Searches, Clicks and Checkout Conversions

When a customer uses the search functionality on your website, their interactions are sent to Klevu in order to influence subsequent search results based on current trends. We do this in three ways:

1. Searches: the customer searches for “shoes”.
2. Clicks: the customer clicks on a red shoe product.
3. Checkouts: the customer purchases the shoe (not necessarily the red one).

Klevu is able to tie these interactions together based on the IP address of the customer, allowing us to detect the whole process from search to conversion. This article will go into more detail on exactly what is happening during each of these steps.

Let’s assume your store has a Configurable Product with 3x Variants:

Product A (ID A)
- Red Variant (ID 1)
- Blue Variant (ID 2)
- Green Variant (ID 3)

1. Search (or Category view)

When the customer searches for “shoes”, a message is sent to Klevu which looks something like this:

https://stats.klevu.com/ ?ipAddress=123.456 &term=shoes

For category navigation pages the analytics is quite similar, but with the category name being used instead of the search term, for example for the Shoes category:

https://stats.klevu.com/ ?ipAddress=123.456 &category=Shoes
Note: These URLs have been shortened and simplified for illustrative purposes. Full details are available in our API Documentation.

2. Click

Let’s say the customer sees Product A appear in the results, specifically the red shoe (ID 1), and they click on it. Something like the following would be submitted to Klevu:

https://stats.klevu.com/ ?ipAddress=123.456 &type=click &term=shoes &product=A-1

Here we are able to determine that the customer searched for “shoes” and found something interesting enough for them to click on. This is useful information we can consume to help other customers find the products they are looking for, more quickly.

When processing a click, Klevu will find the parent of the product with ID: A-1, identify all of its variants (ie. red, green and blue) and increment all of their scores in the same manner. The red, green and blue products will then receive a boosting in subsequent search results, due to their increased popularity in the catalog.

3. Checkout

Now let’s say the customer views this product, but determines they would prefer to purchase the green variant. This is still considered a valid conversion as the customer has been guided from a search request through to a checkout conversion, so we capture this information with something like the following:

https://stats.klevu.com/ ?ipAddress=123.456 &type=checkout &product=A-3

When processing this conversion, Klevu will find the parent of the product with ID: A-3, identify all of its variants (ie. red, green and blue) and increment all of their popularity scores in the same manner. The red, green and blue products will then receive a boosting in subsequent search results, due to their increased popularity.

Which variant will appear in search results?

Since we are applying the same machine learning score to all variants of a compound product, you might wonder which of the variants (eg. red, green or blue) will appear when the customer makes a search or visits a category page.

For search results, the variant which appears will depend on the “relevancy score” associated with the search performed. For example, a search for “red shoe” would result in the red variant being displayed, since it is a better match than the green or blue.

In the case where the customer visits a category, or performs a more generic search like “shoe” where red, green and blue are all as relevant as one another, we have another factor for tie-breaking, detailed below.

Whilst it is true that all variants are allocated the same score for clicks and purchases, the actual variant which was purchased will receive an ever so slightly higher score. This means that in the case of a category page or generic search query, the variant which is most popular in terms of checkouts will be the one displayed to your customers.

Note: Once again these descriptions of our logic are for illustrative purposes only. The actual Klevu machine learning algorithms are too complicated to explain in a short article, so please treat these examples as they are intended, to give a layman’s view of what is happening.