Search Relevancy for Ecommerce
The defining difference between a great search experience that drives sales and a poor experience that loses sales is relevance. How relevant are the product results to your shopper’s intent? The more relevant your search results, the more engaged your customer is likely to be.
Users reward highly relevant search results with more sales, and so it is very worth investing time and money into increasing the relevance of your search results. But confusing results are an all-too-common problem for online shoppers. 60% of online shoppers experience confusion due to irrelevant search results during a shopping session.
Optimizing search relevancy can be a very involved process, but here are some low-hanging fruit changes you can easily capitalize on today.
15% of the top 50 e-commerce sites fail on product name searches. But getting it right is easier than you think.
First, take some time to understand how your site search performs today across the most common types of search. You may not be able to address all of these without the help of a search relevance expert, but you will at least know where you need apply effort.
Types of E-Commerce Searches
Your e-commerce site search needs to perform well against the following types of queries:
Product names searches
These are exact or partial matches against the name of a product, such as “Harry Potter” for the novels in that series, “The Martian” or “iPad Pro”. Perform some representative searches against your site and see what comes back. If your site doesn’t handle product name searches, don’t despair. 15% of the top 50 e-commerce sites fail on product name searches. But getting it right is easier than you think.
Category or product type searches
Your customers will assume you don’t carry a product if it’s not in the category they expect.
Category searches are for categories of products, such as “bedding” or “blenders”. These are the second most common type of query. Customers use category searches to browse whole categories of products. This is especially important when your catalog navigation is organized differently than the shopper expects. Does “blender” go in homewares or the kitchen category? I don’t know either. More importantly, your customers won’t know and will assume you don’t carry a product if it’s not in the category they expect. Catalog navigation is restrained by predetermined hierarchies, but search supports non-hierarchical categories. Put that blender in both categories, search will handle it gracefully. Go ahead and try a few category searches on your site. The results you get back should be relevant and match up well with your filter and navigation options. Category searches should match across the pre-determined categories you set up in your product catalog.
Does your site have Return Policy, Shipping Information or Customer Support pages? Shoppers often use search to find non-product information and your search should support these types of queries. The results should be first in the results list, not buried down in a list of irrelevant products.
Product number searches
For many e-commerce sites, searching by product number is a common query. This is especially true in B2B e-commerce where copy and paste product numbers or SKUs abound. A corollary to this type of search is to make sure your search box can accommodate the entire length of the pasted product code so that the code is visible in its entirely. Usually, at least 30 characters is enough but many sites go as high as 70 or more.
Shoppers are often looking for product attributes such as brand, color, compatibility or typical use. This is tied for second most common query type with category queries and should be handled well by your site search. Search for some common brands or attributes of your catalog. Are the results what you would like them to be?
Category and feature together
Search is a golden opportunity to learn the intent of your shoppers, connect them to products they want to buy, and drive them to a sale.
The most common search type is to combine a category with a feature. You might be looking for an “inkjet printer” or a “silk scarf”. These types of queries make up a third of e-commerce queries and that’s a good thing, since they are specific enough to return highly relevant results. Remember, search is a golden opportunity to learn the intent of your shoppers, connect them to products they want to buy, and drive them to a sale. Try some category plus feature searches on your site. Knowing your inventory, do the top results correspond to your expectations? Are they the best results they could be?
Once you’ve taken stock of your current search capabilities, there are a few things you can do today to improve the overall search experience.
While I recommend all my clients adopt a search engine for site search, make sure you are at least using database full-text indexing. Naive indexing that matches only exact search phrases can’t rank order the results in a meaningful way. Issuing a SQL statement against a database has serious limitations when it comes to search matching, but with full-text indexing, you will at least have some rudimentary scoring of results by which to rank. This may take a bit more time to implement but if you’re not ready for a search engine you should at least take this step.
70% of the top e-commerce sites don’t handle common product synonyms.
Once you are indexing your catalog using a search engine or full-text indexing in a database, ensure your default search is against all product details, including metadata. The way to accomplish this is to have a special field that includes all text data for the product. While the catalog data will separate out title and description, price and categories, you should combine all text into a single field for searching rather than try to search across multiple fields. If you use a database with full-text indexing, you will find that matching across multiple full-text indexes is challenging and produces poor results. Better to match off a single index. If you use a search engine such as Solr or Elasticsearch, having a single catch-all field is a normal search data modeling practice that allows for more fine-grained control over how results are ranked.
Make sure you support partial matches. If you are using a search engine such as Solr or Elasticsearch, this will usually work out-of-the-box. If you are using full-text indexing with a database like MySQL, you will have to configure a hybrid strategy for partial matches.
Optimizing just the top 100 queries on your site can improve the search experience for as much as 30% of your shoppers.
If your search platform supports it, use synonym expansion lists for common product keywords. Check your logs for the actual searches your customers are doing. You can also find published synonym lists for common product categories across different cultures. You never want to lose a sale because of an inflexible choice of words yet 70% of the top e-commerce sites don’t handle common product synonyms such as “hair dryer” and “blow dryer” but getting this right is easier than you think.
Finally, use a search engine such as Apache Solr or Elasticsearch. Although this last suggestion is not a quick fix, it’s the ultimate way to create a truly engaging search experience. Today’s search engines are amazing pieces of technology that have excellent default behavior but can be optimized to deliver a great balance of customer and business need. Best of all, the top search engines today are also free. Apache Solr and Elasticsearch are rock solid open-source products that have all the features of paid products and more. Check them out, you won’t regret it.
Common Issues & Implementation Tips
“My customers try search but always go back to browsing.”
This is a common complaint when the site search experience just isn’t very good. Shoppers that get poor or confusing search results will fall back on browsing (if they stick around at all). To begin to address this, make sure your search results work for product names and categories – failing these erodes trust in the search capabilities of your site. Implementing the changes in this guide will encourage more search-driven engagement.
“Search works great if you know the exact right keywords but can’t handle minor variations of words or even the order of words.”
This can be trickier to fix if you are stuck on a database using SQL. If that’s the case, make sure you are using full-text indexing and combine all product text into a single default searchable field. If you can, switch to a real search engine such as Solr or Elasticsearch. You’ll see better results immediately.
“Customers see no results even when we have matching products.”
Make sure all your product details are searchable. If you require exactly the right text to find a product, consider switching to full-text indexing or, better, a full search engine such as Solr or Elasticsearch. If you can’t switch just yet, give clear examples of searches you know will work in your search box and on the No Results page. Use synonym expansion lists and keep them updated with new terms from your search logs. When no results come back, consider showing suggested products or popular search terms for your site. This can help teach the shopper the search terms that will match your products.
“Promotions and sale items drown out relevant products.”
Search relevancy is tricky and small changes of emphasis can have big impacts. This problem usually occurs within a poorly tuned search engine. Promotional attributes such as “on sale” are given excessive boosting in the relevance calculation. Turn down the impact of these promotions and let the default relevance algorithm do its work. The most relevant items should be shown first. Don’t promote irrelevant products, it only confuses the shopper and distracts from their intention to buy. If you want to promote items that are not relevant to the search, use other placement such as the right-hand column. This is good practice for global ads and site-wide promotions.