🔧Common Issues and Solutions

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.

If you don't find a solution to your problem here, or if the problem persists, don't hesitate to contact our support at [e-mail] with as detailed a description of the situation as possible.

---

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

Possible causes:

• Large volume of data: Processing a very large number of products may take longer than the usual 2-15 minutes.

• Temporary technical difficulties: It may be a temporary problem on the server side or with connection to data sources.

• Configuration error: Less common, but incorrect or conflicting settings in the Cluster Designer could theoretically block the process.

What to do:

1. Check the Operation Manager: Look in the [Operation Manager](../ii.-first-steps/operation-manager.md) to see if the operation is still running or if it displays any error message.

2. Wait: If you are processing a large catalog, give the process more time (e.g., 30-60 minutes).

3. Try again later: If it's a short-term problem, try to start the operation again with a small time delay.

4. Check the configuration: Go through all the settings of your clusterization in Cluster Designer again to see if they contain obvious errors or illogical settings (e.g., extremely low conversion limits combined with a requirement for many clusters).

5. Contact support: If the problem persists, contact Karsa Labelizer support and provide them with the name of your clusterization and the start time.

---

2. The resulting consistency (Consistency) of clusters is low

Possible causes:

• Too much variance in the data: If your products are extremely diverse in terms of the chosen parameters, it may be difficult for AI to create highly consistent clusters.

• Inappropriately chosen clusterization parameters: Some parameter combinations may not lead to clearly separable and consistent groups for your specific data.

• Too high or low number of requested clusters: Extreme requirements for the number of clusters given the nature of the data can reduce consistency.

• Lack of conversion data: If many products or an entire segment has very few conversions, it is difficult to determine their stable performance.

What to do:

1. Analyze [Clusterizations Overview](../iv.-analysis-and-optimization/clusterizations-overview.md): Compare consistency with your other clusterizations or with test variants.

2. Experiment with parameters: Create a new test clusterization with a different combination or fewer parameters. Focus on those most relevant to your business goals. See [Parameter Selection Strategy](../iii.-cluster-designer-creating-clusterization/parameter-selection/parameter-selection-strategy.md).

3. Adjust cluster number limits: Try slightly adjusting the minimum and maximum required number of clusters.

4. Reduce the minimum number of conversions per cluster: If your data allows, try setting a lower threshold for Minimum number of conversions.

5. Analyze products in clusters: Look at the [Products in Cluster](../iv.-analysis-and-optimization/products-in-cluster.md) report to see if low consistency is caused by a few products with extremely different behavior.

---

3. Custom Labels are not properly propagating to Google Merchant Center (GMC)

Possible causes:

• Clusterization is not in ProductionRun mode: Only clusterizations in this state actively generate XML feed for GMC.

• Delay in GMC: Propagation of data from supplementary feeds (which Karsa uses for Custom Labels) to GMC can sometimes take several hours.

• Conflict with another feed or rule in GMC: Another data source or rule in GMC may be overwriting Custom Label values set by Karsa.

• Exhaustion of limit for Custom Labels: In GMC, only 5 slots are available for Custom Labels (custom_label_0 to custom_label_4).

What to do:

1. Verify the clusterization state: Make sure your clusterization in Karsa Labelizer is switched to the ProductionRun state.

2. Check Custom Label settings in Karsa: Verify that you have correctly chosen the Custom label slot and defined the Prefix in Cluster Designer.
3. Wait: Give GMC at least several hours (ideally up to 24) to process the data after the first launch of the production clusterization.

4. Check GMC:

   • Look in the "Feeds" section of your GMC to see if the supplementary feed from Karsa (if implemented this way) is active and without errors.

   • Directly check several products in GMC to see if they display the expected value (including your prefix) in the corresponding Custom label X attribute.

   • Check "Feed rules" in GMC to see if any rule is overwriting your chosen Custom Label.

5. Contact support: If the problem cannot be resolved, contact Karsa Labelizer support.

---

4. Campaign performance declined after deploying a new clusterization

Possible causes:

• Not respecting the learning phase: You made too many changes to Google Ads campaigns (budgets, tROAS goals) too soon after deploying the new structure.

• Too aggressive goals: You set unrealistically high tROAS goals for new campaigns.

• Inappropriate cluster structure for your goals: Even if the clusterization is technically correct, the current division may not suit your specific business strategy or product margins.

• Problems outside Karsa Labelizer: Performance decline may also be caused by external factors (seasonality, competitor activities, technical problems on the website, changes in Google Ads algorithms).

What to do:

1. Respect the learning phase: As described in [Post-Deployment Optimization (Learning Phase)](../v.-deployment-and-strategy/post-deployment-optimization.md), after deploying a new structure, let the campaigns run for at least 4-6 weeks without major changes.

2. Start with realistic goals: Don't set tROAS too high. Start more conservatively and gradually increase.

3. Analyze data in Karsa and Google Ads: Examine in detail the performance of individual clusters (campaigns) and products. Where exactly did the decline occur? Are some clusters performing better than others?

4. Consider adjusting the clusterization strategy: If analysis shows that the current division is not optimal, return to Cluster Designer and try to create a test variant with different parameters or limits.
5. Exclude external factors: Check if the decline is not related to other changes in your marketing mix or in the market.

---

This list is not exhaustive, but it should cover some of the most common situations. The key to problem-solving is a systematic approach, careful data analysis, and, if needed, consultation with support.