Parameter Selection Strategy

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 + Conv. value

This combination is set as the default in Cluster Designer because it provides a natural and strategically very useful view of product performance that combines cost efficiency (ROAS) with volume of generated revenue (Conversion Value).

Why it's interesting: It allows separating products that are highly profitable from those that bring large turnover, even if perhaps with lower margin or higher advertising costs.

Typically Emerging Clusters:

• High-performing (Stars): High ROAS, high conversion value. Key products, deserve maximum support.

• High Potential: High ROAS, lower conversion value. Profitable products with smaller volume, opportunity for growth.

• Volume Drivers: Lower ROAS, high conversion value. Generate revenue, but there is room for optimizing efficiency (costs, bids).

• Problematic: Low ROAS, low conversion value. Candidates for budget limitation or strategy reconsideration.

• (Possibly) Unbalanced Products: Extremely high ROAS with very low conversion value or vice versa. These require individual analysis.

Extension with Frequency: ROAS + Conv. value + Conversions

By adding the Conversions parameter to the previous pair, you get an even more detailed view that also takes into account sales frequency.

Why it's interesting: It distinguishes products with similar ROAS and total conversion value but different sales dynamics (few expensive vs. many cheap sales). Suitable for stores with a sufficient number of orders where you want to better understand the average order value within segments.

Example of New Types of Clusters:

• Premium Products: High ROAS, high conversion value, high number of conversions. The absolute top of your portfolio.

• Expensive Bestsellers: Lower ROAS, high conversion value, high number of conversions. Popular, generate volume of revenue.

• Exclusive Products: High ROAS, high conversion value, low number of conversions. Typically expensive goods with high margin, sold less frequently.

• Effective Low-turnover: High ROAS, low conversion value, low number of conversions. Niche, but effective.

• High-frequency Cheap Products: High/medium ROAS, low conversion value, high number of conversions. Complementary goods, consumables.

• Volume Loss-making Products: Low ROAS, medium conversion value, high number of conversions. May fulfill a strategic role.

Other Interesting Strategies and Parameter Combinations:

1. Conversion Funnel Efficiency: CTR vs. Conversion Rate (Conversions / Clicks)

• Parameters: CTR (Click-Through Rate) and calculated Conversion Rate.

• Why it's interesting: This combination shows you how effective your entire "marketing funnel" is from ad impression to conversion. It helps separate products that have attractive ads (high CTR) from those that effectively convert visitors after clicking (high conversion rate).

• Possible clusters:

  • High CTR / High Conversion Rate: Ideal state, your drivers.

  • High CTR / Low Conversion Rate: "Click catchers". Ads are attractive, but the problem is on the landing page, in the price, or in the offer. (Optimize LP/offer).

  • Low CTR / High Conversion Rate: "Hidden treasures" / Specific demand. Ads don't attract, but those who click often convert. (Improve ads/creatives).

  • Low CTR / Low Conversion Rate: Weak products, require major revision or removal.

2. Profitability vs. Traffic Volume/Visibility: ROAS vs. Clicks (or Impressions)

• Parameters: ROAS and Clicks, alternatively Impressions.

• Why it's interesting: Shows whether products that generate the most traffic or have the greatest visibility are also the most profitable. Helps identify products that "swallow" a lot of budget (through CPC) but don't bring adequate return.

• Possible clusters:

  • High ROAS / High Number of Clicks: Great - profitable products with large traffic.

  • High ROAS / Low Number of Clicks: Profitable, but with low visibility/clickability. (Consider increasing visibility).

  • Low ROAS / High Number of Clicks: "Popular, but loss-making". Generate a lot of traffic, but are not profitable. Often a large part of costs. (Optimize or limit).

  • Low ROAS / Low Number of Clicks: Non-performing and uninteresting.

3. Product Price Level vs. Efficiency: Avg product price vs. ROAS

• Parameters: Avg product price and ROAS.

• Why it's interesting: Reveals the relationship between product price and the effectiveness of its promotion. Are more expensive products more profitable to promote (due to higher margin)? Or conversely cheaper ones (due to easier conversions)? Helps adapt strategy for different price segments.

• Possible clusters:

  • High Price / High ROAS: "Profitable premium products".

  • High Price / Low ROAS: "Expensive to sell". High price doesn't guarantee ad profitability.

  • Low Price / High ROAS: "Effective 'cheap' goods".

  • Low Price / Low ROAS: "Low-margin problems".

Recommendation: Don't be afraid to experiment with different parameter combinations in the test mode (Test) of your clusterization. Watch how the resulting cluster structure and the Consistency metric change in the Clusterizations Overview report. Choose a strategy that best corresponds to your business goals, the nature of your assortment, and the distribution of your data.

In the next section, we will look at How AI Works in Cluster Finding.