Spec sheet

The engine will sort search results in real-time based on a visitor's unique personal preferences (e.g. brand, gender, size, or other attribute in the product feed), using browser cookie or user ID.

The engine will sort search results in real-time based on a visitor's unique in-session purchase intent, using browser cookies or user ID. Intent is inferred from the area/context of the catalogue in which the visitor is interacting during a single session. The contextual mapping of products is done during the algorithm training and is a core component of Loop54's proprietary AI.

Personalised experience can be reflected across all the devices a user might use if the user is identified via login ID or user name.

Query terms, ranked by popularity. Popularity of suggestions are determined by the frequency of searches for that term. The term's popularity is built up by previous searches for that term that led to a good textual match.

The top autocomplete suggestion can be presented with facets (or scopes in other words). To enable users to apply a facet, directly from the search bar. Typically the facets used are brand or category. (e.g. "Prada in Accessories" or "Prada in Shoes")

If a user enters a pre-defined keyword (e.g. opening hours), they are redirected to a specified URL immediately after pressing enter. The engine returns a link to use for the redirect.

Products are returned (along side queries) when a user starts to type in the search box. If a user starts to type "airp" in the search bar, the engine will respond with the "Airpods" product.

Other things than queries and products can be returned. For example results within categories, brands or other content. This can be used to create a rich Autocomplete experience.

By default, all search results returned by the engine are ranked by relevance. Relevancy is determined through Machine Learning, which combines keyword-matching, behavioural insights, business logic and the network of relationships between products. Learn more: loop54.com/howproduct-search-works

Popularity can be determined by collected behaviour on either Click, Add-To-Cart or Purchase (i.e. not a combination of the three). The engine can also use other data sources to determine popularity, such as Sales or CRM data.

Sorting products by any attribute (e.g. popularity, price, newest/oldest, etc.).

Show products in stores based on the visitors location or show stores nearby based on the visitors location.

Any attribute on the product can be indexed, and then, configured and scored independently.

Infinityuse several match types and can find partial matches, reverse partial match, compounded words and fuzzy matches:

• Partial match- users can search for a part of a word and still get a match Reverse partial match - users can search for the end of a word and still get a match

• Compounded words - users can write two words together (that is separate on products) and still get matches

• Fuzzy matches- users still get matches even though they have not typed the query exactly as it appears on the products. This is to accommodate for typos and small variations.

Only show one product as result if search is for SKU or EAN number.

Search for things other than products (e.g. guides, blogs, articles, etc.). List of content results will be returned in a separate list from the product results and will be ranked default by relevance.

When the query contains spelling errors that the engine can interpret with some confidence, it will return spelling suggestions. The engine normally auto-corrects minor misspellings, but if the query is severely misspelled, then the engine will return "best guess" results and provide suggested spelling corrections. The engine is able to determine the most relevant spelling corrections by looking at past queries by other users that led to a direct hit (direct hit = search phrase exists, exactly as typed, in a product's metadata)

If no results are found with regular methods, the engine will edit the search query significantly. The engine tries to break a word down into several words, traversing back through the query until it finds a word that has a match in the catalogue. If the engine still can not find a match from sub-words, then it will drastically edit the search phrase (e.g. by swapping letters, replacing letters, etc.). If users frequently interact with the results the engine will learn the meaning of that word and improve the relevance even further. If users don't interact with the results, the engine will eventually decide to return nothing instead.

In addition to regular results (Direct Results), the engine generates a list of Related Results. These Related Results are products that the engine consider to be very relevant to the query, however, they don't contain the actual word. This way, the engine can still show "pendants" even though the user searched for "necklace".

The engine will display query suggestions on the search page that will lead to similar results as the original query.

Boost or bury results based on any product attribute. For example, promotional campaign, new products, high-margin products, etc. The engine is smart when applying these rules and will only boost a product if it is relevant in the first place. This way, a winter jacket will not be boosted to the top during summer, just because it is on sale.

Faceted filters can be built using any product attribute (date, number or text). Brand, category and price are default. Allow users to select and deselect facets. Applying a facet will narrow the search results and new facet values will be re-calculated based on the filtered results. All faceting options with a value >0 (i.e. the count of products that contain that filterable attribute) will be returned.

The engine can decide what facets are relevant for the current search result and strip away any irrelevant facets. This is particulary relevant for stores with many facets and a lot of broad searches.

Infinitycan limit the assortment of products shown in search results or category listings based on specific customer groups, geographic markets, specific physical stores, or for the purpose of internal diagnostics. The criteria used to customise the assortment does change between syncs. Infinitycan source the groups and their respective assortment information in real-time by communicating directly with retailer's PIM and/or ecommerce platform.

Prices can adapt to the conditions of pre-set customer groups. Prices and customer groups need to be in the feed. Although the number of customer groups used is technically limitless, Infinityimposes a soft upper limit of 100 customer groups. Infinitycan source the groups and their respective price information in real-time by communicating directly with retailer's PIM and/or ecommerce platform.

Unlike many language-specific stemming algorithms that have a set of predefined rules of how to find the root form of a word, Infinityhandles things like stemming and other NLP tasks without any assumptions about how the language is built. Therefore multiple languages can be supported, but each language requires its own engine.

The engine will sort category listings in real-time based on a visitor's unique personal preferences (e.g. brand, gender, size, or other attribute in the product feed).

