Welcome to the world of advanced Google Shopping campaign optimization! Karsa Labelizer is a specialized tool designed to help PPC specialists and marketing managers significantly improve the performance of their Product Listing Ads (PLA) through intelligent, multi-dimensional product segmentation powered by artificial intelligence (AI).

What Problems Does Karsa Labelizer Solve?

Traditional Google Shopping campaign management often faces several key challenges:

• Inconsistent campaigns: Products with vastly different performance metrics (e.g., ROAS, conversion rate, order value) are often grouped together. This makes it difficult for Google's algorithms to effectively predict and optimize their performance.

• Difficult ROAS prediction: Due to inconsistency, it's challenging for Google to estimate the actual return on ad spend (ROAS) that can be expected from a campaign, leading to suboptimal budget allocation.

• Time-consuming manual segmentation: Creating and maintaining a granular yet effective campaign structure manually is extremely time-consuming and nearly impossible for larger e-commerce stores, especially if you want to consider multiple performance metrics simultaneously.

• Overcoming standard segmentation limitations: While common practice often relies on segmentation based on a single parameter (e.g., ROAS) into a predetermined, small number of clusters, such an approach has its limitations. It may not capture all the nuances in performance and might overlook hidden potential or risky segments.

How Does Karsa Labelizer Help You?

Karsa Labelizer approaches product segmentation innovatively:

1. Intelligent segmentation (clusterization): It uses advanced machine learning algorithms to analyze your products' performance data across several dimensions (metrics) simultaneously. Based on this, it automatically identifies and groups products into optimal clusters.

2. Increasing campaign consistency and predictability: The goal of clusterization is to create campaigns (individual clusters) that are internally as consistent as possible in terms of performance. This means that products in one cluster behave similarly. As a result, Google Ads algorithms can much better predict their future performance and more efficiently manage bids and budgets.

3. Campaign number optimization: The tool can suggest the optimal number of campaigns (clusters) specifically for your e-commerce store and your data distribution, instead of fixed, predefined structures.

Who is Karsa Labelizer For?

The tool is primarily designed for:

• PPC specialists: Who are looking for advanced methods to optimize Google Shopping campaigns and want to have detailed control over structure and performance.

• Marketing managers: Who need to ensure maximum efficiency of advertising investments and are looking for a reliable, data-driven solution for their e-commerce clients or their own e-commerce store.

• E-commerce stores of all sizes: From medium to large, who want to fully leverage the potential of their product data and achieve better results in Google Shopping.

Karsa Labelizer is therefore designed to provide you with the necessary tools and data-driven intelligence to effectively structure your Google Shopping campaigns with the goal of achieving higher performance and better predictability.

Why This Documentation?

It provides you with understanding of principles, detailed guides for configuration, strategic recommendations, and practical tips for analysis and optimization.

Key Documentation Sections:

• : What Karsa Labelizer is and why segmentation matters.

• : Interface, service lifecycle, quick start.

• : Design and configuration of clusterization strategies.

Recommendations for Getting Started:

We wish you great success with Karsa Labelizer!

In today's highly competitive digital advertising environment, especially in e-commerce, basic campaign structures in Google Ads often fall short of achieving optimal performance. If you manage an e-commerce store with a diverse product catalog, you've likely noticed that the overall campaign performance view (such as average return on ad spend - ROAS) may seem satisfactory, but often masks significant inefficiencies and losses in specific product groups. This is where advanced product segmentation comes into play.

Limitations of Traditional Campaign Structures

Without detailed segmentation, your campaigns may suffer from several shortcomings:

• One-size-fits-all approach to diverse products: All products are managed by the same or similar rules and goals, despite their performance, margins, sales cycle, or seasonality potentially varying dramatically.

• Products with inefficient budget utilization: Some products may generate many clicks (and thus costs), but low conversion value or low ROAS, reducing the overall campaign efficiency. Without segmentation, it's difficult to identify these products and limit their promotion.

• Untapped potential of "star" products: Conversely, products with high ROAS and high conversion value may not receive sufficient space and budget to fully demonstrate their potential.

• Difficult bid optimization: Setting optimal bids (whether manual or automatic targets like tROAS) is complicated if the campaign contains products with very different profitability or conversion rates.

The Importance of Predictability and Consistency for Google AI

Modern Google Ads campaigns, especially those using automated bidding strategies (Smart Bidding) such as Target ROAS (tROAS) or Performance Max campaigns, rely heavily on Google's artificial intelligence (AI). For these algorithms to work as efficiently as possible, they need quality, structured, and above all, predictable data.

• Campaign predictability: Means that Google can reliably forecast how the campaign will behave – what will be the click-through rate (CTR), conversion rate, average order value, etc., if, for example, the budget or bid changes.

• Campaign consistency: If products in a campaign are grouped so that they have similar performance characteristics (e.g., similar ROAS, similar conversion value), the campaign becomes internally consistent.

How Does Karsa Labelizer Help Google AI? Karsa Labelizer creates highly consistent product clusters using AI. By "cleaning" campaigns of products with significantly different behaviors, it greatly increases their predictability. Google AI can then:

• More accurately estimate future performance: Leading to better bid setting.

• More efficiently allocate budget: Direct resources where there is the highest probability of achieving goals.

• Learn and adapt faster: More stable input data means faster and more efficient learning processes for Smart Bidding.

Benefits of Advanced Segmentation with Karsa Labelizer

Strategic product segmentation, which Karsa Labelizer automates and optimizes, brings a number of specific benefits:

1. Optimized budget allocation: Allows you to invest more in segments (clusters) with products that show the best results (e.g., high ROAS, high conv. value), and conversely, limit spending on segments with low performance or loss-making products.

2. More precise and efficient bidding strategies: Different product segments require different bidding strategies:

   • Best-performing products ("Bestsellers"): You can apply more assertive bids to maximize their visibility and sales volume.

The Role of Segmentation in the Performance Max (PMax) Era

While Performance Max campaigns consolidate various Google advertising channels under one roof, they still strongly benefit from strategic product segmentation through signals from the product feed and targeting within asset groups. Quality data in the feed, enriched with meaningful Custom Labels (which Karsa Labelizer generates), are even more critical for properly directing Google's AI in PMax campaigns.

