Here are some frequently asked questions about Product affinity.
Can I input SKUs from different brands or categories for analysis, or must they belong to the same brand or category?
Yes, you can input SKUs from different brands or categories. The system takes all provided SKUs into consideration when analyzing affinity, regardless of whether they belong to the same brand or category or not. However, it's important to understand that the analysis focuses on customer behavior related to all the provided SKUs as a group, and does not guarantee that they have an interest or even purchased any of the individual SKUs presented. A scattered selection of SKUs from different product areas might therefore end up giving you strange results.
Can I expect that individuals in all groups have purchased the products?
Customers, regardless of the group, indeed may have purchased the product. However, the groups are created more based on how closely interested in it they are, as indicated by their behavior.
Is there any general benchmark for what constitutes a high affinity segment customer? If a search is done for a specific SKU, can it be assumed that this kind of customer has already purchased the product?
While it's likely that the person has purchased the product or one of the products indicated, it's really more about purchase patterns across the entire database. High affinity segment customers have the highest scores in that group.
What determines who is in a specific group (high afifnity, for example)? Is it the top 10% score, or is there a specific scoring threshold to meet or is it based on something else entirely?
The customers in a group is based on strict threshold values for what is considered to be "close" to a product. In other words, it's not a distribution based on top percentiles; instead, the groups are distributed based on the number of scores below/above certain threshold values in the proximity score.
Should low affinity segment customers be interpreted as those least interested?
These are the least likely individuals in the entire customer database to interact with the product.
How fresh is the data on which the model is based?
Freshness depends on when the model was last trained, and currently it's done every week. There must also be a relevant amount of purchases for a product to be included, meaning customers with different product histories must purchase a product to be included in the model. This means that new products may need time, until their sales volume approaches that of the other products, to become relevant for the model. The model compensates for "popularity", weighting bestsellers/popular products less heavily so they're not the sole strong signal (although they're often still important signals!).
What parameters form the basis for the three affinity segments? Just past purchases, or does it also include interactions, such as visiting one of the SKUs on the site (abandoned browse)?
Currently, product proximity is based only on purchase history—what the customers actually bought.
Do the three affinity segments - high, medium and low - cover the whole contact database?
Since groups are threshold-based, there are always customers in between the medium and low affinity groups, corresponding to people with no real signals towards the given products. This means that the suggested groups might not contain many customers, and the not-displayed group might in fact contain almost the whole contact database. So, short answer: no.
When I generate a product affinity group, add it to the Include in the composer, and use it directly or save it, will that product affinity group contain the same contacts forever?
No. Every time the AI recalculates the model (which is currently weekly) that particular group will be regenerated with same given products (SKUs), regardless whether it is part of a segmentation or a target audience.
If a new customer is created in Engage in between one AI-model affinity group calculation and the next, will the customer be immediately included in some product affinity group?
No. It can be included in groups only after the next model calculation (although it also might not end up in any of the groups).
If I save a segmentation with one of the product affinity building blocks, for certain SKU or SKUs, will the number of contacts in that segment differ a lot from week to week?
Yes, this can happen. With frequent recalculation of the AI-model, perhaps the SKUs used in the building block have come back into the assortment, or run out, or models have changed, or they've been on sale, leading to new purchases. These effect are felt not only for the SKUs listed but also one with closeness to the given SKUs. So the number in the group will very likely fluctuate.
I want to find a specific product, for example a jacket of a certain model, but I'm not interested in customer's preferences about its size or color variants, just the jacket brand and model. What SKUs should I enter to find people who are considered close to the jacket itself?
You'll have to enter all the SKUs, one for each variant, that's included for the product/item. It's not possible to use just a single variant to get the system to understand what you want.
Can I find people with affinity to a certain color or size? For instance, those people really interested in black garments (such as goths).
If in your product database (PIM system e.g.) you can fetch all SKUs, regardless of product, for a specific product meta attribute such as color, you can then take all those SKUs and do a product affinity search to get the group with the highest interest in that color, or size or whatever parameter that is available and commonly interpreted for a group of products in your product database.
When using the high affinity segment in a campaign I'd like to know which of these have already purchased the specific product, maybe to exclude them. How would I go about this?
As high affinity segment customers are not guaranteed to have purchased the product, you’ll need to use the specific article transactions filter mechanism found in all data criteria category. You can then logically combine the calculated high affinity group with the ones you are sure purchased the product in whatever way you want in your personalization and communication.
Can I be sure the low affinity segment customers have not bought the given product or products?
They have most probably not bought it. But even if they have, they have also bought so many things from other brands and categories that they can still be seen as low affinity.
Why are so many customers, almost half my customer base, returned in the high affinity segment for a specified SKU?
It depends mainly on how many purchases have come in for the specified SKU. The number of purchases can be linked to the fact that the product is new in the assortment. Or that the specified SKU is really just a new variant (e.g. color or size) of an existing product.
With too few buy signals, Product Affinity can sometimes deliver groups that are far too large. One way to avoid this is to enter all SKUs for the product, or combine them with other filtering (e.g. purchased in the same category, or based on the brand). Otherwise, it's a matter of waiting until more purchases come in for this particular SKU.
Another factor that can affect the result in this way is if there are frequently-bought products in the assortment that, for some reason, "look like" the specified SKU. For some merchants, for example, a "shipping" product can create some gaps in the model calculation. Consider removing that type of product in the integration with Voyado Engage.
Comments
0 comments