Drive Product Discovery with KIBO's Advanced AI Search

Transcript

KIBO offers the industry’s leading search solution built into the ecommerce platform powered by solar. As a true microservices based platform, KIBO offers our on-site personalized search in three ways, as a standalone search solution, part of our personalization offering, and part of our ecommerce offering, all in near real time.

This capability isn’t limited to just search.

KIBO has near real time indexing for our catalog, inventory, pricing, and promotions to power the omnichannel experience.

This means that KIBO ecommerce customers are not getting a serviceable search solution that is a forgotten part of the platform.

Rather, they are getting a search solution that is a stand alone, goes toe to toe with other search and personalization vendors. They get out of the box robust merchandising and ML capabilities native in their on-site search.

This helps KIBO customers drive conversion, margin, and revenue.

So let’s jump in. The first search tool we’ll examine is keyword search. Input settings for this search define the query term and what to look for in that term, such as synonymous expression, multi word matching, bracelet, and more. A great example of this is when a shopper searches for a knapsack.

You may also wanna return results for a backpack or when a shopper searches for living room furniture. We may wanna configure how much emphasis we put on the individual words living, room, and furniture, or how much emphasis we put on the phrase as a whole. Inside of the main tab lives the search controls.

Synonyms can be managed on a per site basis. We can see here that we have a one way synonym set up for backpack so that when a user types in the search term knapsack or pack, we can show them the same results they would get with searching for backpack.

We could also create a two way synonym for shoes and sneakers where a search for one returns that and all the others.

Search term redirects allows merchandisers to redirect a user to a specific URL based on a specific search term. Search redirects apply to site search only. The user will be redirected to the new URL when they press enter or click the search button after typing in a search term. Inside of the search configuration, merchants can manage settings and campaigns across their site.

KIBO enables users to manage on-site search, category, and product suggestions, also known as type ahead search, and product listing pages. A user can enable or disable product slicing for search results. They can manage win match controls and phrase slop, as well as enable out of the box spell correction for autocorrect and did you mean. Marketers can also manage the field weighting of the search engine.

Field weights determine which product field and attributes to look at for that term and what weight to put on each field.

Ultimately, this is used to determine the relevancy.

Vector search is a search method that doesn’t rely on matching words to a query.

After using machine learning on a large dataset, documents are associated with vectors, and then the search query checks those vectors to find the most similar results.

This provides the benefit of retrieving results based on the meaning of the search query rather than simply matching words or phrases, also referred to as semantic search.

When vector search is enabled, it will allow you to select a search mode while configuring site search.

Standard is the basic keyword search.

Hybrid combines keyword and vector search techniques to return results.

There are two types of hybrids that you can use.

Blend mode brings vector results over distance threshold to the same score scale as the top result from a standard search. This is the default hybrid mode.

Boost mode will only boost results that appear in both a standard and vector search. Vector results are first multiplied by a set multiplier if the distance threshold is met.

Lastly, there’s vector only, where keyword search techniques aren’t used to return results.

Let’s take a closer look at standard search results versus vector search. If I go back to my search configurations main page, I’ll make sure my non vector configuration is the default. And if I click on it and then site search, you can see it’s off. Heading to my storefront, I’ll search the phrase ‘hat with SPF protection’, and you can see it doesn’t return anything.

Now if I go back and make my vector search configuration default, then refresh the page, now you can see it returns products.

The most interesting thing about this hat, if I click on it, is nowhere on this product is the term SPF listed. The AI knows that SPF and UPF are related, with SPF typically referring to sunscreen and UPF referring to clothing. So it’s assisting the shopper by understanding the underlying need.

Next, let’s look at boosting and burying where marketers can rerank the order of the results by putting a stronger or weaker emphasis on specific attributes.

Inside of the search settings, I can scroll down to the boost and bury section. I can add an expression based on any core or custom attribute that I have set up. I could boost based on things such as product margin or ratings of the product.

I’m going to go ahead and boost any products that have a rating greater than or equal to four, and I’m going to give them a boost of a hundred so that you see the impact of this.

I’ve added the boost, but I haven’t saved it yet, so let me go to my storefront and search. I’m going to search for pack. We can see the results returned, putting the storm front pack and elevator pack in the first and second slot.

Now I’m gonna go ahead and save this boost and research the exact same query.

If I again search pack, we will see that the elevator pack has now moved into the one slot, and the aether pack has moved into the two slot. If we go into the products themselves, we will see that the elevator pack has a product rating of five, the aither pack has a product rating of four, and the storm front pack only has a product rating of two.

KIBO Search can provide a more granular level of merchandising capabilities with search merchandising rules. These can be created for individual products in an overall site search or within categories on the site.

For example, maybe I wanna quickly configure the way my tents category displays products, pushing my summer tents to the top.

I could create a merchandising rule for the category, preview what the results are going to be, and set up a condition for a booster berry, like what we saw in the search configuration.

We’ll boost the season equal to summer by ten, and then re preview the result.

As we can see, some of the products that have summer search attribute have moved up in our results.

We can also visually move or rank these products and pin them in place.

Then when the front end calls our get search results or get category API, we’re going to return them back in this order.

Search will return them in the appropriate order.

We can see the actual scoring, how we get to that score.

If we’re making boost and barrier, any other search changes will expose the relevancy in the scoring to the merchandiser so adjustments can be made in real time.

The KIBO search solution is fully equipped with robust merchandising capabilities and an easy to use UI.

KIBO’s native ecommerce search is not just a box to check, but a leading personalized search platform made readily available to any ecommerce customer.