By creating thematically and performance-consistent product groups using Karsa Labelizer, you can more effectively structure your PMax campaigns and better align asset groups with specific product segments, leading to more relevant customer targeting and better results.

Using advanced segmentation through Karsa Labelizer thus goes beyond merely organizing products; it's a methodical approach to streamlining advertising investments and supporting growth goals in e-commerce.

Operation Manager is an important component of Karsa Labelizer that provides you with an overview of the status and progress of key operations performed by the system in the background. Its main use for you as a user will primarily be in monitoring the process of creating new clusterizations.

What is Operation Manager For?

• Monitoring operation status: The main function of Operation Manager is to display the current status of long-running operations, typically the clusterization process that you launched from .

• Process transparency: It gives you confidence that the system is working and allows you to estimate when the operation will be completed.

• Diagnostics (if needed): If there were to be an issue with completing an operation, information from Operation Manager can be useful for our support team in troubleshooting.

How Does Operation Manager Work in the Context of Clusterization?

1. Launching clusterization: After you configure and launch the creation of a new clusterization in , Karsa Labelizer initiates a complex computational task in the background.

2. Data export to GMC (manual and automatic): For clusterizations in 'Run' status, Karsa Labelizer ensures automatic daily export of the supplementary XML feed to Google Merchant Center (GMC). This feed is generated to the URL set in the 'Setup' section. Additionally, Operation Manager may offer the option of manually triggering an export. This is useful if, for example, you've just switched a new clusterization to production status and want to generate an XML feed immediately without waiting for the regular automatic update.

Where Can I Find Operation Manager?

You can find Operation Manager in the 'Tools' tab under the 'Operation manager' item.

For most users, Operation Manager will primarily serve as an information panel to confirm that a launched clusterization is proceeding as expected. Detailed technical logs are usually not displayed here, but it provides peace of mind and an overview of background processes.

Welcome to the key subsection of your Cluster Designer settings – the selection of parameters for clusterization. The right choice of metrics by which Karsa Labelizer will segment your products is crucial for creating meaningful, consistent, and high-performing clusters (future campaigns).

Why is Parameter Selection So Important?

The parameters you select directly influence how Karsa Labelizer's AI algorithm "understands" similarities and differences between your products. Based on these parameters, products are grouped into clusters.

Well-chosen parameters will help you:

• Create logical and strategically relevant clusters: Products in one cluster will share similar performance characteristics, making targeting and bid management easier.

• Maximize campaign consistency and predictability: Well-defined clusters are more readable for Google Ads algorithms and allow them to optimize performance more efficiently.

• Uncover hidden patterns in product performance: Multi-dimensional analysis can identify groups of products that would remain unrecognized when using only one parameter.

• Allocate budget more efficiently: More precise segmentation allows better directing investments to the most promising areas of your portfolio.

• Better understand your product portfolio: Analysis of the resulting clusters will provide valuable insight into how different groups of products behave in the advertising ecosystem.

What You'll Find in This Subsection

The following pages will guide you through the process of selecting optimal parameters for your clusterization:

• : Detailed description of all metrics you can use for clusterization in Karsa Labelizer (e.g., ROAS, Conversion value, Clicks, Product price, etc.), including explanations of what they mean and where they come from.

• : Practical guides and examples of various parameter combinations. You'll learn what types of clusters typically emerge when using certain combinations and what goals these strategies are suitable for.

• : Conceptual explanation of how Karsa Labelizer's machine learning algorithms work with your chosen parameters to find the optimal distribution of products and the optimal number of clusters.

Give parameter selection sufficient attention – it's an investment that will pay off in the form of more powerful and better manageable Google Shopping campaigns.

To effectively work with the Karsa Labelizer tool and understand the principles of advanced product segmentation, it's important to familiarize yourself with several key concepts. This glossary provides a brief and clear explanation of them.

Cluster : A group of products that, based on analysis of their performance metrics (using AI in Karsa Labelizer), have been identified as similar and are therefore grouped together. Each cluster in Karsa Labelizer typically corresponds to one future Google Shopping campaign or a specific segment in a PMax campaign.

Clusterization (Clustering) : The process of automatically dividing products into meaningful groups (clusters) based on their similarity within chosen parameters (metrics). Karsa Labelizer's goal is to create highly consistent and predictable clusters.

Cluster Designer : A tool (module) within Karsa Labelizer that allows users to define criteria and parameters for creating a new product clusterization. This is where limits are set, metrics are selected, and behavior for new products or movements between clusters is defined.

Custom Label

Karsa Labelizer offers you a wide range of performance metrics that you can use as parameters for clustering your products. The right selection and combination of these parameters is the foundation for creating meaningful and effective clusters. Below you will find an overview of available parameters and an explanation of their significance.

All the metrics listed below typically relate to the performance of products in Google Shopping campaigns over the selected period (by default the last 30 days, as set in [Setting Limits and Goals](../setting-limits-and-goals.md)).

1. Clicks

After carefully configuring all parameters of your new clusterization in the Cluster Designer – from , through , , , to – you are ready to start the actual process of creating clusters.

1. Launching the Clusterization Process

• Final check:

You have designed, configured, and a clusterization that you are satisfied with. Congratulations! Now comes the key phase: deploying this new structure to your real Google Ads campaigns and its subsequent strategic management and optimization.

From Theory to Practice

In previous sections, you've learned:

Your products' performance is not static – it changes over time due to seasonality, inventory levels, price changes, competitor activities, and many other factors. For your segmentation in Karsa Labelizer to remain effective and current, it's necessary not only to divide products correctly at the beginning but also to actively manage their dynamics.

What Does "Product Dynamics" Mean?

In the context of Karsa Labelizer, product dynamics primarily refers to two key areas:

One of the key advantages of Karsa Labelizer is the ability to dynamically adjust the placement of products into clusters based on their current performance. This means that if a product's performance changes in such a way that it better matches the characteristics of another cluster, the system can automatically move it. However, to prevent these movements from leading to undesirable destabilization of your campaigns and disrupting the learning process of Google algorithms, Karsa Labelizer offers detailed settings of limits for these operations.

