An immersive dance

Search is a unique feature in your website. Your visitors can express themselves in their own words. They have the control and decide in which direction they go. By searching, they experience much more freedom compared to browsing with your menu. And search came a long way from being a “message in a bottle”, in the times where visitors used to type long searches by themselves and hoped for the best when pressing enter. But now, since the first character typed, visitors are guided contextually with personalized queries, products, content, and communication messages. At every keystroke, they can choose to continue leading by typing or to be guided by clicking. This delicate, open, and multistep experience is what we call the search dance.

Open

This dance between the visitor typing and the system guiding can be highly immersive as the visitor discovers not only what matches his search, but also what he could be searching for. On the other side, thanks to all the typing and clicking behaviors, your website can continuously learn to improve its guidance and become a “better dancing partner”.

In this White Paper, we will look at the most important roles and the impact this immersive search dance can have on the success of your online business.

Win by focusing on interactions rather than features

Without solid technical features, it is not possible to provide a good search experience. So much is clear. But search features, like error corrections, synonyms, or even natural language processing, are too often presented as the central piece of the solution. We will discuss such features as well and they have their importance for sure. But they are only there to support the visual experience flow.

At the macro level, Customer Journey optimization focuses on all visits throughout all your channels over time. But here, the experience flow can be viewed as each footstep in this search dance. We focus here at the micro-level, as the experience flow is defined by each event happening not only between pages but within each page of your website (and in the search case, even between each keystroke).

The 10 winning interaction search steps

Open

1: Invite to search

The large white search bar over the dark header background of amazon.com says it all: “I am big, I am important, you should consider me if you want to find something”. If we consider that the large majority of visitors search for only one or two words, it’s on purpose that it was made disproportionally large, so it can send the message: “Here is the most important feature of the website, use it!”

In order to figure out the right way to display your search bar, for its position, size, and visual appearance to be as effective as possible, it is important to do A/B Testing on the layout, content, and visual style (maybe even on default texts inviting visitors to search or exemplifying what they can search for).

This can be done using Boxalino Narrative Layouts with A/B Testing and your online optimization team can improve all these aspects with close to no involvement of your IT team.

2: Individualize right away

No need to wait until visitors already started typing before guiding them. Start immediately displaying the Autocomplete, with for example the visitor’s last searches, as soon as the search bar is selected (a very useful feature which shows that personalization can really be helpful even in its simplest and most direct form). You can also inspire with “current search trends” or even better, with searches related to the prior clicks of each individual visitor.

For example, galaxus.ch displays the last searches the user made, which is already helpful and a good start.

This can be done by using Boxalino Semantic Autocomplete which can tap both into the visitor’s search history as well as on frequent searches matching his prior behavior. Personalized product and content recommendations can also be displayed using Boxalino Neural Collaborative Filtering (NCF).

3: Guide since the first keystroke

4: Bring focus early on

5: Inspire with refinements

6: Promote with merchandising

For example, in flaschenpost.ch Boxalino Automated Semantic Filters automatically extract the meaning of each search word and display a promotional marketing message on top of the search results.

7: Match meaning

The definition of the search results becomes more complex when faced with the following situations:

But in fact, all these situations have something fundamental in common: the learning of what the user actually means and the learning of what is a good match for it.

It is therefore important to correct or adapt the input of the user to match your terminology and it is also important to filter and rank the results to match what the user actually expects.

The adaptation of the user input to your terminology can be done by Boxalino Semantic Parsing which automatically corrects the user input, learns important synonyms, finds the word roots, and matches the word compounds.

The filtering and ranking to match the user expectation can be done with Boxalino Semantic Filters to define the relevant segments of your product assortments and with Boxalino Self-Learning Ranking for the precise ordering of the results in these segments.

8: Recover fast

Sometimes, there is just no matching results. In can be because of a wording problem, because you don’t sell the searched products or because only some of the searched words have any results, but not their combination.

In such cases, pretending that there are search results and showing something which is not what the user expects is likely to be counter-productive. Instead, it is better to indicate that there are no matching products and to suggest interesting alternatives as well.

