Improving customer experience is top-of-mind for every digital business. Billions are spent each year on mobile apps, content, personalization and omnichannel capabilities. And hundreds of hours are spent on redesigning websites and conversion optimization.
Yet CX plans often overlook a fundamental piece of the equation: site search performance.
Visitors who search are explicitly telling you what they want and are more likely to convert than visitors who browse (50% higher according to an Econsultancy survey).
Today’s best-of-breed search tools have come a long way from simple keyword matching, boasting varying degrees of autocorrection, semantic matching, machine learning and personalization.
But too often, merchandisers “set and forget” search, relying on their solutions to just work their voodoo. Rarely do merchandisers take advantage of the tuning capabilities they’ve paid for.
The result is site search behaving badly, or at least underachieving its potential. Leaving search on default settings may be efficient, but neglecting to check under the hood to ensure search shows the right products for your “money keywords” costs dollars and sense.
The scale of individual “pinhole leaks” in your system can amount to significant lost revenue every year. The good news is you can identify and correct these leaks with a simple auditing process and site search tweaks.
Auditing site search
Step 1: consult your analytics
Pull out your search analytics, and set your date range to one year. Look for high volume searches with underperforming revenue and conversion rates, or high abandonment and refinement rates, and create a short-list of 10-30 “money keywords” to optimize.
Bonus tip: During this exercise, scan your report’s top 100-500 keywords and jot down common abbreviations, misspellings and product attributes that appear. This intel can help you improve your search application’s thesaurus, and may identify helpful category and search filters.
Step 2: test your searches
Now the fun part — roll up your sleeves and play customer! Check for anything irrelevant or out of place. Wear your “business hat” as you do this, and look for opportunities to tune results to better match your merchandising strategy.
For every search you audit, note what needs to improve about the experience. For example, investigate why sunglasses are appearing in searches for “grey jackets,” or why iPhone accessories outrank iPhone handsets.
Note what issues you need to correct for every search term you audit
Optimizing relevance with search logic
Many modern enterprise search applications and digital experience platforms provide merchandiser-friendly admin tools to adjust search logic, the business rules that inform the algorithm. If you don’t have access to such business tooling, enlist a developer’s help to tune the back end (most search applications are built on Solr or Elasticsearch).
There are several levers you can pull to maximize search relevance for your “money” keywords:
Just like Google’s ranking factors, your site search algorithm calculates relevance based on index factors such as product title, category, product description, product specs (attributes), keyword tags and other metadata.
Adjusting index factors across the board, or for specific products or categories, can tune results in favor of your merchandising strategies, and improve relevance, click-through and sell-through.
For example, if you sell high ticket electronics and find accessories and lower ticket items are sneaking their way into top search positions, your engine may be weighting product name at 200% (which would boost accessories’ score), descriptions at 150%, specs at 100% and category relevance at 75%. You can improve results by reducing product name, description and spec weighting and boosting category relevance and price.
Advanced engines may include additional index factors such as popularity (clicks, favorites and sales), product ratings, price, margin, date added, inventory count, semantic relevance and custom attributes (e.g. brand, genre, format or category).
Some product types benefit from a specific keyword or attribute boost or bury. For example, a search for “patio furniture” should boost sets above individual items like patio chairs, and bury accessories such as cushions and covers.
Boosting patio sets within results for “patio furniture” better matches customer intent than individual pieces, and can improve basket size and revenue
Bonus tip: Use your site search’s autosuggestions (or a high-volume competitor’s) to identify terms to boost or bury, per keyword.
Modern search applications do a decent job of recognizing synonyms out of the box thanks to their robust dictionaries and thesauri. However, most ecommerce catalogs benefit from custom synonym mapping to handle colloquial terms and jargon, brand and product names that aren’t standard dictionary terms, and their respective common misspellings. After all, one man’s “thumb drive” is another’s “memory stick,” and one woman’s “pumps” are another’s “heels.”
A usability study by Baymard Institute found 70% of ecommerce sites failed to map synonyms and only return results that match search terms as entered. Considering brands and manufacturers often describe the same things in different ways, this hurts recall and customer experience. It can also stifle sales for products that don’t match the most frequent variants of popular product searches — the two-piece swimsuits in a world of bikinis.
And don’t forget model numbers! Baymard’s research found only 16% of ecommerce sites do.
Most search tools employ fuzzy logic to handle plurals, misspellings and other near-matches. This increases recall (number of results returned) for a given search, and often improves results, especially for misspellings.
For example, a search for “pyjamas” would return matches for “pajamas.” Using stemming, a search including “floss” could match “flossing,” “flosses,” “flosser” and “flossers.”
However, fuzzy logic doesn’t always improve results, particularly when fuzzy matching or stemming a product or brand keyword matches attributes of other products, or other product types altogether.
For search engines that use “or” operators in their algorithms, results can appear when only one word in the query matches product information. For example, any search that includes “orange” (attribute) would return results for “Orange Boss” (brand).