Enabling Product Movements

First, it is necessary to decide whether you want to allow automatic product movements for the given clusterization at all.

1. Allow moving products when updating

• What it is: A basic switch that enables or disables automatic product movements between clusters when Karsa Labelizer updates data daily and recalculates optimal placement.

• Recommendations:

  • Enabled (Yes): In most cases, it is desirable to have this option active so that your segmentation is continuously optimized and reflects current performance.

Setting Limits for Product Movement

If movements are allowed, the following limits will help you control their intensity and prevent sudden, extensive changes that could negatively affect the optimization of Google campaigns.

1. Each cluster can lose at most % of products

• What it is: Defines the maximum percentage of products (out of the total number of products in a given cluster) that can be removed from this cluster (moved elsewhere) during one update cycle (usually daily).

• Example: If you set 10% and the cluster has 100 products, a maximum of 10 products can be moved away from it on a given day.

• Importance: Prevents massive "outflow" of products from one cluster at once.

2. Each cluster can gain at most % of products

• What it is: Defines the maximum percentage of products (relative to the original number of products in a given cluster) that can be added to this cluster (moved from other clusters) during one update cycle.

• Example: If you set 15% and the cluster originally had 100 products, a maximum of 15 new products can be moved into it on a given day.

• Importance: Prevents sudden massive "inflow" of products into one cluster, which could also affect its characteristics and behavior.

3. Waiting days before moving product again

• What it is: The minimum number of days that must elapse since the last movement of a product before this specific product can be moved to another cluster again.

• Example: If you set 7 days, a product that was moved today cannot be moved again for the next 7 days, even if its performance would suggest suitability for another cluster. It must first "settle" and demonstrate performance in the new cluster.

• Importance: Prevents constant "pouring" of products between clusters, gives products time to acclimatize and collect relevant data in the new environment.

4. Limit (in percent) how much can be stabilization value changed

• What it is: The maximum allowed percentage change of the so-called stabilization value for each cluster during one update cycle.

• Stabilization value: This is a metric that represents the total conversion value, number of products, or another parameter chosen by you or determined by the system. Products are moved between clusters so that this defined stabilization value of each cluster changes only as much as this percentage limit allows.

• Example: If the stabilization value is the total conversion value of the cluster and the limit is set to 5%, then product movements (outgoing and incoming) must not cause a change in the total conversion value of this cluster by more than +/- 5% during one update.

By carefully setting these parameters for allowing and limiting product movements, you ensure that your clusterization will not only dynamically respond to changes in performance but also maintain the necessary stability for long-term effective functioning of your Google Ads campaigns. In the next section, we will focus on .

This report is your central place for managing and comparing all created clusterization strategies. It allows you to quickly gain an overview of their settings, status, and basic characteristics. You will typically find it by clicking on the highest level of the category tree in the left menu, which represents all clusterizations, or on a specific parent category if you have organized your clusterizations this way.

What the Report Is For:

• Strategy comparison: Easily compare the results of different clusterization designs that you created in (e.g., with different parameters or limits).

• Status monitoring: Quickly determine which clusterization is currently active in production mode (ProductionRun), which is in test mode (Test), or paused (Pause).

• Consistency evaluation: Assess how stable and predictable the clusters are within individual clusterizations.

• Custom Labels management: Get an overview of which Google Custom Labels (custom_label_0 to custom_label_4) are already being used by your clusterizations.

• Optimization planning: Identify older or less performing clusterizations that may require updating, adjustment, or replacement with a new design.

Key Columns and Their Meaning:

Clasterization name

• The name you entered during configuration in .

• It is recommended to use descriptive names for easy identification of the purpose and goals of the clusterization (e.g., Test_ROAS_Price_08-2025).

Created time

• Date and time when the given clusterization (design) was created.

• The report is usually sorted from newest to oldest, which makes it easier to track the latest experiments.

Status

• Current status of the clusterization:

  • ProductionRun: The clusterization is active in production mode. This means it runs daily, updates data, allows dynamic product movements between clusters (if enabled), and exports Custom Labels to Google Merchant Center. In the left menu, this category is marked in green.

Custom label

• The number of the Google Custom Label (from 0 to 4) that was selected for this clusterization in for marking products in Google Merchant Center.

Clusters count

• The total number of clusters (product segments) that were created within this clusterization. For an existing clusterization, this number cannot be changed; you need to create a new clusterization with different limits.

Products count

• The total number of products that are included in all clusters of this clusterization.

Consistency

• A very important performance metric. It indicates the degree of internal consistency (stability and similarity) of product performance metrics within individual clusters of the given clusterization over time.

• A higher value means a better result. Clusters with higher consistency are more internally homogeneous, their behavior is more stable and better predictable for Google Ads optimization algorithms.

• For proper comparison: It is recommended to compare Consistency values between clusterizations that were created using the same number (and ideally the same types) of parameters/dimensions.

Clastering parameters count

• The number of parameters (metrics, e.g., ROAS, Conversion value, CTR) that were used to create this clusterization.

Clastering parameters names

• A list of specific parameter names that were used in the clusterization process for this design (e.g., ROAS, Conv. value).

This overview serves as a starting point for quick orientation in your clusterization strategies and for deciding which ones deserve more detailed analysis, deployment to production, or conversely, adjustment or archiving. For a deeper insight into individual clusters, proceed to the report.

Creating a clusterization design in Cluster Designer is only the first, albeit very important, step. For your new segmentation strategies to bring the expected results, it is essential to carefully analyze, evaluate, and based on the findings, potentially further optimize or select the best variant for deployment to production mode.

Karsa Labelizer provides you with a set of clear reports and analytical tools for this purpose.

Why is Analysis Key?

• Strategy verification: Analysis allows you to check whether the AI algorithms have created clusters that match your expectations and strategic goals.

• Understanding structure: You will see in detail how your products were divided, what characteristics individual clusters have, and which products they contain.

• Identification of performance differences: Reports will help you uncover strengths and weaknesses of individual clusters and the products within them.