On puresense.ch a search for a combination (category and brand) that doesn’t exist will show both results separated (first the results of the category and then the results of the brand) while clearly indicating to the visitor that his full search didn’t match any results.

On haarprodukte.ch a search for a brand that is not sold will not bring any results but will show a selection of personalized product suggestions to motivate the user to continue his visit regardless.

This can be done by using Boxalino Narratives Layouts to manage the layout of your search page, so that you can configure different layouts for different situations (e.g.: standard layout if there are results, sub-phrases groups layout if they are results only for parts of the search and personalized product recommendation slider in case there are no results at all).

9: Empower to explore

Showing products without requiring scrolling is quite important, just as showing the expected product first. This is absolutely clear. But what if there are several pages of results and that it is not clear what the user really wants to buy? In such cases, the filtering options (of facets) play a very important role, but only if they are visually appealing.

Here are the main things to consider:

1. Top facets: showing a small selection of important facets on top of the results can play a very important role

2. Only relevant facets: hide all facets which are not related to the search results (for example, a size facets, when only 5% of the search results have any value for size)

3. Pre-expanded facets: Instead of showing all your facets collapsed, and then show all values when expanded, select a few facets (3-5) and show them already expanded with the top 5-7 values (display a “show more” link below to see the other values). This will highlight the most helpful filtering options without requiring the user to click on anything.

4. Facet Icons: using icons for each facet facilitates the user experience and makes the facet look more helpful, easier to understand, and to remember

5. Different visualizations: use a slider for some of your numerical facets and a search box for facets with a large number of values, as well as many other visualizations depending on each facet’s data.

Here is an example for flaschenpost.ch with top facets and 3 expanded facets on the left. All this without the user needing to click on anything.

This can be done using Boxalino Dynamic Facets and defining different visualization parameters for different groups of facets.

10: Personalize relevance

The products (and content) shown on top of the search result list are crucial. If they are relevant, the user might not only click on them but will be much more likely to look at more products in the list below. Therefore, to select the results not only as a “best for all” but in a personalized way is a very important feature of a good search.

But personalizing the products appearing on the search result page is a delicate process for several reasons:

1. Prior purchase behaviors don’t always help (e.g.: a brand the visitor likes for a certain type of products might not be what he likes when searching for a totally different type of products)

2. Search results should not change too quickly (the visitor might want to return to his search result page later and the behavior he had in-between should not change the listing, especially if the user was in the process of paginating)

3. Some personalization criteria might be much more important than others depending on the search term (e.g.: visitors searching for a need might care much less about brand preference than when searching for a function)

For these reasons, we do not advise to use general “one-size-fits-all” personalization algorithms blindly, but instead to compare the effect of different personalization logics with A/B Testing.

One Personalization logic we highly recommend to test is Customer Clustering. Find out more in our recent Case Study with PerfectHair.

There is another type of personalization that can be very impactful, Boxalino Real-time Results Rating. Instead of changing the list of results with personalization behind the scene (implicit personalization), you can let the visitor indicate the results he likes or doesn’t like (explicit personalization). This way, the visitor can simply select a thumbs-up/down on any of the returned results (you can also use an X and a pin or other visual indications of what the visitor considers a good match or a bad match). The search results are then updated with the user feed-back taken into account (e.g.: similar products to a thumbs-up go up in the list and similar products to a thumbs-down go down in the list).

The underlying technology

In the 10 winning interaction search steps, the following technologies are referred to. We highlight here the specific ways Boxalino Winning Interactions Platform covers these functionalities with Data Science and Artificial Intelligence.

1. Narratives Layouts with A/B Testing

Boxalino Narrative REST API is the backbone of any integration and is a full-fledged Layout API. This means that, unlike other systems, Boxalino doesn’t provide you only with personalized content to be displayed on the page, but the entire layout of the page with the final content, ready to be rendered as provided. This means less work to develop your front-end, as most of the business logic of any page is done in the configuration of Boxalino, but also full freedom for your marketers to personalized, contextualize, target, and A/B test any aspect on any page driven by Boxalino Narratives.

