When e-commerce search works, it’s fast, convenient and efficient.
It’s no wonder that so many users prefer searching over clicking
categories. Unfortunately, our recent large-scale usability study and
top-50 benchmark of e-commerce search finds that search often doesn’t
work very well.
On-site search is a key component of almost any e-commerce website.
That’s why we at Baymard Institute have invested months conducting a
large-scale usability study, testing the e-commerce search experience of
19 major e-commerce websites with real-world end users.
We’ve boiled our findings down to 60 usability guidelines for
e-commerce search design and logic. Based on these findings, we’ve
benchmarked the search experience of the 50 top-grossing US e-commerce
websites, rating each website across all 60 search usability parameters.
In this article, we’ll present some of the findings from this vast
research data set and explore the state of e-commerce search.
Benchmarking the search experience of the 50 top-grossing US
e-commerce websites reveals a surprisingly grim state of affairs. Here’s
a walkthrough of just a handful of the most interesting statistics:
- 16% of e-commerce websites do not support searching by product name
or model number, despite those details appearing on the product page! An
additional 18% of websites provide no useful results when the user
types just a single character wrong in the product’s name.
- 70% require users to search by the exact jargon for the product type
that the website uses, failing to return relevant products for, say,
“blow dryer” if “hair dryer” is typed, or “multifunction printer” if
“all-in-one printer” is typed.
- Searches with symbols and abbreviations are not supported by 60% of
e-commerce websites. For example, the websites do not map the double
quotation mark (often substituted for the double prime symbol: ″) or
“in” to “inch.”
- Autocomplete suggestions are found on 82% of e-commerce websites.
While some implementations greatly enhance the search experience, 36% of
implementations do more harm than good.
- Only 34% allow users to easily iterate on their query by prefilling
it in the search field on the results page, despite the fact that,
according to our tests, users frequently need to iterate on their query —
on average, 65% of test subjects required two or more query attempts
during testing.
- Only 40% of websites have faceted search, despite it being essential
to e-commerce search because it is the foundation of contextual
filters.
- While an e-commerce website eases navigation by offering two types
of breadcrumbs, 92% of the top-50 websites display only one breadcrumb
type or none at all.
Below, we’ll walk through each of the above statistics and provide insight on
how to improve the search experience and success rate on your e-commerce website. We’ll round the article off with a general analysis of the current state of e-commerce search.
1. 34% Do Not Support Users’ Searches By Product Name
At the heart of a good e-commerce search experience is the search
engine’s ability to return relevant results for the query. While our
usability study identified
12 unique types of search queries
that users make on e-commerce websites, let’s stick to the most basic
ones here, because even those prove troublesome for a significant
portion of the top e-commerce website.
The simplest query type is an exact search, whereby
users search by a specific product name or model number. Yet 16% of the
top benchmarked e-commerce websites do not support searches for product
names or model numbers that appear on the respective product pages. The
reason is probably that the search logic of these websites is based
entirely on matching the user’s keywords against the product title or
main product description, instead of the full data set. Whatever the
reason, it makes for a poor experience, with grave consequences.
During testing, users would (rightfully) conclude that a website that
displays no results for a query so specific means that the company
doesn’t carry the product. And if you’re wondering how many users
actually searched by product name or model number, they frequently did
during our testing. Multiple test subjects began their initial product
research on an external website, starting with a Google search, or on a
manufacturer’s website, a review website, etc. Only when they had
decided on one or more products did they copy and paste the product’s
name or model number directly from the external website into the search
field on one or more e-commerce websites.
nother search pattern for exact searches that is ill-supported by the
top e-commerce websites is phonetic misspellings. 18% of websites
handle phonetic misspellings so poorly that users would have to be able
to pass a spelling test in order to get relevant results. For example,
the query “Kitchen Aid Artysan,” rather than “KitchenAid Artisan,”
yields no results. While misspellings can occur in any scenario,
phonetic misspellings are particularly common when users have only verbally heard a product’s name
and later try searching for it. This isn’t unusual in the discovery
process, if we consider how often products are recommended by friends
and colleagues.