• Selection of the best design: If you have created multiple test clusterizations with different parameters, analysis helps you objectively compare their quality (e.g., using the Consistency metric) and select the most suitable one.

• Basis for decision: Based on analytical outputs, you will decide whether the clusterization is ready for and integration with your Google Ads campaigns.

• Continuous optimization: Even after deployment, it is important to monitor reports so you can respond to performance changes and potentially adjust your strategy or product movement settings.

What You'll Find in This Section

This documentation section will guide you through the individual reports and analytical capabilities of Karsa Labelizer:

• : Your central place for managing and comparing all created clusterization strategies. Here you will monitor their status, consistency, and basic parameters.

• : In-depth analysis of performance metrics for each individual cluster within the selected clusterization. It helps you understand the character and performance of each segment.

• : Detailed view of individual products assigned to specific clusters, including their individual performance metrics.

Let's now take a closer look at the individual reports and what information you can gain from them.

Creating a new clusterization in Karsa Labelizer begins in the Cluster Designer. The first step is defining the basic identification details of your new design.

Cluster Designer works on the principle of configurations. It allows you to open individual saved configurations, save them under a new name, and manage them. This enables you to create your own sets of settings (configurations) and later return to them, modify them, or reuse them.

1. Clusterization Name

• What it is: A unique name that will identify your new clusterization in the system. This name will also be used to create a directory (category) that will contain the individual clusters (subcategories) generated by this clusterization.

• Recommendations:

  • Choose descriptive and understandable names that will help you and your colleagues easily identify the purpose and main characteristics of the given clusterization (e.g., Test_ROAS_ConvValue).

  • You can include key parameters, target market, period, or specific product group.

• Automatic date addition: For better clarity, a timestamp is also added to each name of an automatically created directory (clusterization category). The created clusterization can be renamed later.

2. Description of Clustering

• What it is: A text field where you can enter a more detailed description or notes about the clusterization being created.

• Recommendations:

  • Use this field to record specific goals, hypotheses, or specific settings that you used for this clusterization.

3. Destination Parent Category

• What it is: Determines where in your category structure in Karsa Labelizer the newly created clusterization (with all its clusters) will be placed.

• Recommendations:

  • For better organization and clarity, especially if you manage multiple clusterization strategies or tests, you can create main categories (e.g., "Test Clusterizations", "Production Clusterizations", "Archived Designs").

Carefully filling in these basic details will make it easier for you to navigate your clusterization strategies and ensure that the segmentation process will be applied to the correct set of products. After setting these basic parameters, you will continue by defining .

When you select parameters for your clusterization in the Cluster Designer and set the necessary limits, Karsa Labelizer's artificial intelligence (AI) takes the initiative. Unlike simple rule-based systems that segment products based on fixed threshold values, Karsa Labelizer uses advanced machine learning (ML) algorithms to identify optimal and natural groupings (clusters) in your product data.

Basic Principles of AI in Karsa Labelizer

1. Multi-dimensional analysis:

   • AI can analyze products based on multiple parameters (dimensions) of your choice simultaneously. This means it doesn't evaluate products in isolation according to one criterion (e.g., just ROAS), but looks for complex similarities across the entire spectrum of selected metrics (e.g., ROAS, Conversion Value, CTR, Product Price, etc.).

2. Goal: Consistency and Predictability:

   • The main goal of AI is not just to divide products in any way, but to create highly consistent clusters. A consistent cluster contains products that behave as similarly as possible in terms of the chosen parameters.

   • This consistency leads to higher predictability of future campaign performance, which is key for effective optimization by Google Ads algorithms.

3. Hybrid algorithm:

   • Karsa Labelizer uses a sophisticated hybrid machine learning algorithm. This approach combines the advantages of various models and is designed to best suit the specifics of product data in e-commerce.

   • Key features of our algorithm:

What Happens "Under the Hood"? (Simplified)

1. Data preparation: Your chosen parameters and product data are normalized and prepared for analysis.

2. Structure search: The AI algorithm iteratively searches through the data and looks for "clusters" of products that are similar to each other in the chosen dimensions (parameters).

3. Optimization: The process is optimized to:

The result is a data-based segmentation proposal that is ready for your evaluation and subsequent deployment to Google Ads through Custom Labels.

Understanding the basic principles of how AI works will help you better interpret the results of clusterization and more effectively use Karsa Labelizer to optimize your campaigns. In the next section, we will look at practical [Custom Label Setup](../custom-label-setup.md).

Welcome to the section dedicated to Cluster Designer – a key tool in Karsa Labelizer that gives you full power over the design and configuration of your clusterization strategies. This is where you'll transform your product data and business goals into an intelligent structure for Google Shopping campaigns.

What is Cluster Designer?

Cluster Designer is a specialized module of Karsa Labelizer, whose main purpose is to enable you to:

• Define criteria for segmentation: You select which performance parameters (metrics) you want to use to divide your products (e.g., ROAS, Conversion value, Number of conversions, etc.).

• Set limits and goals: You determine the boundaries for the clusterization process, such as the minimum and maximum number of resulting clusters (campaigns) or the minimum number of conversions that each cluster should contain.

• Configure advanced behavior: You set how the system should handle new products without history or how dynamic product movements between clusters should be managed to maintain stability.

• Test different scenarios: You can create multiple variants (proposals) of clusterizations with different settings and then compare them to find the most effective strategy for your e-commerce store.

In short, Cluster Designer is your "operations center" for strategic planning and technical configuration of how your products will be intelligently grouped to achieve maximum performance.

What You'll Find in This Section

This documentation section will guide you through all aspects of working with Cluster Designer:

• : How to start with a new design, basic naming, and description.

• : Detailed explanation of individual limits (number of clusters, minimum conversions, data period) and how to set them optimally.

• : A key subsection that will help you select the right metrics for segmentation and understand how different combinations affect the resulting clusters.

Careful configuration in Cluster Designer is a fundamental prerequisite for creating an effective and high-performing structure for your Google Shopping campaigns. Give this section the attention it deserves.

