Skip to main content
Behavioral Affinities

Learn more about Nosto's Segmentation & Insights

Lari Lehtonen avatar
Written by Lari Lehtonen
Updated over a week ago

At the core of Nosto’s segmentation tools are affinities and affinity settings. In this article you’ll learn why you can and should use them, what you can achieve with them, and the basics of the principle how they work. 

Vast majority of buying intent and affinity signals are created by users who simply browse a site without buying or even adding anything to the shopping cart. Majority of customer platforms and different customer segmentation tools rely only on transactional data, thus these often miss a lot of invaluable information about shopping preferences, as these can be learnt only after a purchase is made. In other words, many opportunities to improve shopping experiences are missed, which is especially a problem when new visitors without any buying data visit the site the first time, but also when existing customers who have bought something change their scope and want to shop something else than they did before. 

Nosto's affinity based segments help you to solve both of the example cases and many more. Affinity segments link products, product ranges and shoppers together with the highest degree of relevancy, updating on real-time, adjusting on each store's unique browsing and shopping patterns, while segment sensitivity settings are the mechanism to control the degree of personalization and volume of users belonging to each segment.

Value of Using Nosto's Dynamically Adjusting Affinity Segments Instead of Traditional Rule Based Segments

Vast majority of systems include manual, rule based tools to create segments, which support creating segments based on pre-defined settings such as: Shoppers who viewed three or more products from sneakers category have affinity towards 'sneakers'.

This is arguably suitable for analytics, but in context of e-commerce, product ranges and availability change all the time, while changes on the website can easily render rule based segments obsolete. 

For instance, if store has 100 pairs of Adidas sneakers, but temporarily the selection is reduced to only 10 pairs, it's likely that shoppers will browse less. On the contrary, if thanks to conversion optimization efforts such as excellent use of recommendations, time-on-site and number of browsed products increase, it's possible that many shoppers would browse more Adidas sneakers than before. In both scenarios, rule based segments would require adjustments, which is always manual work and foremost would be undetected. 

The other benefit of using Nosto's dynamic segments is simply based on time & performance, as creating rules potentially separately for each category and trying to figure out what different levels of affinity mean, require a noticeable effort in web-analytics plus overall inventory management skills and understanding. 

To solve these problems, Nosto's affinities adjust automatically to changed shopping & browsing patterns and are based on real-time updating scoring. In order to support different merchandising and personalization scenarios and use cases e-commerce merchants have, Nosto features segment controls called Affinity Sensitivity.

How Does Affinity Sensitivity Work And How To Use It?

Affinity sensitivity effectively works as a control mechanism and campaign configuration by adjusting the strength of  segments affinity and consequently reach or volume of users belonging into each. When creating a segment you can choose from three different sensitivity settings: Top, High and Low

  • Low produces the highest reach, but also includes affinities towards ranges that might've been considered, but where the link is not as strong as in Top and High. On majority of the websites, where average amount of products viewed per visit is around 5-7, viewing 2-3 products in a product range will likely add the user to the low sensitivity affinity segment.

  • Top will include shoppers who have a high affinity and also include the ones with a strongest preference for a given product range. For example, a shopper who is balanced between three different brands - say, Nike, Reebok & Adidas - but who starts interacting noticeably more with Adidas is eligible for an Adidas Brand Affinity segment set with a Top sensitivity rule. In other words, he doesn't have only high preference yet for Adidas, but it is also his most preferred brand. In order to belong into top sensitivity affinity segment, a shopper needs to have multiple views [of a certain product range] across multiple visits and likely add a product to the shopping cart or buy it as their sole purchase. Following the example, a shopper can only belong to one top brand affinity at a time, but can have top affinity based on different rule such as category (sneakers) or product attribute such as color (black). 

  • High will include the strongest affinities instead of only the most favorite one (top), but by disregarding the feeble ones (low). This narrows down the reach, but also increases the quality of the audience to only include the biggest fans. Following the same example as above, it would take some time for the shopper to be included in an Adidas Brand Top Affinity segment, but by browsing multiple Adidas products and by adding an Adidas product to the cart or buying it, will in most cases add a user to the high affinity segment.

How Does The Scoring System Work?

The affinity system is based on scoring, indicating the strength of affinity between a shopper and a range of products. A product range is based on a product property such as product category, brand, or a custom product attribute such as a colour or a product feature to name two examples. In brief, shopper's strong affinity indicated as high affinity score towards a product range suggests a close relationship, undoubtedly increasing likeliness of a purchase conversion if the on-site experience is personalized by exposing more the preferred product ranges.  

The scoring is individually calculated for each shopper and based on all interactions a shopper does while visiting the site, re-calculated in real-time, across multiple sessions and taking time into equation. Below an example of a few self-explanatory interactions specific for e-commerce websites which are used for scoring, starting from feeble intent signals towards very strong affinity:

  • Product listing view (weakest)

  • Product page view (weak)

  • Adding a product to a shopping cart (strong)

  • Buying a product (strongest) 

Because website design, product types, buying cycles and in general businesses also affect the number of interactions per visit and interaction types, the scoring is self-learning, not static or limited only to given interaction examples as other interactions weigh in too. To avoid possible machine learning errors caused commonly by anomalies, some global safety rules are in place and regardless of site’s parameters. For example, an interaction of adding a product to a cart is always stronger signal than simply opening up a product page in a browser. As another example, opening the same product three times across three different visits to the site produces a stronger affinity score than eying out the product twice during the same session, while interactions done months ago are less impactful than those made just a few days ago.

An outcome of the automated process could be visualized as a network of links how a shopper is connected to product ranges, including the strength of the connection: An affinity score.

Note: if you are using Low as a sensitivity rule, we recommend you to take advantage of our Segment Prioritization tool to prioritize one experience over another.

Did this answer your question?