Suggesting the correct spelling
on the results pages is recommended, then. And if the search engine
automatically processes the query when it finds no matches or a few poor
matches for the user’s spelling, even better.
Implementation Tip
Assuming that users will spell perfectly is unreasonable. And it’s
important that the search logic broadens the query’s scope and looks for
closely related spellings, especially if only a few results of low
relevance would be returned. Furthermore, the logic should search the
entire data set of products to include matches for product names and
model numbers.
2. 70% Require Searches In The Website’s Jargon
Another common search query type in e-commerce is the product type
query, whereby a user knows the type of product they want but doesn’t
know of or hasn’t decided on a particular product. So, they simply
search for, say, “blow dryer” or “multifunction printer,” two queries
that often lead to no results because the “correct” name for that
product type is “hair dryer” or “all-in-one printer.”
To fully support product type queries, the search engine’s logic must
go beyond the exact titles and descriptions of products, and look to
the categories that products are placed in, as well as take product
synonyms into account.
However, only 30% of e-commerce search engines include keywords from each product’s parent category or
map a set of synonyms to product titles, product types and category names.
As a result, users have to use the website’s exact jargon for their
product type queries on 70% of websites, as seen in the two examples
above.
Product type synonyms were observed to have a particularly large
impact on the user’s search experience during testing, because it can be
very difficult for users to tell whether they are using a wrong term,
and even more so to guess what the “correct” term is. Therefore, a lack
of synonym support was observed to be a direct cause of website
abandonment, because users assumed that the website doesn’t carry the
products that they searched for.
Implementation Tip
At the very least, manually map common product type synonyms to the
actual product types and category names. A better long-term solution
would be to build keyword synonym logic, because this can more easily be
updated, tweaked and even personalized on a regular basis.
3. 60% Don’t Support Searches With Symbols And Abbreviations
Some products have specifications that are vital to the user’s
purchasing decision. When searching for such products, our test subjects
frequently included those specifications directly in their search
queries (for example, “13″ laptop sleeve”), making them feature queries.
But do users usually search for “13 inch laptop sleeve” or “13″
laptop sleeve”? Of course, both queries should present exactly the same
results, because
users have no way of knowing how each
website denotes such units of measurement, and all ways are equally
“correct” and used interchangeably across the Internet. In case you
think we’re stating the obvious here, 60% of the top e-commerce websites
do not support searches with symbols and abbreviations. So, users will
miss out on perfectly relevant products if they search for “inch” and
the website uses the double prime symbol (″) or the abbreviation “in,”
or vice versa.
Breaking down such technical hindrances and mapping common symbols
and abbreviations is important, so that users find the products they are
looking for and get the same results regardless of what a website or
its suppliers have decided to use. Most websites are small enough that
synonyms, abbreviations and full spellings could be manually mapped for
the most common units.
Implementation Tip
Map all common symbols, abbreviations and full spellings to each
other, so that all results are shown regardless of how a query is
written. A quick way to identify candidates for mapping is to look for
units of measurement in all product titles. A more thorough way is to
compile a list of units of measurement by going over all product
specifications.
4. Auto-Suggestions Found On 82% Of Websites
Auto-suggestion is a convention of e-commerce search, with 82% of large websites offering it.
The value of autocomplete suggestions isn’t that they speed up the
typing process, but that they guide users to better queries. When
auto-suggestions are done well, they teach users the types of queries to
make, show them correct domain terminology, help them avoid typos and
assist them to select the right scope in which to search.
During usability testing, autocomplete suggestions directly
influenced and altered what the test subjects decided to search for.
While this is their purpose, it also means that
autocompletion can do more harm than good if not implemented carefully.
Among the websites that do have autosuggest, 36% of them have
implementations with severe usability problems. Two problems frequently
observed in testing are query suggestions that either are repetitive or
lead to a dead end.
