Autocomplete

Overview

Athos Commerce's autocomplete function is the helpful list of suggestions that appears instantly while typing in the search box. It surfaces the most relevant search term predictions to your shoppers. Suggestions are dynamically refined as customers type additional words and phrases, delivering the fast and relevant results that your customers expect. Spell Correction is also baked into our latest releases.

Autocomplete adds many benefits for your store. It increases conversions by creating a seamless flow to guide your shoppers, reduces friction and clicks, increases shopper engagement, and pulls in existing merchandising rules and content to give you more control over its behavior.

Interested in showing popular searches to your shoppers when they engage with your search box? Read More »

How It Works

AutoComplete improves the quality of your shoppers' experience by performing several intelligent calculations frequently (on a de-duplicated schedule [50-200ms] when the user pauses in typing). We check for similar terms in your product data, and we run spell correction, if needed, to generate the most relevant search suggestions possible.

All of this happens in the instant before we even show the AutoComplete pop up. In most cases, we'll also do a micro search for the top suggested search to show a preview of its results when the shopper hovers their cursor over a suggested term. Then, once the shopper clicks on a suggested term, AutoComplete reroutes the shopper to the results page for the suggested search term. (If the user clicks a specific item, they are routed to the product detail page instead of the product listing page.)

The Process

When a shopper enters text into a search field, they may click <Enter> to search for their original query, or they may select a query suggestion from the autocomplete container. If the shopper decides to search with their original query, the search passes through the Athos Search API before generating a result. The Search API uses Expanded Search to look for a similar match in the product catalog (as opposed to a Strict Search, which attempts to find an exact match to the entered term). If this generates a result, the search results are presented to the user. 

Note: Additional configurations can be enabled to make expanded search provide results that are more relevant than in the past.

For example, when strict noun detection is enabled in advanced search settings (available to Athos admins only), expanded search will require the search query to match a detected noun in addition to the requirement of "at least one term matching in your data". To activate Strict Noun Match, speak to your customer success representative, as these settings are restricted to Athos admins only.

If this does not generate a result, the system looks for a suggested query. If a suggested query exists, then the search instead will be conducted with the top suggested search term. If no suggested queries exist, then the shopper will see a zero results page.

A search with a suggested term may or may not generate a result. The latter, for instance, may occur if an item is out of stock, and the user might be rerouted to a zero results page, or the out-of-stock item might be pushed to the bottom of the search results. (This behavior is determined by the merchant.) However, more often than not, the shopper is rerouted to a results page for the suggested term, which is accompanied by a note that the search was conducted with a suggested term. 

Displaying to the User

The pop-up modal that the user sees will usually look something like this:

As the user types out their search query, the Spelling Correction Algorithm returns search suggestions in real-time, and presents these suggestions in the AutoComplete container.

The suggestions themselves are listed on the left side. If the shopper hovers their cursor over a suggestion, the autocomplete container displays a preview of the product results for that query.

If the user hovers their cursor over a suggested search query, the autocomplete container displays the results that the user would see if they complete the search with the suggested term. 

In the example above, the shopper typed "dresz". The top suggestion is "dress", for which product results are previewed to the right — including filtering options. 

When shoppers press <Enter>, they are taken to a product result page that corresponds to the original search query entered by the shopper. If the search query has a matching result in the product catalog, the shopper will see search results for that query. If the search query does not have a matching result in the product catalog, the shopper will see search results for the top suggested search query. The latter includes a notice that there were no results for the intended search query and so we instead searched for a suggested term. 

If the original search query does not generate a result, we instead conduct a search for the top suggested search query and inform the customer that "Hmm, we couldn't find anything for "what you searched for", so we searched for "a suggested query". In the example below, a search for "matte" found nothing, so the system provided the search result for "mats" instead.

Want Something Different?

Older versions of Spell Correction may function differently. V3 of Spell Correction forces shoppers to use suggested queries each time they hit enter. This also included a “Search instead for” prompt to allow shoppers to return to their original query. If you’d like to upgrade to the new version of Spell Correction, please reach out to your Customer Success Representative to request an upgrade to your site's behavior.

AutoComplete Matching Logic

While a user is typing, we look at both your product data and popular queries to match against. 

Note: To learn about how we determine popular queries to match against, see How Autocomplete suggestions are created.

Order of Suggestions

Suggestions must always be a prefix match with words. For example, "dr" can find "dress", but not "eardrops".

1. Exact match from either popular queries OR product-data appears first (e.g., Exact SKU match)
2. Popular queries
   • "Complete the sentence" using the difference between user's query and popular query
   • Shortest query first
3. Product data generated (Name and SKU core fields)
   • "Complete the sentence" using the difference between user's query and product data
   • Shortest query first

Process

We begin by processing popular search data to create a list of the top 5,000 most popular queries we can reference for your site. When a user searches, the Suggest API looks at the query and returns suggestions in this order:

• Inferred queries to top of group
• Show shorter similar sibling queries first
• Looks at site product's "Name" and "SKU" to find matches from the user’s query
  • Fully typed words (space after) must fully match terms from Name or SKU field (SKU prioritized)
  • Partially typed words prefix match terms from Name or SKU field (SKU prioritized)

Inferred Queries (Complete the Sentence)

We look at this group of suggestions and see if the user’s query has a prefix match for any words in the current suggestions and then create a new suggestion. This is intended to find a closer representation of a suggestion to what the user is typing — attempting to complete the currently typed word.

For example, if the user types "blue dress p" and one of the existing suggestions is "blue dress with polka dots", we'll intelligently grab the next word from that phrase that starts with "p" and append it, resulting in "blue dress polka"

Shorter Similar Sibling Queries

If, within the popular queries, there is a suggestion that is a prefix exact match with one of its neighbors, the shorter one will be swapped above it. 

For example, if both "green dress" and "green dress with stripes" are suggestions, we'll show "green dress" first.

Frequently Asked Questions (FAQ)

What about terms that are set up as synonyms?

Synonyms created in Athos will be reflected in autocomplete suggestions.

Is it possible to apply background filters to the autocomplete results?

Yes. Background filters can be added to the autocomplete results, although you may need to ask us for the specific code needed depending on your ecommerce platform.

Can IntelliSuggest be removed from autocomplete?

Yes, though you will need to reach out to Athos support for additional assistance.

Why does autocomplete not close when I select a filter?

Autocomplete allows for filtering down the results within the autocomplete window so that the results can be narrowed; therefore, it remains open until a user clicks on a product or outside the autocomplete window. Behind the autocomplete window, the search results page automatically updates according to the selected filters.

How do I update from V2 or V3 to Snap autocomplete?

An upgrade to autocomplete requires a mockup and an update to your contract with Athos. Please reach out to your Customer Success Representative to discuss this upgrade to your account.

Why does autocomplete flash and then go away?

If your site integration is new, autocomplete has not had time to learn user interaction. As a result, it is defaulting to Google Analytics data, which contains search terms that users have entered. These terms may or may not be products in your store. When autocomplete determines there is not a match in the database for the search term, it automatically closes, giving this flash. This resolves itself with time and user interaction.

Why are the autocomplete suggestions and the search results sometimes different?

The autocomplete API will only search for the core Name and SKU fields, while the regular Search API will search every field that is marked as "searchable" in the Athos dashboard, thus returning a larger and more varied selection of products.

When to contact Athos support:

• Product links go to the wrong product pages
• Prices aren't accurate
• Image URLs aren't populating