“Or” operators match products to any keyword in a multi-keyword query
Understanding context and adding exclusion rules for specific searches tightens recall and maximizes the precision of your results.
plant, planted and planter
salt, salts and salted
blue and blues
boot and booties
belt and belted
print, printer, and printing
cook, Cook, cooker, cooking
rock, rocker, rocking
Many of today’s enhanced search platforms offer semantic matching, natural language processing and learning algorithms out-of-the-box. Some are intelligent enough to detect when a keyword is intended as a product, attribute or utility of the product (such as “for older dogs”). Nevertheless, even Cadillac tools can miss some important contextual variables specific to your catalog, customer and merchandising strategy. When auditing your top search terms, look for fuzzy product matches that should be excluded or buried.
Bonus Tip: Excluding stemming variants“-ing” and “-er” and “-ed” in general across all searches can tighten search results, optimizing for relevance and sell-through.
Showing fewer matches reduces the “paradox of choice” effect which can lead to slower decision making or even indecision. A tighter set also supports mobile shoppers who have a harder time browsing and comparing products within a list on a smaller screen, and who struggle with applying filters and facets.
Searchandising with slot rules
Slot rules tell your site search engine specifically how you want to populate your product grid for a specific search. For example, you may always want the first row to show your house brand for searches that don’t include a specific brand. Or, to show only full price products in the first three positions, and flexibly rank the rest. (Not all site search tools support slot rules, but many enterprise solutions do).
Keywords that span multiple categories such as “jackets” (men’s, women’s, boys’ and girls’) and thematic searches (e.g. “Valentine’s gifts,” “white marble,” “LA Raiders” or “safety equipment”) benefit from slot rules that diversify results rather than front-load from a popular category. This helps your customer understand you carry a breadth of products and may help them refine their results, especially on mobile where fewer results are visible per screen.
Slot rules can diversify results to ensure results from certain categories aren’t overrepresented in top positions
Bonus Tip: The most efficient way to leverage slot rules is to apply them to your category lists and apply search redirects for exact-match queries. If you uncover high-volume searches that don’t have associated categories, create them! This helps customers who browse rather than search, supports guided selling and can boost SEO.
Search redirects to category landing pages can optimize the buying experience for exact-matched terms
Search engines and DXPs (digital experience platforms) with machine learning capabilities are gaining popularity, promising to optimize relevance and performance with minimal effort from the business.
- Semantic relevance returns product matches even when queried keywords don’t appear in descriptions or metadata.
- Natural language processing identifies search intent and context such as a navigational query (looking for a category) or searching by attribute or product function (e.g. “dry food for older dogs”).
- Aggregated behavioral data can match a visitor to past activity and look-alike customer segments, using predictive analytics to provide personalized recommendations.
Despite their intelligence, advanced tools suffer as much from set-and-forget implementation as their less sophisticated counterparts. Shipping with the most powerful searchandising controls, these platforms are designed for merchandising logic. But many users of these engines fail to leverage their capabilities, and never experience the full value of their technology.
Why you still need to “searchandize” your personalized search engine
Machine learning takes time to get good. Highly trafficked sites with relatively evergreen catalogs benefit most, while less trafficked sites with large catalogs (thus a long search tail) or higher catalog turnover may struggle to build reliable affinities between search queries and products.
Default settings create bias. It’s well demonstrated that top search slots receive higher click-through, on average. When algorithms favor popularity metrics, the “rich get richer” over time. Search satisfaction can dwindle as SKU variants such as sizes and colors sell out, and fresh, full-margin product may be buried under discounted stock.
Tools are agnostic to your merchandising strategies. With data and time, intelligent search tools can recognize buying trends, seasonality and more. But they still lack insight into when it trend forecasts, promotional calendars, anticipated shifts in demand and other variables. By the time they catch up, this context may be stale!
To ensure personalized search serves your business in real-time, leverage index weighting, boost-and-bury and slot rules the same way you’d tune non-personalized search.
DXPs that integrate with CRM and ERP systems allow you to shape merchandising logic for individual catalogs, geographics and customer segments. For example:
- Boost new items and prestige brands for high-spending segments, or boost heavy puffer jackets to New Yorkers and bury them for Californians
- Strongly boost SKUs and brands previously purchased to individual B2B accounts (even if ordered offline)
- Strongly bury products that aren’t available for international shipping to non-domestic visitors
Target should bury “not available for intl shipping” products for non-US shoppers
Search tuning shouldn’t happen in a vacuum. Document your strategies every time there’s an update to merchandising logic. An audit trail ensures other team members (and future members) know what was tuned and why, and can revisit strategies as data is collected and business strategies and objectives evolve.
Consider time-limited strategies. Certain searches will benefit from tuning around seasonality, promotional events and other variables. Site-wide adjustments may also be relevant. For example, boosting sale items December 26 through January 31 helps clear excess inventory and matches buyer expectations for traditional retail. Some tools allow you to set start and rollback dates for merchandising rules. If yours doesn’t, ensure someone’s assigned to revert changes at a designated time.
Should you A/B test your tuning strategies? Your enterprise search tool or DXP may natively support A/B testing. However, because split testing requires sufficient traffic to produce reliable results for each keyword, and sends half of your traffic to untuned results, it’s often unnecessary — especially when you’re closing an obvious experience or relevance gap.
Site search doesn’t have to remain a black box. Make search tuning a regular part of your searchandising strategy to optimize your customer experience, built trust and loyalty, save lost sales and ensure search results are always in step with your ever-evolving business strategies.