While avoiding dead ends might seem obvious, such suggestions were
observed multiple times during benchmarking. They were often the result
of auto-suggestions being based on the prior searches of other users or
old catalog contents, regardless of whether they proved to be useful.
Developers should at least internally query all suggestions on a regular
basis and weed out those without any results.
To provide high-quality search suggestions, look at how suggestions
are generated. Many suggestions that are redundant, of low quality or
typos are likely the result of developers sourcing suggestions from the
website’s search logs. If this strategy doesn’t take into account the
success of those queries (that is, whether a decent percentage of users
found and purchased products after performing those searches), then it
is flawed.
Unless you track the success of search queries,
don’t use search logs to generate auto-suggestions
because that would result in redundant and low-quality suggestions.
Aside from outright duplicate suggestions (as seen in the Overstock.com
example above, which could easily have been filtered out), redundant
suggestions are ones that overlap and make it difficult for the user to
select one over the other. Notice in the Overstock.com example how five
variations of “coffee table” are suggested, despite the user having only
typed “coffee” at this stage. These nuances might have made sense if
the user had typed “coffee ta–.”
We also found that copying a suggestion to the search field when the
user focuses on it with their keyboard (as illustrated above) is an
important detail in autocompletion design because it enables users to
iterate on a suggestion.
(We identified
eight autocomplete design patterns during testing.)
Implementation Tip
Given that autocompletion design and logic will directly alter what
most users search for, ensuring the high quality of suggestions by
weeding out dead ends and being selective in the inclusion of
suggestions is vital. Suggestions based on other users’ past queries
should be carried out with special care or avoided entirely; ideally,
any machine learning should be based on the success rate (or conversion
rate) of each query.
5. Only 34% Prefill The User’s Query On The Results Page
During testing, 65% of all test subjects’ attempts at searching
consisted of two or more queries in the same search. However, only 34%
of e-commerce websites allow users to easily iterate on their query by
prefilling the query in the search field on the results page.
On websites that do not persist the user’s query in the search field,
the iteration process became needlessly cumbersome and easily
frustrated subjects. Not persisting a query introduces friction at the
worst possible time because redundant typing is added to the already
disappointing experience of not receiving relevant results.
The amount of time that test subjects spent retyping their query is
insignificant, but as observed in all of our prior usability studies
that involve filling out forms , the user has a negative perception of a
website that forces them to retype the same data within a short
timeframe, which often sparks remarks such as “Tedious,” “Idiotic” and
“Do you think they’ve tried using their own website?” This is especially
true on touch devices, where typing is particularly taxing.
The picture was completely different on those websites that persist
queries on the search results page. Here, test subjects weren’t forced
through a needless halt-and-retype process each time they wanted to
iterate on their query, but instead made swift changes by adding or
removing a word or two from their original query, as seen in the Zappos
example above, where the user simply added “11″” to his prior query.
Implementation Tip
Given how relatively simple this is to implement, persisting the
user’s query in the search field on the results page can be considered
low-hanging fruit in search optimization.
6. Only 40% Have Faceted Search
In a perfect world, we would have little need to filter and sort
search results because users would make precise queries, knowing exactly
what they want, and the website’s search logic would return just the
right results.
This is far from reality, however;
filtering and sorting are vital ways
that users find the right product among the results. This is partly due
to the challenge of getting search logic and design just right (as
we’ve hinted at in the preceding five points), but also partly due to
how and when users search. Users will not always be able to perfectly
specify their queries, simply because many still haven’t fully decided
or realized what they are looking for.
In both cases, being able to modify search results by filtering and
sorting is a powerful and important tool. During testing, the quality of
the filtering and sorting features and their design often meant the
difference between success and failure in the subjects’ search
experience.
Our testing confirmed that the foundation of a
contextual filtering experience
in e-commerce search is faceted search. With faceted search, the user
is presented with a list of filters for product attributes, filters that
apply only to a part of the search results. For example, the search
results for “Tom Hanks” could have a “movie duration” filter even though
the results include books, and the search results for “down filling”
could have a filter for “sleeping bag temperature rating” even though
the results include other product types.