Welcome to Karsa Labelizer! This guide will show you how to quickly create your first test product clusterization. The goal is to familiarize you with the basic workflow and show you how easy it is to get started. For a detailed understanding of individual steps and settings, we recommend studying the relevant sections of this documentation afterward.

Steps to Create Your First (Test) Clusterization:

Follow these steps and within a few minutes, you'll have your first custom product segmentation proposal ready:

Choosing the right parameters is a key step when designing an effective clusterization in Karsa Labelizer. Parameters determine based on which performance characteristics your products will be divided into individual clusters (future campaigns). Karsa Labelizer uses an advanced hybrid machine learning algorithm that can work with multiple dimensions (parameters) simultaneously and find the optimal number of clusters to maximize their internal consistency and predictability.

Recommendation: Start with 2 parameters. Too many parameters can lead to excessive segmentation and dilution of data, especially for e-commerce stores with a smaller number of conversions.

Default and Most Common Combination: ROAS

While Performance Max (PMax) campaigns are highly automated, their success still significantly depends on the quality of signals you provide them, especially through strategic segmentation of the product feed. generated by Karsa Labelizer are a key tool for effectively structuring your PMax activities.

Why is feed segmentation important for PMax?

• Providing stronger signals to Google AI:

After entering the basic identification details of your new clusterization in the next key step is defining limits and goals. These settings determine the boundaries within which the Karsa algorithm will work, and help ensure that the resulting clusters will be not only statistically optimal but also practically usable for your Google Ads campaigns.

1. Minimum and Maximum Number of Clusters/Campaigns

This setting allows you to determine the range of how many resulting clusters (which usually corresponds to the number of future campaigns or segments) you wish to create.

After you have defined the , , and , the next step in the Cluster Designer is setting up the Google Custom Label. This setting is key for transferring the resulting segmentation from Karsa Labelizer to your Google Ads campaigns.

What is a Custom Label and Why is it Important?

Custom Labels are attributes custom_label_0,

Welcome to the Karsa Labelizer glossary. This section provides more detailed explanations of key terms you may encounter when working with our tool, in the areas of Google Ads, product advertising, and data analysis. If you don't find a term here, also check the .

AI (Artificial Intelligence) : A broad field of computer science concerned with creating systems that exhibit intelligent behavior. In the context of Karsa Labelizer, AI is primarily used for advanced analysis of product data and automated creation of optimal segments (clusters) using machine learning algorithms.

API (Application Programming Interface) : A set of definitions, protocols, and tools for building software and applications. It allows different software systems to communicate with each other and exchange data. Karsa Labelizer can use APIs to download data from Google Ads or to receive order data from clients.

Attribution / Attribution Model : The process of assigning credit for conversions to various marketing channels, ads, or interactions that contributed to a customer's conversion journey. Google Ads offers various attribution models (e.g., Last Click, Data-Driven) that can affect how conversions are reported.

This page collects links to official Google help, professional articles, and other resources that can help you deepen your knowledge in areas related to Karsa Labelizer and optimizing your campaigns.

Official Karsa Labelizer Resources

• Karsa Labelizer Website: https://karsa.ai/en/ (Link to the main product page)

The Products in Cluster report provides you with the most detailed level of analysis in Karsa Labelizer. It allows you to explore the performance metrics of each individual product assigned either to a specific cluster of your choice, or across all clusters within the entire clusterization.

Its main purpose is to:

• Identify the most successful and problematic products: Quickly discover which specific products are driving your business (high ROAS, high conversion value) and which ones are lagging behind or generating losses.

• Optimize product assignment to clusters: Analyze whether products are correctly assigned based on their performance and cluster characteristics.

• Monitor performance at the product level: Track key metrics such as ROAS, CPC, CTR, number of conversions, and their value for each product individually.

How the Report Works and What It Displays

After selecting a cluster or an entire clusterization, the "Products" report will show you a table with the following information for each product (data typically relates to the last 30 days):

Displayed Columns (Metrics) for Each Product:

• Product ID: Unique identifier of the product in your system/feed.

• Name: Product name from the product feed.

Report Features:

• Filtering: Option to display products only from a specific cluster or from all clusters within the selected clusterization. You may also have more advanced filtering options based on the values of individual metrics.

• Sorting: Products can typically be sorted by various columns (metrics). The default sorting is often by conversion value (Conv. value) or ROAS, which helps you quickly identify key products.

• Visual Performance Highlighting (Percentile Coloring):

How to Use This Report for Optimization:

• Identification of "stars" and "dogs": Using sorting and color coding, quickly find products with the highest ROAS and conversion value ("stars") that deserve attention and possibly increased support. Similarly, identify products with high costs and low or negative ROAS ("dogs"), for which you should consider limiting promotion, adjusting price, optimizing the landing page, or completely removing them from active advertising.

• Checking the correctness of cluster assignment: Check whether the performance of individual products corresponds to the overall characteristics of the cluster to which they are assigned. For example, if you find a product with very high ROAS in a "Low ROAS" cluster, it may be a candidate for movement (or a signal that its performance has recently improved).

• Optimizing bids at the product level (if your strategy allows it):

The "Products" report is an essential tool for anyone who wants to truly go in-depth with Google Shopping campaign optimization. It provides transparency down to the level of individual items and allows you to make highly informed decisions.

The Cluster Details report (labeled as "Clusters" in the system) is a key analytical tool in Karsa Labelizer. After selecting a specific clusterization you want to analyze in the Clusterizations Overview report, this report provides you with a detailed view of the performance metrics of each individual cluster contained in that clusterization.

Its main purpose is to allow you to:

• Evaluate the effectiveness of product distribution: Assess how individual segments (clusters) are performing.

• Identify the highest and lowest performing clusters: Quickly determine which segments generate the best results and which ones are lagging behind.

• Uncover anomalies and opportunities for optimization: Find clusters with unusual metric values that may require further investigation or strategy adjustment.

• Optimize bidding strategies: Based on metrics such as ROAS, CPC, and CTR, adjust bidding strategies for individual campaigns corresponding to these clusters.