The engine will sort category listings in real-time based on a visitor's unique in-session purchase intent. Intent is inferred from the area/context of the catalogue in which the visitor is interacting during a single session. The contextual mapping of products is done during the algorithm training and is a core component of Loop54's proprietary AI.

Personalised experience can be reflected across all the devices a user might use if the user is identified via login ID or user namer

The standard sorting for category listings is "relevance" which incorporates any combination of: user bahaviour, sales statistics, business rules, personalisation and other.

Popularity can be determined by collected behavioud on either Click, Add-To-Cart or Purchase (i.e. not a combination of the three). The engine can also use other data sources to determine popularity, such as Sales or CRM data.

Sorting products by any attribute (e.g. popularity, price, newest/oldest, etc.).

Boost or bury results based on any product attribute. For example, promotional campaign, new products, high-margin products, etc.

Faceted filters can be built using any product attribute (date, number or text). Brand, category and price are default. Allow users to select and deselect facets. Applying a facet will narrow the search results and new facet values will be re-calculated based on the filtered results. All faceting options with a value >0 (i.e. the count of products that contain that filterable attribute) will be returned.

The engine can decide what facets are relevant for the current result and strip away any irrelevant facets. This is particulary relevant for stores with many facets and a lot of broad categories.

Infinitycan limit the assortment of products shown in results or category listings based on specific customer groups, geographic markets, specific physical stores, or for the purpose of internal diagnostics. The criteria used to customise the assortment does change between syncs. Infinitycan source the groups and their respective assortment information in real-time by communicating directly with retailer's PIM and/or ecommerce platform.

Prices can adapt to the conditions of pre-set customer groups. Prices and customer groups need to be in the feed. Although the number of customer groups used is technically limitless, Infinityimposes a soft upper limit of 100 customer groups. Infinitycan source the groups and their respective price information in real-time by communicating directly with retailer's PIM and/or ecommerce platform.

Infinitycan be the source of product page information (e.g. variant images, product specifications, etc.). It is often the case that information for the product page is sourced faster from Infinitythen from the eCommerce platform.

This list is meant to be used on a product page and is made up of the products that are similar, or related, to the product being viewed. Filters can be applied to for example show a more expensive product, or higher margin.

The engine can return lists with products that fulfill certain requirements. For example new products, campaign products or popular products. These results can also be influenced by the same personalisation as mentioned earlier.

Our platform is built to be flexible and we can accommodate if you have any particular need in the setup. Please chat with us to get to know how we can help you with your custom solution.

Examples of things we have done for current customers:

• Data synchronization, enrichment and cleaning

• Custom API request/response format (excluding those used for filtering, faceting, sorting)

• Advanced object-based filtering and property inheritance

• Advanced parent-child variant handling

• Communication with third party API

• A new API endpoint

Product feed is synced at least once a day. More frequently if needed.

Delta-information/changes pushed to our API in near real-time, but it should always be complemented with a complete feed that Infinitycan synchronise with at least once a day. Infinitycan fetch changes or retailer can push changes.

If different information needs to be synced, it does not have to be in the same feed as products. For example, Infinitycan enrich the feed with data from other sources, like images, CRM or sales data. Data can also be in different formats. Engine can index different formats from different sources and merge them. (e.g. CSV, XML, JSON).

Products that should be removed from the indext can be pushed through the API.

Infinityalso allow for partial delta updates, that means you can send only the data that needs to be updated.

Engine can handle any static filters based on any attribute. These are filters that do not change between syncs of product feed. (e.g. out of stock, available in store, not yet released). Filters are defined by the API client, so adding a new facet on an attribute.

Filters that are created based on live dynamic user information. Product assortment is filtered based on, for example, specific customer groups, geographic markets or specific physical stores. Can also be used for internal diagnostics.

Global events are the user behaviour events (e.g. clicks, add-to-cart, purchases, etc.) that occur across the entire website. These global events are tracked and used to improve the sorting of search results and category listings (i.e. to determine personal taste and popularity). The engine keeps track of where the event occurred - either within search or navigation. Read developer docs.

Infinityis SaaS (software-as-a-service), this means our engines our hosted in the cloud and new versions can be rolled out without any extra work or cost. To access some new features, retailers may need to implement new APIs.

Monthly Uptime Percentage (MUP) of 99.98%. All network traffic is saved and analysed for volume and customer behaviour in MSSQL DB. Traffic data and server monitoring is sent to ELK (Elastic - Logstash - Kibana) system and Grafana where different stats like response time, sync, memory usage are monitored.

There is the possibility to cache responses.

Possibility to distribute load across multiple servers and work with load balancers.

Infinityhas failover proxies hosted in Sweden and England.

Encrypted SSL (HTTPS) and, if requested, IP-blocking to backend servers.

Unlike Google Analytics, which only provides unique searches, Infinitywill provide total search volume with adjustable time periods. The Insights Dashboard will also display analytics regarding search trends over time, changes in product rank order, and conversions rates for all queries (click, add-to-cart, purchase).

Ability to see queries that the engine could not make sense of "as is" (i.e. the words do not exist anywhere in the catalogue). When "MakeSense" is False, the engine will edit the query to find a match. Being able to view search analytics on "MakeSense False” queries gives an indication of what new words the engine may have already learned or may eventually learn (i.e. in order for engine to learn these new words, "Make Sense False" results must be displayed to the user and behaviour events must be sent to Loop54).

Infinityprovides API so some search insights collected by Infinitycan be exported - only from the backend. The specific insights that can be exported are: all search queries that led to a purchase, and the count of units purchased per product for that search query.