The traditional way of suggesting only generic scope filters
(categories) and site-wide filters (price, brand, availability, etc.)
for site-wide search results is insufficient for a good experience.
Product-specific filters based on the user’s query must be suggested,
too. However, only 40% of e-commerce websites currently do this via
faceted search filter suggestions.
While faceted search is a crucial component of search filtering, it
doesn’t make for a good filtering experience on its own. Also crucial
are the filtering types (such as thematic filters), the filters’ design
details and the filtering logic (for example, avoiding mutually
exclusive filters, as explained in the third point of “
Best Practices for Designing Faceted Search Filters”).
Furthermore, faceted search filters were observed to have usability
issues of their own. One challenge is that when the filters also invoke a
higher-level scope, they need to clearly indicate this in their label.
Otherwise, users will likely be misled because they have no way to
accurately predict the implications of applying the filter, as
illustrated in the REI example above.
Faceted search’s
labelling issues aren’t solved simply by including the filter’s context in the filter label
(for example, permanently having the filter read “Sleeping Bag
Temperature Rating”). That would hinder users who have already applied a
search scope (and users who are using category navigation) because it
would make the labels needlessly difficult to scan due to a lack of
front-loaded information and a poor signal-to-noise ratio. Therefore, if
faceted search filters invoke a scope, then a dynamic labeling system
is needed to keep the filter labels concise and scannable when the user
has already selected a context (for example, navigated to a category or
applied a search scope) and then dynamically rename the filter labels to
indicate the scope-related implications of applying that filter.
Implementation Tip
Don’t simply rely on generic site-wide filters, such as category,
price and brand. Rather, provide product-specific filters that relate
directly to the user’s query (through faceted search). If the faceted
search filters invoke a scope, then the filter labels need to be
dynamically renamed to indicate this. Also, consider whether sufficient
filtering types are available. For example, thematic filters such as
style, season and usage context often map closely to users’ purchasing
parameters.
7. 92% Have Only One Breadcrumb Type Or No Breadcrumbs At All
During testing, breadcrumbs proved to be helpful for test subjects
when navigating both search results and when looking through categories
to find just the right product. Interestingly, testing also revealed
that e-commerce websites need two different types of breadcrumb links —
namely, hierarchical and history-based breadcrumbs. Yet, 92% of the 50
top-grossing e-commerce websites display only one breadcrumb type (72%)
or no breadcrumbs at all (20%).
Without breadcrumbs on the product page,
users will find it difficult to efficiently browse a collection of products,
because they have no way to go one level up in the hierarchy to the
product category or to return to the search results page. In practice,
this often forces users to make a drastic jump in scope, such as
selecting a generic top-level category, or else perform a new search or
remain stuck on the product page.
With traditional hierarchical breadcrumbs, any user who doesn’t find a
particular product to be a good match can use the breadcrumbs to
traverse up the website’s hierarchy and navigate to a related category.
This is paramount for non-linear navigation such as search, because it
enable users to see other products in the same category as an item in a
search result. The hierarchy essentially acts as a cross-navigation link
for finding related products, regardless of whether the user has
accessed the category from a completely different part of the website.
(The same non-linear behavior was observed to hold true for all external
traffic landing directly on product pages.)
During testing, it quickly became evident that most subjects had a
strong desire to go “one step back” after exploring a product page. This
typically meant going back to the search results list, which
history-based breadcrumbs are well suited to. History-based breadcrumbs
are, as the name implies, based on the user’s history, giving the user a
way back to previously visited pages.
While this functionality is also available in the browser’s interface
via the “Back” button, test subjects repeatedly got stuck or were
misguided on websites that offer only one type of breadcrumb. For
example, when only hierarchical breadcrumbs were available, many
subjects confused them as a way back to their search results. As a
consequence, they unwittingly switched their product-finding strategy
and lost any filter or sorting settings they had applied, thinking the
last hierarchical breadcrumb link would take them back to the search
results page.
