!
If you’re interested in learning more about enhancing the relevance of your shoppers’ search queries, check out the technical details in our piece, Clothes in Space: Real-time personalization in less than 100 lines of code.
Ready for the next step, but wondering how to evaluate different AI driven personalization platforms? Here are 10 Must-Have Features for a Relevance Platform in 2021.
Not sure? Here’s plenty of other information on personalized ecommerce experiences, just for you.