This AI-powered tactic enables targeting based on your customers' relationship with a certain product or products.

When you use the Product affinity tactic, advanced machine learning models will examine your customers' behavior. Based on their purchases, it will present building blocks you can use to generate customer groups with different closeness or affinity to the specified product/products, product category or brand.

The results are approximations and come with recommendations on how to use them.

How it works

Approach A - Using SKUs for specific products and variants:

1. Go to Targeting tool, select Products on the side menu and then Product affinity.
2. From your product database, get the SKUs of the products you are interested in (one or several).
3. Enter these SKUs in the field (separated by commas). Now hit Generate.
4. Now you will be presented with the product affinity score graph. Adjust the slider to configure your desired segment size and add to the composer using the Include, Exclude, and Restrict buttons.

Approach B - Using brands and categories

Instead of targeting customers based on their interest in a specific product or brand by entering SKUs, you can instead configure Engage to allow you to target groups of customers using drop-downs. The fields available depends on your setup but is usually product metadata like brand, category and sub category, the same as those available in the “Specific article” filtering tool.

You can also combine these fields along with SKUs as shown in Approach A.

For example, imagine entering two options, “boxer shorts” and “briefs”, in the "Categories" field. Engage will examine all your contacts for their affinity to those two categories, building up some imaginary “underwear” category. This is functionally the same as adding all the underlying SKUs in a request to the product affinity model. This is therefore an OR statement, or a union in set theory, creating a group larger than the two groups (“boxer shorts” and “briefs”) used to create it.

Imagine instead if you added to the "Brands" the values "Nike" and "Adidas". Then you also selected "sneakers" in the "Categories" dropdown. This queries the model for those SKUs that are sneakers but only from Nike or Adidas. This corresponds to an AND statement (intersection in set theory) and results in a smaller group than the groups used to create it.

Reach out to your Voyado representative if you need more fields.

Product affinity score graph 

Recent improvements make the experience more transparent, flexible, and actionable by replacing the previous fixed segments (Fans, Lookalikes, Detractors) with a fully configurable graph-based interface.

The graph shows the distribution of all customers based on their affinity scores.
You can use a slider to:

• Define a narrow, high-affinity segment (top scorers only)
• Create a broader audience by lowering the threshold
• Preview how many contacts match your settings

This gives you precision and flexibility when building segments tailored to specific campaign goals.  

Example use cases

• Promote a specific product or category to the most relevant customers
• Limit an email campaign to your most engaged contacts
• Avoid over-communication by excluding low-affinity contacts
• Run A/B tests with different affinity thresholds

To learn more about working with Product Affinity, see the dedicated FAQ.

Using and interpreting the Score Graph

Below are some generalised examples of the histogram score graph output, and possible ways of interpreting them. Please be mindful that your product catalogue and purchase history is the baseline for the  results. The general examples below might not represent the underlying reasons for your specific outcome. 

Scenario 1: Majority of contacts around high affinity

In this scenario a majority of your contacts have high affinity to your queried products.

If your graph looks like the one above with a lot of contacts' scores distributed to the right and a high average affinity it could be likely to the following reasons:

• Your query is of very popular products, linking together many contacts' purchases to the queried products
• Your query includes "super categories", "super brands" or even "super products". These categories, brands or specific products could be very popular as mentioned above. The brand or category could also be broad and include many common products 
• Your product catalog is relatively small, meaning that a lot of contacts has purchased relatively many products in your catalog, linking together your query to a lot of contacts
• If your query relates to a low volume product and your result looks like above you can investigate your queried products compared to the contacts having purchased it. For example, an edge case could be a new or low volume product which is purchased by a highly active group of contacts. These contacts are then linking the queried products to a lot of products from their purchase history, and subsequently connecting your queried product to numerous similar contacts with high affinity

Scenario 2: Majority of contacts around middle - zero affinity

Most contacts are neither close nor far (high nor low affinity) to your queried products.

If your graph looks like the one above, a lot of contacts' scores distributed around 0 affinity, it could be likely to the following reasons:

• Your queried products could be relatively unknown to the model, it could be new or relatively low frequency purchases products. This results in the model having a hard time identifying contacts with a strong high or low affinity to the query
• In contrast to the statement above, your queried products could also be very popular products, but your regular purchase pattern "splits" your contacts into clusters around your products. The queried products thus acts as a bridge between contact clusters, having no strong relationship to any cluster
• An edge case could be "cold contacts", meaning that a majority of contacts have few purchases, thus not creating clear patterns in affinity. This should mitigate itself over time as more purchase history is incorporated in the model. This is a common pattern in "young" environments likely without having imported purchase history

Scenario 3: Majority of contacts around low affinity

A majority of contacts' scores are gravitating to strong negative affinity.

A histogram graph showing a strong separation between the queried products and the majority of the contacts could be due to the following reasons:

• The queried products could be bought a low amount of times and be a niche cluster of products
• The contacts who would score high to the queried product is a disjoint community or subset of your contact base, not representative by the majority
• Your queried products could be old products, with few relationships to current and active products

Scenario 4: Less spiky distributions of contacts

Contacts are distributed more smoothly, covering large parts of the scale 

If your graphs is similar to the smooth, more evenly distributed graph shown above, it could be related to the factors posted below:

• Your query contains "contradicting" products, giving every contact some affinity but no strong signals in either end for a majority
• Your query contains weakly connected or correlated products
• Seasonality could be at play; your queried products are highly seasonal, resulting in few of your non-seasonal products being strongly connected, but contacts are still connected through other purchases, and thus contacts are spread out around the full spectra

Scenario 5: Two or more peaks of contacts

Contacts are distributed into two or more distinct peaks or clusters

If your graphs contains two or more distinct peaks or clusters, it could be due to the reasons stated below:

• A strongly connected (high affinity) cluster and weakly connected cluster is present in your contact base. Your queried products have distinct relations to different clusters amongst your contacts. For example, your contacts could be divided into communities with different purchase patterns. Such as premium vs budget, or discount product chasers 
• You could have queried a combination of products where each product itself have distinct peaks, and the combination of products further strengthens the peaks
• Your query contains substitutes or mutually exclusive products, resulting that come contacts have a strong affinity, whilst another has a distinct low affinity (buying "other brand" or substituting the queried products)
• A last scenario could be that you have queried "stale" or non-recent products, resulting in "stale" or old contacts have a high affinity, while newer purchasers have a lower affinity (low relationship to the old products)

Article last reviewed 12 November 2025 13:13