A simple “Back to results” link alongside the standard hierarchical
breadcrumbs enables users to seamlessly go back to their search results,
with filters and sorting choices intact. History-based and hierarchical
breadcrumb links are an ideal combination, allowing users to
efficiently continue their current search session or switch to a new
navigational mode.
Implementation Tip
Implement two types of breadcrumbs on product pages: hierarchical
breadcrumbs, which allow users to infer and jump to categories that
contain the current product, and history-based breadcrumbs (such as
“Back to Results”), which minimize misinterpretation of hierarchical
breadcrumbs as a way back to the search results. Testing confirms that
history-based breadcrumbs can be both appended (as on Macy’s) and
prepended to hierarchical breadcrumbs.
The State of E-Commerce Search
To give you a more general analysis of search performance in the
e-commerce industry as a whole, we’ve summarized the entire benchmark
data set in the scatter plot below.
Each of the 3,000 benchmark scores is divided into the six major
areas of e-commerce search usability: query types, search form and
logic, autocompletion, results logic, results layout, and results
filtering and sorting. Thus, each gray dot represents the summarized
score of one website’s score across the 6 to 15 guidelines within that
area.
The blue circles represent the actual benchmark average for each
column (an average of the gray dots). The red triangle and green circle
are reference scores that we’ve created for comparison:
- The green circle represents the score for what is to be considered a
“good” search experience — here defined as a website that partly
adheres to all 60 of the search guidelines. That is, the green circle
represents the standard that an e-commerce website should reach (or,
better yet, surpass) in its search experience.
- The red triangle represents the score for a “mediocre” search
experience — here defined as a website that partly adheres to 48 of the
60 guidelines. That is, search engines and designs that reach this
standard can be assumed to directly hinder (or even obstruct) users as
they search.
Besides noting the very scattered score distribution in each column,
the columns to pay attention to are those that show the industry average
(blue circle) significantly below a “mediocre” search experience (red
triangle). This is the case for query types, results layout, and
filtering and sorting — all areas of the search experience where the
vast majority of e-commerce websites have significant room for
improvement.
Query are the very core of e-commerce searchtypes,
yet support for the 12 essential query types is lackluster at best.
Points 1, 2 and 3 of this article are just the tip of the iceberg (you
can find all 12 query types in a
white paper that we recently published), but they clearly demonstrate poor support:
- 16% of e-commerce websites do not support searches by product name or model number.
- 18% handle misspellings so poorly that users would have to pass a spelling test in order to get relevant results.
- 70% require users to use the jargon of the website, failing to return relevant results when users search with common synonyms.
- 60% do not support searches with symbols or abbreviations of units of measurements (or vice versa).
Given its key role in the search experience, query types are an area
that sorely needs to be prioritized on the vast majority of e-commerce
websites, and they should be seriously considered and evaluated in any
optimization project. When evaluating the resources required, remember
that improvements to search engine logic would benefit all platforms
(desktop, mobile, tablet, etc.), whereas layout changes are typically
platform-specific.
Testing revealed that
the results layout is a balancing act
of designing a clean overview of search results and providing
sufficient information for users to accurately evaluate and compare
results. However, the benchmark of this metric tells a grim story, with
the best websites doing merely OK, and the other half of websites
performing poorly.
A common cause of poor results layout is that the website relies on
the same (static) layout for both search results and category product
lists. From our testing, search results clearly need a more dynamic
layout that adapts to the user’s query and context. This could include
altering how much and which information is displayed for each result,
which product thumbnail is displayed, how large the thumbnail is and so
on. All of these elements should dynamically adapt to more closely match
the user’s query and expectations. To some extent, this also includes
the product page’s layout, which could have dynamic links, such as
history-based breadcrumbs, along with the traditional hierarchical
breadcrumbs (see point 7 in this article).