How the Report Works and What It Displays

The "Labelizer - List clusters" script works with the selected clusterization category (which contains individual clusters as its subcategories). For each cluster within this selected clusterization, it collects and calculates the following performance metrics for the last 30 days:

Displayed Metrics for Each Cluster:

• ROAS (Return on Ad Spend): Return on advertising investment (Conversion Value / Cost).

• CostPerConversion: Average cost of acquiring one conversion.

Data Visualization and Interpretation

The "Cluster Details" report uses several visual elements for easier and faster data interpretation:

• Transposed view: Data is often presented in a transposed table, where:

  • Rows represent individual metrics (ROAS, CPC, CTR, etc.).

  • Columns represent individual clusters of your clusterization.

How to Use This Report for Optimization:

• Comparing cluster performance: Quickly identify which clusters are achieving your ROAS goals and which need attention.

• Analyzing clusterization parameters: Look at how the average values of key metrics differ between clusters. For example, if you clustered by ROAS and price, you will see here how these metrics translated into the characteristics of individual clusters.

• Monitor the number of conversions and products in individual clusters. If these values approach the minimum (threshold) values set for clusterization, consider reducing the total number of clusters being created. The reason is that if some products are discontinued or due to seasonal changes, the cluster could fall below the threshold of sufficient data volume. Such a cluster would become less predictable for Google algorithms, which could negatively affect its performance. When setting up and evaluating clusterization, it is also important to take into account your expert knowledge of catalog management for the given e-commerce store – how often changes occur in the product portfolio, when seasons begin and end for key segments, etc.

The "Cluster Details" report is a powerful tool for in-depth analysis of your segmentation strategy. By regularly monitoring and interpreting this data, you can make informed decisions leading to gradual optimization and performance improvement of your Google Shopping campaigns.

New products added to your product feed present a specific challenge for any performance-based segmentation system. Because they have no history of performance data (such as impressions, clicks, conversions, ROAS, etc.), their placement purely based on current (zero) values would be misleading. For example, a new, potentially very successful iPhone would end up in a "non-performing" cluster if its initial zero performance was taken as final.

Karsa Labelizer offers several strategies for intelligently handling these new products to give them a fair chance to prove their potential. In the Cluster Designer, you can choose one of the following options for Select the behaviour of newly added products:

1. Automatic derivation of values

• Principle: For new products, Karsa Labelizer will try to derive (approximate) their potential performance values based on similar, already existing products in your catalog. For this derivation, attributes such as:

  • Manufacturer (brand)

  • Product category

2. Add to category for new products

• Principle: All newly added products are automatically placed in a special, dedicated category (and thus also campaign in Google Ads).

• Duration of stay: The product remains in this "incubation" campaign for a defined period of 30 days, during which it collects real performance data.

• Subsequent placement: After this period has elapsed and sufficient history has been collected, the product is automatically moved from this special category to the cluster that best corresponds to its actual performance.

3. The product is placed with zero values

• Principle: The new product is assigned to one of the existing clusters based on how the algorithm evaluates "zero" or very low initial values of its performance metrics.

• Placement: This often means that the product ends up in a cluster designated for low-performing or "other" products.

• Gradual data acquisition: The product gradually acquires real performance data and based on this can later be moved to a more suitable cluster (if is allowed).

Which Strategy to Choose?

The choice of optimal strategy depends on the specifics of your assortment and your goals:

• Automatic derivation of values is often a good choice if you have enough similar products in your catalog and believe in the system's ability to estimate the potential of new items. It is more aggressive in trying to quickly place the product correctly.

• Add to category for new products provides the most control and is suitable if you want to have a specific regime (budget, ROAS) for new products and a clearly separated data collection phase. It is a safer, more conservative approach.

Proper handling of new products is an important part of maintaining dynamic and effective segmentation. After setting this strategy and rules for movements, you can proceed to the final .

After logging into Karsa Labelizer, you'll find yourself in a user interface designed for intuitive control and efficient work. This page provides a basic overview of the main parts of the interface.

Main Interface Components

The Karsa Labelizer user interface primarily consists of these parts:

1. Left Navigation Menu (Category Tree and Tools)
2. Main Workspace

Let's look at them in more detail.

1. Left Navigation Menu

The left menu is your main tool for navigating the Karsa Labelizer system. It allows you to access various clusterizations, reports, and configuration sections.

Key features of the left menu:

• Cluster Tree: In the left menu, you'll primarily find one main tree that displays a hierarchical list of your clusterizations. Each clusterization can be expanded to show the individual clusters (product segments) it contains. This structure serves as the main navigation to details and reports of individual clusterizations and their clusters.
• Interaction:

  • Clicking on an item (e.g., the name of a specific clusterization) displays its content in the main workspace.

2. Tools (Switchable Tab)

In addition to the main category/clusterization tree, you can also find a section with key tools in the Karsa Labelizer interface, accessible via switchable tabs. This section brings together the main modules for working with the system:

• Cluster Designer: This is your primary tool for creating, configuring, and managing your clusterization strategies. Here you define segmentation parameters, set limits, control product dynamics, and prepare proposals for deployment. (Described in more detail in section III. Cluster Designer).

• Operation Manager: A module that provides an overview of the status and progress of operations running in the background, especially the process of creating new clusterizations. It allows you to monitor whether an operation is running, how long it takes, and whether it was successfully completed. (Described in more detail in section II. First Steps / Operation Manager).

• Notes: This feature is used to create and manage your own notes directly in the system. You can record important information here, such as the date of launching a new clusterization, specific test goals, observations from performance analysis, or any other relevant comments on your work with the tool.

3. Main Workspace

The main workspace is a dynamic part of the interface that displays the content selected in the left navigation menu. This can be:

• Configuration forms: For example, when creating or editing a clusterization in the Cluster Designer.

• Data reports and analyses: Such as Clusterizations Overview, Cluster Detail, Products Overview, or Product Movement History. These reports often contain tables and metrics for performance evaluation.

The content of the main workspace changes depending on your current selection in the left menu, allowing you to focus on a specific task.

