Whether you’re at an online marketplace that relies on search for navigation, or a retailer or brand with a sizable catalog, site search and discovery are a big part of the user journey. Search needs to be fast, accurate, and seamless with your site to engage customers and keep bounce rates low.
After someone lands on your site, good search and discovery can be one of the biggest advantages to boosting conversion rate. Up to 30% of e-commerce visitors will use your site search box and they’re 2-3 times more likely to convert.
First, a brief overview on the difference between relevance and ranking. Then, we’re going to walk through solutions for improving site search conversion rates. Lastly, we’re going to talk about how to deploy site search to gain a competitive edge for on-site conversion improvement.
What’s in this article:
- AI search and machine learning
- Personalization (and your omnichannel strategy)
- A/B testing e-commerce search
- Speed (zero latency)
- Mobile search
- E-commerce search metrics and keyword campaigns
- Headless e-commerce search
- Instant indexing
- Understanding user intent
- Visual search
Relevance vs ranking: What’s the difference?
A note on of the terms search relevance and search ranking used throughout this article:
What is search relevance?
Search relevance is a measure of how related an item is to a search query. Relevancy depends on a variety of factors, such as the number of times a keyword appears on a product page, and can be influenced by a variety of signals such as clicks.
What is search ranking?
A search ranking is the order in which results appear.
You can have multiple items that are relevant to a search. For example, if someone searches for “shirt” there could be many tops (t-shirts, blouses, tank tops, etc.) that are relevant, and some may be more relevant than others.
How they are ordered could depend on different factors such as most clicked, best selling, top rated, highest margin, newest, past purchase history, recent product pages viewed, etc.
E-commerce search solutions
Here are a dozen examples of e-commerce search solutions that anyone using a modern search service can use today to improve on-site conversion rates. I’ve started the list with machine learning because it’s the foundation upon which many other solutions are built. However, there’s no particular order for what follows after.
AI search and machine learning
When it comes to search, everyone is familiar with Google’s internet search and it’s incredible quality. Google improves search results based on hundreds of factors across billions of daily queries. Replicating that for site search has been difficult because all e-commerce datasets are much, much smaller. However, that’s changing as search-as-a-service companies integrate new AI-based technologies which work even with smaller datasets.
Powerful AI models such as reinforcement learning make it possible to improve results on a much smaller scale — you no longer need to be Google to get great results. You can now boost search results that lead to higher conversion rates and better margins.
Machine learning for search offers a natural revenue flywheel: more searches lead to more clicks which lead to more sales, and that feeds back into search results, which lead to better clicks and conversions.
When shopping for a search service, be sure to select a site search platform that offers machine learning to automatically and continuously improve results based on conversions — however you define conversions. Some questions you may want to consider:
- Does the search solution improve results automatically over time?
- Does machine learning cost extra, or does it require professional services fees?
- Can you adjust results manually as well?
- What data is the AI using to improve results? (e.g., shopping cart activity, sales, signups, etc.)
Personalization (and your omnichannel strategy)
Whether you’re selling on a single website or omnichannel — social media, AR (augmented reality), web, in person, etc. — personalization is a big key to success.
You’re already collecting data about your customers across channels — how they shop, what they’re looking for, and what they’re buying. That information can be used to personalize shopping experiences from recommendations to search results.
You can build solutions such as:
- Order search results by brand or lifestyle preference
- Deliver recommendations in the search bar as users type
- Combine and “weight” factors (past purchase history, location, products viewed, etc.) to deliver truly custom results
Personalization can help deliver the right products to the right users at the right time.
A/B testing e-commerce search
Did you know you can A/B split test search results to increase conversion rate? If we take an example from above with factors such as past purchase history, brand preference, location, and product pages viewed — which one(s) is (are) best to optimize for? Should new search results put more emphasis on past purchase history or recent pages viewed?
A split test allows you to try two or more configurations to determine which one outperformed the other. For example, in Sajari you can create and preview two or more pipelines with different relevance settings. Each pipeline is saved so you could continually test outperforming pipelines against each other.
