Personalization has long been a rallying point for e-commerce companies trying to emulate the success of behemoths like Amazon and Netflix. It’s no wonder why — Companies using advanced personalization report a $20 return for every $1 spent.
It’s now a part of shopper's expectations, too: 91% of consumers say they are more likely to shop with brands that include relevant offers and recommendations.
Product recommendations are one way to offer personalization and were once considered the best form of personalization. But consumers are savvy... you have a good idea already what you're looking for and want more from the sites your browsing! That’s where search comes in. Up to 43% of shoppers will use your site’s search box, so personalizing results can be a much more powerful form of recommendation.
Imagine you do a search and the results are better than expected! Not only can search results include personalization, they can even replace traditional product recommendations.
And, the same e-commerce search engine that powers your site search can also be used for merchandising. Retailers often create product collections to upsell and cross-sell visitors for top products or collections, but AI-powered personalization can help by adding another layer of recommendations.
In this article, let's look at everything you need to know to add search personalization to your site.
But first, some terminology...
I'll use many different terms in this article — sometimes interchangeably — so let me explain what we mean.
Search engines vs recommendation engines
- Recommendation engines use an algorithm to return results based on-site activity — such as page views and purchase history.
- Search engines do something similar but have an additional piece of information to work with: a search query.
Most of what you can accomplish with a recommendation engine can now be handled by an AI-powered search engine which uses the query, past search history, and other metadata such as browser location (country, state, city) or time of day to power results — either as search results or as on-page recommendations.
Recommendations vs personalization
- Recommendations are product suggestions that can be one-to-many or one-to-one.
- Personalized recommendations are suggestions that are always one-to-one.
A one-to-many recommendation is something like: people who buy product X also tend to buy product Y. For example, people that buy baby strollers also tend to buy onesies and baby bottles.
Personalization takes into account specific attributes, past purchase history, past page viewing history, social media activity, product ratings, etc., to build a profile of each user from which to suggest new products or boost certain categories. A goal of personalization is to deliver value to customers as fast as possible with smart 1-to-1 recommendations.
What is personalized search?
Personalized search includes search results that are tailored to each individual based on their profile, which can include past search history, purchase history, brand preference, product ratings, gender, and more.
Site search solutions like Search.io can be used for e-commerce personalization and recommendations:
- Search results can be ordered or optimized with personalized products
- Collections pages can be personalized
- Recommendations can be included throughout the buyer’s journey
Later on in this article, we'll show an example of just how easy it is to set up.
Search personalization data
Personalization starts with data. The more demographic and psychographic data you can collect about your customers and visitors, the more sophisticated your personalization can be. This includes stuff such as:
- Location (geo)
- Past purchase history
- Site search history
- Pages viewed
- Social media likes
- Member status (if you have a membership or rewards program)
It can also include contextual data such as browser and viewing preferences, off-site information such as emails clicked, rewards program credits, ads clicked, and more.
Mid-to-large sized e-commerce companies have begun to invest in building data lakes to store, analyze, and leverage customer data across omnichannel marketing touch points for real-time search. Even if you are not investing in a data lake, many e-commerce platforms store enough data on their own which can be used for personalization.
And there are other systems that have a lot of customer data already — for example, your marketing and email platform. All of these systems can be used for search personalization.
Known vs unknown visitors
Generally speaking, the more data you have about someone, the more you can personalize results. If your customers are logging into their accounts and have a shopping history — pages visited, products purchased, gender, age, etc. — you can personalize results.
However, even if they’re not logged in and you don’t know who they are, search can still be personalized using browser type, IP location, time of day or year, mobile vs desktop, and other attributes.
For example, you might promote different winter clothing to someone in Florida versus someone in Minnesota. Even though you don’t know their gender, age, or purchase history, you can promote products based on their IP location.
Search relevance is key to search results. Search personalization won’t work unless the search results match the user’s intent.
While many customers use single keyword searches, it’s becoming more common for search terms to be much longer. Google has trained everyone to use longer terms — more than 40% of all Google search queries are 4 words or more. Voice search has also changed how we search — the way we type is different from how we speak.
