Did you know that over 70% of shoppers never make it to the second page of a product category and almost 90% never make it past the first page of search results? How many do you think even make it to the bottom of that first page?
In today’s age, it is critically important that the products your customers are most likely to purchase appear above the fold on the first page. Customers expect you to show them what they’re looking for, and your margin for error is minimal.
Unfortunately, most of today’s eCommerce search engines rely on dated approaches to determining relevancy, and often fall short of their primary objective as a result. Unless all your traffic goes straight to product pages, it’s almost a certainty that your search engine is leaving money on the table.
Fortunately, getting more out of your search engine doesn’t have to mean going through a costly and time-intensive replacement process. In this article we’ll help you understand how and why your search engine is falling short, discuss methods for “tuning” search to get better results, and discuss the next generation of search technology.
Everything Comes Back to Relevance
While a search engine may have different functions, its central goal is always the same: showing the customer the most relevant products possible.
When customer attention is limited, the selection of products shown and the order in which they’re displayed is the most crucial decision your search engine has to make. These two elements are defined by what search engines consider a product’s “relevance” to the user.
Search engines on many of today’s eCommerce sites generally determine relevance using the following approach:
A user submits a search query.
The search engine parses the terms in the query and looks for matches in product attributes.
Each match is assigned a score that is weighted based on the product attribute where the term showed up.
Each product is assigned an overall relevance score, made up of the summation of all match scores.
The result set is then sorted so that products with the highest overall relevance score are listed first.
What’s Wrong With This Approach?
In a perfect world, the traditional approach to search makes sense. Unfortunately, we don’t live in a perfect world.
This approach assumes that data is normalized (e.g. brand names are included in all product names), has a consistent density across all products (e.g. every description is a similar length), and generally aligns with the terms that users enter in their queries (e.g. both “overalls” and “bib trousers” are included for a site with a heavy presence in U.S. and the U.K.). Unfortunately this is unlikely to be the case, which means that relevance will be inconsistent at best, and will often result in irrelevant products being prioritized by the search engine.
As you can see, it is easy to imagine numerous situations where a user might quickly become frustrated with the search experience, or simply assume that the site doesn’t have the product they’re looking for.
What Can We Do To Improve Relevancy?
When clients come to us for help improving the relevance of their result sets, we typically recommend a two-pronged strategy:
Product Data Enrichment
Search Engine Tuning
Product Data Enrichment
The foundational approach to improving search relevancy involves addressing issues with your product data. Without a basic level of product data, there’s not a search engine on earth that can provide a satisfactory experience for all users. For more information on how rich, reliable product data is critical in ensuring users have a consistent and positive experience with your brand, and advice on improving your data issues, see our earlier blog post on the subject.
Unfortunately, addressing issues with product data is often a daunting task that is not realistic in many circumstances. The good news is that relevancy can be improved, even with poor data.
Search Engine Tuning
The goal of search engine tuning is to effectively improve the accuracy of the relevancy score by changing the way the engine interacts with the data you have. Your ability to tune search will depend heavily on the search platform your site uses, and the features it offers. However, there are a few features that are generally included in most search engines, and which offer low-hanging-fruit for improvement. These include:
Synonyms: Ability to indicate that one term should be considered the same as another (e.g. Kleenex and Tissue).
Keywords: Determine keywords that a customer is searching for, explicitly add them to the indexed product data.
Prefer Phrases: In scenarios where a user enters multiple terms in a query, a match on the entire phrase can be given preference over products where the terms are matched separately.
Allow for Slop: This comes in 2 forms:
Incorrectly spelled terms: e.g. vaccum instead of vacuum.
A search on a phrase may include additional terms: e.g. “chrome wheel” should match “chrome plated wheel”
Tune Attribute Weighting: When determining a product’s overall relevance score, attributes are weighted differently based on their importance. For example, matches on the name of a product should logically be weighed more heavily than matches on the product’s description.
Because of the number of moving parts, tuning is a challenging task, especially if you haven’t performed it before. For this reason, it is important that one takes a methodical approach so as not to create cyclical issues, where fixing one problem creates a new one and so on.
Advanced Solutions for Improving Relevance
Tuning the search engine will likely result in improvements to relevancy. However, the challenge with tuning is two-fold:
Product catalogs, user preferences, and trends are constantly changing. This means that maintenance can be time-consuming and costly.
Tuning still relies on an assumption that all data is created equal. The worse the data, the more likely it is that a “fix” to one challenge will break something else.
At some point you’ll find that tuning begins providing diminishing returns. The good news is that there are a number of modern solutions available that are less reliant on product data, and offer built-in personalization that optimizes relevance for individual users or personas.
Avatria Convert does just this by leveraging shopping data and machine learning to improve product relevance. We’ve designed our tool to include all the bells and whistles—one-to-one personalization, manual merchandising rules, configurable models, and support for custom data. The results are astounding; our average customer sees a 10% lift in revenue and 14% lift in conversion.
Best of all, Convert avoids the biggest drawback of most other modern solutions built to solve this problem: implementation time and complexity. Unlike these tools, Avatria Convert isn’t a search engine itself, and is instead designed to augment your existing search engine, rather than replace it. As you might guess, this ends up being a huge time and cost saver.
Search is critical to the user experience and the gateway to conversion. If you’re not paying attention to the performance of your search engine, it probably means that you have a number of problems that have gone undetected. Tuning search is an option, but will only go as far as your product data will take it. In the end, if you can’t normalize your data and create a consistent taxonomy, tuning will be more trouble than it’s worth. At that point, you’re better off using an AI-driven solution that considers multiple personas, like Avatria Convert, to augment your existing search solution.