2. Semantic Autocomplete

Boxalino processes your products and behaviors data in the Boxalino Data Science Eco-System built in the Google Cloud Platform (GCP) with Google’s best technologies (incl. Google BigQuery). As a result, Boxalino creates an index of queries to be suggested to the visitors, built with:

• Smart grouping of synonyms queries together (so that you don’t recommend two queries which are in fact just two different ways to write the same thing)

• Promoted position of all your important product properties (e.g. brands will appear on top if searched, even if the most frequent typed version doesn’t match exactly the brand name)

Personalized Query Suggestions

Boxalino analysis the online behavior of each one of your visitors and can include the most important and effective attributes to be used as personalization criteria:

• One-to-one personalization based on the prior visitor’s query

• Behavior-based positions (the searches are smartly ranked according to their popularity, based on a defined time-moving window letting new search trends quickly go to the top).

• Personalization (most important criteria like key product attributes or customer segmentation like Customer Clusters)

3. Dynamic Facets

Boxalino developed an advanced configuration model for Dynamic Facets, including:

• Many visualization parameters (type of facet (enumeration, slider), order of the values, number of values shown before the “show-more”, etc.)

• Contextualization: dynamic logic to only show facets that are needed for a specific search or navigation context

• A/B Testing

• and much more

4. Neural Collaborative Filtering (NCF)

Boxalino personalization and contextualization algorithms (up-selling, cross-selling, …) are driven by state-of-the-art algorithms build in Google Tensorflow and based on Neural Collaborative Filtering (NCF). As described here NCF is supercharging collaborative filtering with neural networks and showed by far the best results we have seen so far.

5. Personalized Search Suggestions

Boxalino Winning Interactions Platform unifies the configuration of any product or content recommendations strategy for any use-case, search (and in this case Autocomplete-Search) included.

This means that all the capacities of the platform (including AI-based personalization strategies) can also be used in the context of Search Suggestions (including the above mentioned NCF).

To learn more about these functionalities of Boxalino’s Platform, please contact us!

6. Semantic Filters

A Semantic Filter is a logic that changes the search results, by either boosting a segment of the results to appear on the top or directly removing any results not matching the segment.

Semantic Filters can be defined by your team for important cases in Boxalino’s Intelligence Admin, but the most important, and largest quantity of, semantic filters are automatically generated by Boxalino’s standard Data Science processing.

This means that Boxalino analyzes the statistics of what your visitors do after they make a search and based on their scrollings, clicks, add to baskets and purchase behaviors, the system will identify the segments of results which should be promoted or even filtered.

7. Omnichannel Campaigns

In Boxalino Intelligence admin, you can define your Marketing Campaign in one place, and have them automatically:

1. Display campaign banners on your start pages (like the home page) with self-learning and personalization built-in

2. Generate an entire Landing Page automatically for your campaign without any additional manual step

3. Provide all the content automatically for your newsletter (and other outbound channels) without any content edition in the templates editors of your marketing systems.

8. Semantic Parsing

Boxalino Semantic Parsing covers all the standard features of a state-of-the-art search engine, including:

1. Stemming (removal of the endings of the words to avoid plural and other similar words to be considered as different)

2. Compound words tokenization (separation of the key words parts in composite words like “Kindermaske” => “Kinder” and “Maske”

3. Synonym self-learning (Boxalino is abale to learn automatically synonyms based on the user behaviors and the descriptions of your products)

4. Error corrections (typical spelling mistakes, but also unusual ones and possibly several errors per queries, all corrected while the user is typing)

9. Self-Learning Ranking

Boxalino’s platform analyzes continuously the behavior of each search visitor to draw conclusions about which products perform and which one does not, based on:

• Scrolling behaviors

• Pagination

• Click-Through

• Add to Basket

• Buy-Through

• (many other KPIs can be set as goals and will be also considered)

As a result, for each of the top 5K searches, Boxalino computes a ranking of the products based on the their Click-Through and Buy-Through performances and uses them to change the ordering of the products in the search results.

10. Real-time Results Rating

Boxalino’s can integrate real-time feed-backs on the list, including thumbs up/down (or similar) ratings the user can give directly on the search result list.

The page can then simply be refreshed and better results (according to similarities with the products which received a thumbs up/down) will be returned, in a fully personalized way.