Speed (zero latency)
The pandemic has changed online shopping behavior, maybe forever. According to Shopify’s most recent Future of Ecommerce report, consumers demand immediacy, convenience, and speed more than ever. By some estimates, digital transformation by retailers, brands, and online marketplaces can unlock $2.95 trillion within a decade.
Every millisecond delay on a site costs companies money. Worse still, more than 50% of mobile users (which now account for more than half of all e-commerce traffic) will abandon an e-commerce website that takes longer than 3 seconds to load.
Page load and search result speeds must be fast. Three ways you can improve search speed include:
- Test: Don’t buy a product you haven’t tested on real data on a non-production site.
- Data center: Pick a search solution that’s hosted in a data center close to your application.
- Optimize your site: Your site’s structure and set up impacts search performance. Try our free search health report to see how your site scores.
It’s also recommended that you implement search suggestions, autocomplete with typeahead results, or instant search with product previews. The more you can bring forward for customers as they initiate a search the better.
The same goes for search results and collections pages. For example, by building dynamic search facets that change to match the context of the query (or collection), you can provide a faster, more dynamic discovery experience.
You can read more about e-commerce filters and facets here.
Smartphones have limited screen space, so there are different considerations for search design. However, mobile search can still be full-functioning with features such as autocomplete (or predictive search), instant search, product previews, filters, sorting, etc.
More traditional search with a separate results landing page can work, too, but it requires customers to go through one more step to find what they are looking for — and mobile users aren’t patient!
E-commerce search metrics and keyword campaigns
Your site search engine’s metrics can help in at least two ways:
- Improving search results for customers who are already on your site
- Improving your PPC advertising by giving you real keyword suggestions
Most e-commerce search engines offer search metrics to help you identify popular searches, ineffective searches, and highest converting searches. Knowing this can help you uncover product description pages that need improvement or tweak search relevance settings to deliver better results.
It can also help improve your keyword purchases. The data from trends and popular keywords on your site can be leveraged to bid on Google or Bing PPC keywords. Conversely, if no one is searching for certain keywords on your site search, it may make sense to drop certain keyword spending.
Low-maintenance search APIs for headless e-commerce
Companies are adopting headless, API-driven e-commerce platforms because they offer tremendous flexibility for customization. The search solution you choose needs to work with platforms such as Shopify Plus, CommerceTools, or SAP Commerce, and a large ecosystem of products for catalog management, inventory and ERP, email, order management, CRM, etc.
For that they will need an extensive set of REST APIs to reliably index and manage data in real-time from various sources.
But here’s the catch: Interacting with search APIs can be very overwhelming, especially when you start tapping into the more complex and dynamic functionality available. Some of the more well known open source solutions have spawned an ecosystem of query builders to give users a simplified interface to their APIs.
We feel it’s better to have a solution that can fit into any environment without massive teams of specialist search engineers. Your team has enough to do to implement and manage the headless e-commerce solution. Pick a search API that offers enough flexibility without expensive search engineering.
During the pandemic, stores that were accustomed to selling in-person had to quickly reconfigure their business for online sales. One of the toughest areas to transfer is in-store merchandising, or making recommendations as people shop. It’s easy to make recommendations; it’s hard to make good recommendations that actually drive conversions.
E-commerce search merchandising (or searchandising as it’s sometimes called) borrows concepts from brick and mortar merchandising, but there are some important differences.
Where brick and mortar setting retailers can set up attractive displays and discounts to catch a buyer’s attention when they walk in, online retailers need to adjust for online searchers.
In an online search context, retailers can move stock by building criteria to emphasize seasonal items, on-sale products, trending products, higher margin items, or just about any product metadata. Merchandised items could appear as featured products either in a search preview, on a collection page, or as a related item in product recommendations.
Look for a search solution that offers data-driven, AI-powered merchandising capabilities.
Unfortunately, many site search solutions have indexes that can take 30 minutes or more to update. That’s too long. With new inventory, flash sales, and updated pricing, you don’t have hours to wait. Your site search should be indexed instantly with every change.
Every search provider says they have real-time indexing, but what most mean is “soon.” Best practice is to take a search solution for a spin before you buy. Test it with real product data and make changes to see how quickly result pages are updated in the index.
