Karsa labelizer is a specialized service that solves the problem of product segmentation into individual campaigns according to their performance.
Thanks to machine learning, consistent campaigns are generated, which is difficult to achieve by manual segmentation according to strict rules and almost impossible to achieve when using more parameters/more dimensions for product segmentation.
The optimal campaign is one where Google can predict its behaviour. Therefore, if the number of impressions increases, the number of clicks on the promoted products will increase accordingly and the number of orders in the e-shop will also increase. It is also important that the conversion value is as stable as possible.
Google's algorithm has a simple job in the ideal scenario. If I display 1000 products I get 100 clicks, 100 clicks will be 1 order, each order will have a conversion value of 500. Google can then react precisely to an increase in budget or a change in target ROAS.
Google cannot predict the behaviour of the products in the campaign and selects only those products that are predictable for them, but this selection process costs money because the user has to click on the product. Two scenarios follow. The ROAS target is not reached or the campaign does not spend the set budget, because Google focuses only on a few products that are predictable. It cannot increase sales volumes and does not take market share.
Products are distributed into campaigns using machine learning models, where the consistency of product values is optimized to make the campaign as predictable as possible.
Each created campaign meets the set criteria of minimum number of orders. Meeting the minimum number of transactions/orders is essential for Google to evaluate the campaign. It is often the situation with manual product distribution that these criteria are difficult to achieve, Karsa solves this automatically.
With Cluster designer, you can easily create campaign distribution designs. You can compare them against each other using their consistency and choose the one that best suits your business.
The new products have no history and would be placed in a non-performing campaign, which usually has a limited budget. The products would not be able to prove their performance. In the configuration it is possible to set the derivation of values from related products and the product is placed in the potential best cluster. The second option is to insert products into a campaign dedicated for this purpose, where the products are moved after 30 days from this campaign according to their performance.
The best number of campaigns is found according to the product data structure and the parameters used for clustering. This is essential because each store has a different value distribution and product structure and it is not efficient to use a static template to divide into, for example, 4 campaigns. It is possible to create 20-50 campaigns for large stores and a lower number is more efficient for smaller stores.
Products change their performance over time and migrate between campaigns. This is a required functionality because it improves performance. The downfall is the transfer of large numbers of products, when the values change by leaps and the campaigns stop being predictable. Google starts to relearn and the campaign loses performance. In Karsa, you can set limits on both the percentage of product that can migrate and the % of conversion value that can change in a campaign over time. This helps to increase stability.
Every day, new data from Google Ads is automatically downloaded, calculating the product placement in campaigns and transferring the information to GMC.