Optimizing the results layout is a relatively manageable project,
which mainly entails switching from reusing the static (category)
results layout to a dedicated and slightly more dynamic search results
layout. It should, therefore, be considered low-hanging fruit, given the
large impact it can have on the overall search experience, especially
during the product-selection process.
Filtering and sorting search results is a somewhat overlooked area.
Notice the highly scattered plot in its column and the fact that nearly
all websites miss out on important aspects of it, as indicated by the
threshold for a “decent” search experience (red triangle). Just like the
results layout, filtering and sorting features should adapt to the
user’s query and context.
For example, while faceted search (see point 6 in this article) is
the foundation of a contextual filtering experience, only 40% of
websites have it. Worse, the multiple elements of sorting site-wide
search results that we identified during testing are overlooked
entirely, with more than 70% of websites lacking key sorting types, and
90% having no scope options or suggestions when users try to sort
site-wide results.
Given that filtering and sorting are much less resource-intensive to
get right than query support, they should be a part of almost every
optimization project for e-commerce search. Moreover, many of the
improvements are manageable enough to be implemented and optimized on an
ongoing basis, and much of it can be reused to improve the sorting and
filtering experience in category navigation.
Search: A Competitive Advantage
The gloomy state of e-commerce search doesn’t mean that users cannot
perform and benefit from search on the benchmarked websites. However, it
does clearly indicate that e-commerce search isn’t as user-friendly as
it should be and that users’ success rate could be improved dramatically
on most websites — even those of these 50 e-commerce giants.
While catching up with the few websites that have done really well
from years of focused investment would require a serious prioritization
of the search experience, it is achievable. Furthermore, because the
poor state of search is industry-wide, most websites have an opportunity
to gain a truly competitive advantage by offering a vastly superior
search experience to their competitors’.
A good start would be to look into the seven points we’ve presented in this article:
- If few results of low relevance are returned, the search logic
should broaden the scope and look for closely related spellings (18% of
websites don’t). Furthermore, the logic needs to search through the
entire product data set, to include matches for product names and
copied-and-pasted model numbers (16% of websites don’t).
- Map common product-type synonyms to the spellings used on your
website to ensure relevant results for a query such as “blow dryer” if
you refer to it as a “hair dryer,” or a query such as “multifunction
printer” if “all-in-one printer” is used (70% of websites don’t).
- Map all commonly used symbols, abbreviations and full spellings to
each other, so that all results are shown regardless of how something is
written in the product data. For example, map “inch” to the double
quotation and double prime symbols and to the abbreviation “in” (60% of
websites don’t).
- Be cautious about auto-suggesting based on other users’ past queries
because that often leads to low-quality and redundant suggestions.
Furthermore, regularly check that auto-suggestions don’t lead to a dead
end (36% of the websites with autocompletion don’t do this).
- Allow users to easily iterate on their query by prefilling it in the search field on the results page (66% of websites don’t).
- Implement faceted search to suggest filters that match the user’s
query more closely. For example, suggest product attribute filters that
apply to a subset of the search results (60% of websites don’t do this).
- On product pages, provide both traditional hierarchical breadcrumbs
(to support non-linear patterns of search) and history-based
breadcrumbs, such as “Back to results” (72% of websites offer only one
type).
Because a poorly performing search experience can look as good
aesthetically as a high-performing search experience, gauging one’s own
or a competitor’s search experience requires extensive testing and
evaluation. The fact that search experience and performance are heavily
influenced by non-visible factors, such as search logic and product data
integration, is actually good because the competitive advantage you
would gain from investing in them cannot be easily copied by competitors
(unlike, say, a home page redesign). So, while creating a truly great
search experience will probably require substantial resources, it’s also
an opportunity to create an equally substantial and lasting competitive
advantage, one that competitors cannot easily piggyback on.
As a final note, the findings from our usability study give owners of
small e-commerce websites a fair shot at improving their search
experience, because roughly half of the 60 guidelines relate to user
interface. This is especially true of the results layout and the
filtering and sorting experience, which are areas that are usually easy
to change but whose performance on most websites is currently below
expectations.