Real time updates mean you know customers will always be looking at the most current products and catalog.
Finding out the product you want to buy is out of stock after you land on the product details page can be disappointing — and worse, searchers may abandon your site to look elsewhere. (Finding out your product is out of stock after you’ve purchased it is even worse! See screenshot above.)
A solution to this is building search and discovery rules that push low-inventory items to the bottom of the search results, labeling results clearly (and with an option for users to be notified when items are back in stock), and/or to hide no-inventory items altogether. Sync product information such as price, description, and images with SKU counts to display relevant products that are in stock.
When you combine inventory rules with other criteria such as promoting high margin items or personalized results, you can be sure to deliver the most relevant search results to visitors.
Understanding user intent
Buyers have been trained to search sites like they search on Google: about 50% of Google queries are 4 words or more. Yet most sites still struggle to deliver relevant results with longer keyword phrases, misspellings, different words (synonyms), or search terms that are out of order (e.g., “red shoes” versus “shoes red”).
There are a variety of search features for all the challenges mentioned above, including:
- Natural language processing: Accenture reported that on marketplaces, “search is the brand” — meaning that consumers aren’t typically looking for brands on large marketplaces; rather they’re looking for solutions. They’re typing in searches for “best,” “cheapest,” or “highest rated” product. These types of queries require natural language processing to parse the meaning to deliver meaningful results. Whether you’re a marketplace looking to improve search, or a retailer looking to use search to outperform marketplaces, NLP will help.
- Synonyms: Most site search solutions offer synonym dictionaries to return results whether a user has typed “jacket” or “coat”
- Vectors: Vectors turn words into mathematical expressions to search for similar-terms. If you’re using vectors, you probably don’t need synonyms.
- Exact search: The site search experience can massively vary because queries range from vague, fuzzy terms to specific product searches. The site search solution you pick needs to handle both types of search functionality.
- Typo tolerance: Search solutions should deliver great results whether a user searches for “rd shoes,” “versachi” or other typos and misspellings.
- Stemming: Word variations such as “treats” and “treating” stem from the word “treat” but search engines may, ahem, treat them differently by mistake, so a robust stemming function is required.
Online sellers need to invest in visual search for the following reason alone: visitors are 50% more likely to be influenced to buy from visual search results. Most marketers are already aware of this and have optimized site images and metadata for SEO to deliver great results in Google and on product listing pages.
Two visual search solutions you could build today include:
- Use an image as a search query: Allow customers to upload an image of what they’re looking for to find similar products. Images can be represented mathematically which enables visitors to search across images just like you would text.
- Extract visual information from images: Online APIs such as Google Vision enable you to extra information from an image (as shown above) to enrich your catalog or even sell items from photoshoots. We have used the Google Vision API to extract colors from images which are then used for faceted search.
E-commerce search pricing and costs
Marketplaces have the burden of providing accurate search for millions of items across thousands of categories, but smaller retailers and brands have their own problems to solve when it comes to building a great user experience — often with fewer resources.
The bottom line: to remain competitive is to find a search and discovery solution that enables you to be agile without added costs, complexity, and overhead.
Up til now, building custom search solutions, recommendation engines, and dynamic personalized results was something only the largest companies like Amazon could afford. There were no true off-the-shelf solutions; products required low-level customization and ongoing optimization and maintenance.
Today, products like Sajari enable any company to build sophisticated search and discovery solutions with minimal engineering resources. Most of the solutions mentioned above such as machine learning, stemming, typo tolerance, metrics, instant (real-time) indexing, and autocomplete simply work out of the box.
The key to managing costs is how you manage advanced site search. We offer simple relevance settings and pipelines so our technical and non-technical customers can easily configure, test, and deploy custom search algorithms. Powerful, custom search can be built in minutes with comparatively little effort.
Modern e-commerce search solutions are available and easier to use than ever before. We hope you’ll try Sajari free for 14-days, or if you prefer, book a demo with our team today. Our self-service e-commerce platform is easy to use and we offer a full-service enterprise option, too.