A machine learning algorithm takes several parameters into account when establishing what a customer may like and eventually buy on the store. On this page, we have tried to simplify the personalization process and tried to answer a few important questions.
What is Personalisation?
When a shopper enters an online shop, he or she usually browses through a catalog to find a relevant product of his or her choice. Many shoppers prefer to use search if they know exactly what they are looking for. If the customer is using search and has clicked on a few products, in most cases this is a good indicator that he or she may have spotted something of interest.
Personalization is all about identifying where the customer's interest is and highlighting those products that are in alignment with the customer's preferences.
Who can enable Personalization?
You can enable personalization on your store If you are
- using Klevu JS Library
- using a custom integration using our API
How does it work?
As part of the semantic analysis, a relationship between the query and the products either visited by the customer or found to be relevant from the other customers' activities is established. Only those products that are relevant to the current search query are picked up and analyzed to identify common factors within them. It is possible that a few features are found as common (e.g. their brand, color) and others have some or no similarly. But this commonness is used as boosting weights to boost certain types of products higher up in the search or category page results.
Does it have any impact on relevancy or manual promotions?
It is important to mention that personalization is not about filtering products away. It is about boosting one or more types of products that are found to be in better alignment with the customer’s preferences. However, within what we find relevant to the customer’s preferences, it is also important that we boost products that are promoted by you or what our self-learning module thinks are in-trend on the website. Below, we have highlighted the different merchandising features and how do they get affected by enabling personalization:
Keyword specific product promotions (guide)
Keyword specific promotions are preserved and the products promoted to be shown at the top are shown at the top of the search results. In this case, personalization is applied only to the rest of the products.
Keyword specific product exclusions (guide)
Products excluded for specific keywords are excluded from the search results, even though they may look very relevant to the customer’s preferences.
Individual and rule-based product promotions (guide)
Here, there are no hard rules but in general, if the products boosted manually are matching customers’ preferences, they are boosted even higher. For example, let us assume that
- For a query, some products from brands A, B, C, and D are relevant.
- As a merchant, you have equally promoted products of brand A and brand B.
- Thus, without personalization, products relevant to the query that are from brands A and B are going to be promoted and shown towards the top of the search results.
- If we realize that the customer is interested in brand B (more than the brand A), the personalization engine will boost products of brand B higher up.
- If we notice that brand C is what the customer is interested in, depending on the boosting scores of brand A and B products, it may happen that they will see products of brand C appearing higher up but also maybe together with the products of brand A and brand B.
Category-specific product promotion (guide)
These are the hero products chosen by you to be displayed at the top of the category and they are shown at the top as configured. After the top products have been displayed, personalization will be applied to the rest of the products.
Category-specific rule-based promotions (guide)
Similar to how personalization impacts results on search results pages, here too, there are no hard rules and in general, if the manually boosted products are matching customers’ preferences, they will be boosted even higher. The same example explained earlier is true here as well.
Is it possible to specify what attributes and attribute values to use for personalization?
Here, the value should be provided in the following format:
- attribute key of an existing attribute that has been indexed as a facet (the facet doesn't need to be visible)
- a valid value for this attribute
- some weight (between the range of 1 to 99). Any negative score is discarded and a score > 99 is brought back to 99.
- you can specify more than one attribute-value-weight separated by a double semicolon.