After you have created and configured your clusterization in the Cluster Designer and thoroughly reviewed it using analytical reports, you are ready to deploy it to production mode. This step means that Karsa Labelizer will start exporting Custom Labels for this clusterization to your Google Merchant Center (GMC) daily and possibly also activate dynamic product movements between clusters (if you have enabled them).

Steps for Production Deployment:

1. Switching the Clusterization to "ProductionRun" in Karsa Labelizer

• Select the clusterization: In the left navigation menu of Karsa Labelizer, find the clusterization category you want to deploy.

• Change the status: In the detail of this clusterization category (or in its settings – the exact location may vary depending on the UI version), find the option to change its Status. Choose the status ProductionRun (Production mode).

What happens after switching to ProductionRun:

• Daily export to GMC: Karsa Labelizer will start automatically exporting Custom Label values (with your defined prefix and cluster number, e.g., mycompany_1) for all products included in this clusterization to the XML supplementary feed every day.

• Feed URL: You can find the URL in the left menu Tools > Setup. Use this address when creating a data source in GMC. Set the file loading time to 6-7am.

2. Setting Up/Adjusting Campaigns in Google Ads

Simply switching the clusterization to ProductionRun in Karsa Labelizer and exporting Custom Labels to GMC does not automatically create or modify your campaigns in Google Ads. You must perform this step manually in the Google Ads interface.

Principle: In Google Ads, you will now use the Custom Label values that Karsa Labelizer sends to GMC for product segmentation.

Procedure for Standard Shopping Campaigns:

• New campaigns (recommended for a new structure):

  1. Create new campaigns: For each significant cluster (or logical group of clusters), create a new Shopping campaign in Google Ads.

  2. Campaign settings: Choose the appropriate product feed from GMC, set the budget, bidding strategy (e.g., Target ROAS, Maximize Conversion Value), and other relevant targeting (country, language).

Procedure for Performance Max Campaigns:

You have two main strategic options for using Karsa Custom Labels:

• Strategy A: Multiple separate PMax campaigns (High control - recommended)

  • Create several separate PMax campaigns, where each campaign will target one or more logically related Karsa clusters.

  • In the settings of each PMax campaign (or through its Listing Groups), filter products so that it contains only items with the appropriate Custom Label values from Karsa.

More detailed information and comparison of these strategies for PMax can be found on the page.

3. Setting Budgets and Bids

• For each newly created campaign (or segmented group) in Google Ads, set the appropriate daily budget and bidding strategy (e.g., Target ROAS).

• Initial ROAS targets should be realistic and based on the historical performance of that product segment or on the overall average with a slight reserve for the learning phase.

  • More information:

4. Verification and Monitoring

• Verify in GMC: Check in your Google Merchant Center whether the Custom Label values from Karsa are correctly propagating to products (this may take several hours to one day).

• Verify in Google Ads: Make sure your campaigns and product groups correctly target products with the appropriate Custom Label values.

• Monitor performance:

Remember that after deploying a new campaign structure, an important follows, during which you should not make fundamental changes to the settings.

Deploying a new campaign structure created using Karsa Labelizer to Google Ads is a significant step. After switching the clusterization to production mode and properly setting up your campaigns (whether Standard Shopping or Performance Max), it is key to understand what happens next, and how to approach subsequent optimization. One of the most important concepts is the learning phase of Google Ads algorithms.

What is the Learning Phase?

When you launch a new campaign or make a fundamental change to an existing campaign that uses Google Ads automated bidding strategies (such as Target ROAS - tROAS, Maximize Conversion Value, etc.), the system needs time to "learn" how to best achieve your goals. During this time:

• Algorithms collect data: Google analyzes impressions, clicks, conversions, and other signals associated with your new campaign structure.

• Various bids are tested: The system experiments with different bid levels to find out what works best for a given segment of products.

• Performance may fluctuate: During the learning phase, campaign performance may be unstable – you may notice fluctuations in ROAS, conversion volume, or costs. This is normal and expected behavior.

Recommended Length of Learning Phase

• Generally, count on a learning phase lasting approximately 2 to 4 weeks from the launch of new/significantly changed campaigns.

Key Rule: What (Not) to Do During the Learning Phase

MOST IMPORTANT RULE: Avoid fundamental changes in campaign settings during these first 2-4 weeks!

Frequent or large changes can disrupt, prolong, or even restart the learning process, which delays achieving stable and optimal performance. Changes you should avoid include:

• Changes to bidding strategy goals: For example, frequent adjustments to the target ROAS value.

• Significant changes to daily budget: Avoid abrupt increases or decreases (small adjustments of +/- 20% are usually okay, but with caution).

• Changes to campaign structure or ad groups/asset groups.

What you can do:

• Monitor performance: Track key metrics to stay informed, but don't make hasty conclusions from short-term fluctuations.

• Check feed quality: Make sure your product feed in GMC is in order and error-free.

• Ensure proper conversion measurement: Accurate measurement of conversions with their values is absolutely crucial for the functioning of automated bidding.

Optimization AFTER the Learning Phase (after 2-4 weeks)

Once the initial learning phase has passed and campaign performance begins to stabilize, you can proceed with gradual optimization:

1. Data Evaluation

• Analyze the collected performance data for the past few weeks (ideally at least 30 days with a sufficient number of conversions – see ).

• Look at the actual ROAS, cost per conversion, conversion volume, and their total value for individual campaigns (clusters).

2. Transition to/Adjustment of Target ROAS (tROAS)

• If you started with a different strategy (e.g., Maximize Conversion Value without a target), and now have enough conversion data (ideally 15-30+ over the last 30 days for a given campaign), you can switch to the Target ROAS strategy.

• Setting the initial tROAS target:

  • Be realistic: Don't set the tROAS target too high beyond what the campaign was actually achieving. Too high a target can significantly limit impressions and spending.

3. Gradual Budget Adjustments

• Change the daily budget slowly, ideally not more than +/- 20% every few days (e.g., once every 5-7 days). Avoid sudden large jumps.

• Increase the budget for campaigns that meet or exceed the tROAS target and have potential to grow (e.g., they are not losing impression share due to budget).

• Consider reducing the budget for campaigns that consistently fail to meet the tROAS target and other optimization steps are not helping.

