Choosing your product attributes for product affinity segmentation

This article explains how to choose the right product attributes to use with product affinity segmentation for broadcast campaigns.

Not all product attributes will work well for product affinity campaigns.

Here's what to consider:

🚀 Actionable business case

We recommend choosing attributes that lend themselves to clear marketing actions or personalisation strategies, e.g. "Fragrances", "Skincare", etc.  

An Established affinity for fragrances segment suggests a clear marketing strategy, whereas "sky blue" might be a changeable preference. 

Attributes should also be relevant to your business, e.g. "Season" is a good way to organise the product catalogue, but as an attribute for product affinity it's not as useful because a seasonal collection likely contains lots of different attributes like autumn knitwear, autumn pyjamas, etc. 

Choose attributes that genuinely differentiate customer affinities so that the segments represent distinct and identifiable customer groups.

The attributes should also have some strategic importance or high business value.

We recommend using high level product groups, styles, price groups, sizes, etc. (if the data is in good condition).

📊 Data volume

To calculate stable affinity scores, the attributes need enough volume. 

This means they need to apply to a good number of products and those products should have sales. 

It's also important to have customers who haven't ordered products with this attribute, e.g. If every customer adds socks to the basket on checkout, there will be no significant uplift for "socks" when used in product affinity campaigns.

• If the attribute has very few products and low order numbers, not many customers will have an affinity to it, making the affinity score ineffective.
• If the attribute applies to all products except for a small number, the affinity score won't add much value. 

🔀 Overlap

Technically, choosing attributes that overlap with other attributes within the categorisation is not a problem - but it might suggest that the attribute doesn't represent distinct products.

Example:

When most products are unisex and have the attribute "male" AND the attribute female, the affinity to "female" isn't valuable, since it doesn't distinguish the customers from those with an affinity to "male".

🫧 Clean attributes

You might need to clean up your data before you can successfully use product affinity in broadcast campaigns. 

Example:

If a product group has multiple different spellings e.g. Socks, socks, sox, SockS, etc., or several attributes with the same name, this means that customers with an affinity for 'socks' end up split across many different segments.

This means that if you choose "Socks" but not "socks, sox, etc." when setting up a product affinity campaign, you'll only get a subset of customers with an affinity for socks.