Understanding Data Granularity and Aggregation in Cassandra

Proper data granularity and aggregation are essential for ensuring accurate Marketing Mix Modeling (MMM) results. The level of detail in your dataset directly impacts model performance, prediction accuracy, and budget allocation strategies.

This guide explains when to use daily, weekly, or monthly data, how campaign aggregation affects model accuracy, and best practices for structuring historical data.

1. When to Use Daily vs. Weekly vs. Monthly Data

Daily Data (High Granularity)

Best for:

• Short-term campaign optimizations (e.g., flash sales, limited-time promotions).

• Paid media platforms with frequent spend adjustments (Google Ads, Meta, TikTok Ads).

• Analyzing short-term trends and external influences (e.g., weather, news events).

Considerations:

• Requires high data volume and computational power.

• Can be noisy due to daily fluctuations in marketing performance.

• Best suited for digital performance marketing where day-to-day changes impact results.

Weekly Data (Balanced Granularity)

Best for:

• General MMM modeling – Provides a balance between detail and stability.

• Multi-channel media analysis, especially if some channels update performance weekly.

• Budget allocation decisions over weeks or months.

Considerations:

• Reduces daily noise while maintaining meaningful trends.

• Ideal for evaluating cross-channel interactions and optimizing spend.

• Works well for offline and online media mix analysis.

Monthly Data (Low Granularity)

Best for:

• High-level strategic planning (e.g., annual budgeting, brand impact analysis).

• Traditional media channels with long-term effects (TV, print, OOH advertising).

• Subscription-based businesses, where customer acquisition effects are cumulative.

Considerations:

• Lacks precision for campaign-level optimization.

• Best suited for long-term trend analysis and macroeconomic factors.

• Not ideal for fast-moving digital campaigns that require quick adjustments.

2. How Campaign Aggregation Affects Model Accuracy

Why Aggregation Matters

• Helps reduce multicollinearity by combining similar campaigns.

• Prevents overfitting when too many granular variables are included.

• Ensures the model captures broad trends instead of campaign-specific noise.

Best Practices for Campaign Aggregation

• Group by Campaign Type: Instead of analyzing each ad individually, aggregate by campaign type (e.g., Prospecting vs. Retargeting).

• Separate Branded vs. Non-Branded: For search and display campaigns, keep branded and non-branded activities separate.

• Combine Similar Channels: For example, if Google Display and YouTube serve the same purpose, they can be analyzed together.

• Factor in Spend Thresholds: If a campaign accounts for <3% of total spend, consider merging it with a larger category.

Examples of Aggregation Strategies

3. Best Practices for Structuring Historical Data

Ensuring Data Completeness

• Include at least 24 months of historical data for stable trend analysis.

• Ensure data covers all active marketing channels, even those phased out recently.

• Capture seasonal and macroeconomic effects by maintaining a continuous dataset.

Standardizing Formats

• Use consistent date formats (YYYY-MM-DD for daily, YYYY-WK for weekly).

• Align all cost and performance metrics across channels.

• Separate fixed costs (e.g., annual sponsorships) from performance-based media spend.

Handling Missing Data

• If data gaps exist, use imputation techniques (e.g., rolling averages) instead of zero-filling.

• Clearly label estimated or interpolated data to avoid misinterpretation.

4. Summary & Next Steps

• Daily data is best for short-term optimizations, while weekly data provides a balance, and monthly data supports strategic planning.

• Campaign aggregation improves model efficiency by reducing noise and collinearity.

• Maintain clean, structured, and complete historical datasets for the most accurate MMM results.

• Regularly review aggregation levels to ensure alignment with business decision-making needs.