There are a number of techniques that site search platforms use to help parse the meaning of a user search, including:
- Natural language processing (NLP) is the process of analyzing unstructured text to infer structure and meaning.
- Semantic query understanding is the process of actually trying to understand the intent of queries.
- Word embeddings, vectorization, query segmentation, scoping, and other techniques are available to help search engines make sense of a query.
Spell checking can be an issue too. Somewhere around 10-25% of queries can be misspelled, and Baymard reports that, “27% of sites are incapable of handling misspelling of just a single character in the product title.”
All this is to say that it's important to pick site search solution that does a good job of query understanding. Personalization can only work if your search engine can return good results from the outset.
Search AI and A/B testing personalization
Search results can be ordered differently based on other factors such as gender, brand preference, or page view history. Well-optimized and personalized search results can deliver a better experience for visitors as they scan your site.
So which data should optimize for? If you had five attributes stored about each customer — age, gender, search history, product ratings, and purchase history — which one(s) are most important and what’s least important? How can you know what’s most effective?
There’s no way to completely know without testing. In Search.io, our customers can use relevance sliders (or query pipelines) to weight, or prioritize, all the various customer attributes they’ve collected and preview what results would look like.
For example, are you more likely to increase clicks and revenue if you put more emphasis on results for gender versus search history? You can try both, preview results, and even test both to see which wins.
Split testing search results or personalized recommendations is a best practice for boosting conversions through search.
Search A/B testing divides users into A, B, C…, etc., groups and looks for which configuration has the highest conversion rate(s). With Search.io, you can save every configuration so it’s easy to test different settings against one another or roll them back to previous versions if needed.
Search.io also uses machine learning to constantly improve search relevance and results. Machine learning adds another level of optimization by automatically promoting best selling items in search results, or adjusting results based on customers’ near real-time search and purchase trends.
The Gini coefficient
What tests are best to run? While you could create a location-based split test to sell winter jackets between people in St. Paul, Minnesota and people in Minneapolis, Minnesota, as you might imagine, it's highly unlikely there will be a clear winner.
That's because it’s often the biggest differences between customers that are most likely to see results. This idea is embodied in a concept known as the Gini coefficient.
The Gini coefficient is defined as “a single number that demonstrates a degree of inequality in a distribution of income/wealth.” In the context of personalization data, the Gini coefficient states that major differences between visitors — male vs female, member vs non-member — can be used very effectively for personalization.
One example is the difference between customers in the United States northeastern states compared to those in the southwest. If you’re selling winter gear, you’ll want to send heavy winter gear suggestions to customers in the northeast, but light gear to customers in the southwest. It’s a much bigger difference than compared to customers in St. Paul and Minneapolis.
How to configure search personalization with Search.io
It can be incredibly difficult on most search platforms to personalize search results. It might require a team of dedicated engineers to adjust and version config files.
By rebuilding search from the ground up, we’ve massively simplified it. Search.io includes intuitive relevance settings that customers can use to adjust results based on parameters passed to the search query.
Let’s look at a scenario.
Let’s say you sell electronics and peripherals. If you know a customer has purchased an Apple iPhone in the past or if they’re browsing on iOS, you may want to boost Apple-compatible product results for searches on “phone case” or “headphones.”
That brand affinity can be passed to the search engine as a parameter. In the above example that parameter is called “brandAffinity”. Now all that’s left to do is to set up a filter that compares the brand to the “brandAffinity” parameter and boosts matching products. A similar parameter can be set up for shoppers using iOS devices who are browsing your site. If they're using iOS, there's a pretty good likelihood that it's better to display Apple-based products at the top of search results.
The data you generate from personalization will pay off in the long run. Personalization success will increase over time. Using reinforcement learning (a type of machine learning), Search.io will learn which queries and suggestions led to more revenue, and it will use that information to automatically boost higher-converting results.
The data you need to personalize your shopping experience is within your reach. It’s embedded within your e-commerce platform of choice and other systems you’re already familiar with.
More than half (60%) of consumers say they will likely become repeat buyers after a personalized shopping experience with a retailer. It’s not too late to start personalizing results at every interaction with shoppers.