4. Ongoing Optimization and Monitoring

• Search queries: Regularly check the search query report in Google Ads and add irrelevant terms as negative keywords.

• Feed optimization: Continuously work on the quality of data in your product feed in GMC (names, descriptions, images, prices, availability, etc.).

• Karsa dynamic movements: Remember that if you have dynamic movements enabled, Karsa Labelizer will continuously optimize product placement. Monitor the .

By following these procedures – from respecting the learning phase to gradual, data-supported adjustments – you maximize the chance of successful and long-term sustainable results for your campaigns structured using Karsa Labelizer.

The Product Movement History report ("Moving history") is an advanced analytical tool in Karsa Labelizer that allows you to track and analyze in detail how individual products have moved between different clusters (campaigns) over time within your active clusterizations.

Its main purpose is to provide you with:

• Detailed timeline of changes: An exact record of when, from where, and to where a product was moved.

• Performance metrics at the time of movement: What values the product achieved in key metrics at the moment it was moved, and what its values were before the move.

• Basis for evaluating movement effectiveness: Whether product movements actually led to improved performance of the products or entire clusters.

• Audit and transparency: Complete history of changes for reporting and retrospective analysis purposes.

• Ability to evaluate ML algorithm recommendations: Compare whether movements initiated by the system (based on ML recommendations) bring better results than the previous state.

How the Report Works and What It Displays

This report typically displays chronologically sorted records of each product movement. For each record (movement), you can expect the following information:

Key Information Displayed for Each Movement:

• Product identification:

  • Product ID, Name

• Movement information:

Data Visualization and Interpretation

Similar to other reports in Karsa Labelizer, the Product Movement History may use visual elements for easier orientation:

• Metric color coding: Different types of metrics may have different color highlighting (e.g., green for ROAS, blue for conversion value, etc.).

• Highlighting changes:

  • Previous values may be displayed in a darker shade.

How to Use This Report for Optimization:

• Monitoring movement effectiveness: Analyze whether product movements between clusters led to improvements in their key metrics (e.g., whether a product's ROAS actually increased after being moved to a "better" cluster).

• Evaluating ML recommendation accuracy: Compare whether movements made based on ML algorithm recommendations consistently bring better results.

• Auditing changes: Have a complete and transparent history of all automatic changes in product assignment for reporting, control, and possible retrospective analysis purposes.

The Product Movement History report is an invaluable source of information for understanding the dynamic aspects of your AI-driven segmentation. It provides detailed insight into how Karsa Labelizer continuously tries to optimize the assignment of your products, and allows you to verify and further fine-tune this process.

When optimizing Google Shopping campaigns, advertisers have several approaches to product segmentation to choose from. Karsa Labelizer offers a unique solution based on artificial intelligence that differs significantly from traditional methods. Let's look at the main differences:

1. Manual Product Segmentation

Manual segmentation involves manually dividing products into different campaigns or ad groups based on criteria that the advertiser sets themselves (e.g., category, brand, price, seasonality, or rough performance estimate).

Even with careful setup and use of Karsa Labelizer, you may encounter some common problems or situations that require your attention. This page will offer an overview of several typical scenarios and recommended procedures for solving them.

1. The clusterization process did not start or is taking unusually long

Possible causes:

To fully understand how Karsa works in the background and how it contributes to optimizing your Google Shopping campaigns, it's important to familiarize yourself with its typical lifecycle. This cycle describes a sequence of automated and user-controlled steps that regularly repeat.

The following diagram and description illustrate the key phases of the service lifecycle:

1. Daily Data Download from Google Ads

One of the biggest challenges of dynamic product segmentation is maintaining the stability and performance of campaigns in Google Ads. Although the goal of moving products between clusters (campaigns) is long-term optimization, the movements themselves can, if not properly managed, lead to temporary performance declines. This happens because Google Ads algorithms (especially Smart Bidding) need a certain amount of time and consistent data to "learn" how to optimally work with the new composition of products in a campaign.

Karsa Labelizer is aware of this problem and offers robust tools to manage the dynamics of product movements and thus minimize the risk of destabilizing campaigns.

Why is stability so important?

• Learning phase of Google algorithms: Any significant change in the composition of products in a campaign (or ad group/asset group) can trigger a new learning phase for Google algorithms. During this phase, performance may fluctuate.

• Data consistency for Smart Bidding: Automated bidding strategies work best with stable and predictable inputs. Frequent and extensive changes in the portfolio of products in a campaign reduce its consistency and make AI's job more difficult.

• Maintaining achieved ROAS: The goal is for optimization movements to lead to improvement, not to temporary or permanent deterioration of return on advertising investment.

How does Karsa Labelizer help manage stability?

In , you have several key settings at your disposal that allow you to precisely define how intensively and under what conditions products can move between clusters:

1. Limitation of the percentage of products being moved:

   • Each cluster can lose at most % of products: Limits how many percent of products a given cluster can "lose" during one update.

   • Each cluster can gain at most % of products: Limits how many percent of products a given cluster can "gain".

Recommendations for stability settings:

• Start more conservatively: If you are unsure, set the limits for movements rather strictly (smaller percentages, longer waiting times). You can gradually release them based on monitoring performance and stability.

• Monitor Product Movement History: This will show you how often and in what volume products actually move, and what impact this has on their metrics.

• Consider the lifecycle phase of campaigns: For completely new campaigns or after major changes in the assortment, it may be appropriate to have movements temporarily disabled for 2-3 weeks or very limited, as described in

Stability management is a continuous process. Regularly evaluate the settings of your clusterization in light of current performance and don't be afraid to make gradual adjustments to achieve the best possible results.

In this section, you will find answers to some of the most common questions regarding the use and functionality of Karsa Labelizer.

Q1: How often is data from Google Ads updated in Karsa Labelizer?

A: Karsa Labelizer automatically downloads new performance data from your connected Google Ads account every day. This ensures that all analyses and clusterization decisions are based on the most up-to-date information. The system also takes into account retrospective changes in conversions that Google may make. You can find the operation launch time in the Operation Manager.