Common Troubleshooting & Debugging Tips

If you encounter issues with your Cassandra models, this guide provides solutions for common problems related to data integration, model accuracy, budget allocation, and more.

Step 1: Troubleshooting Data Issues

1. Data Not Uploading or Missing Values
   • Ensure the file format is CSV and adheres to Cassandra’s YYYY-MM-DD date format.
   • Check for missing required columns (e.g., spend, conversions, revenue).
   • Verify data consistency (e.g., no mixed date formats or text in numeric columns).
   • Large files? Try splitting them into smaller parts before uploading.
2. Incorrect or Mismatched Data in Reports
   • Ensure column mappings are correct when uploading data.
   • If API-connected, check if the latest sync was successful.
   • Look for duplicates in your dataset that may be inflating values.
3. Discrepancies Between Cassandra & Platform Reports
   • Cassandra uses incremental attribution, which differs from platform-reported last-click attribution.
   • Compare platform reports with Cassandra’s multi-touch model.
   • Run incrementality tests to validate model accuracy.

Step 2: Troubleshooting Model Accuracy

1. Why is My Model’s Accuracy Low?
   • Ensure you have at least 2 years of historical data for best results.
   • Remove variables with low spend (<3% of budget) or low variance.
   • Check for multicollinearity—if multiple channels have highly correlated spend, try aggregating them.
2. Why Do Some Channels Show Zero Contribution?
   • Channels with low historical spend or infrequent campaigns may lack statistical significance.
   • If a channel is new, wait until at least 4-6 weeks of data before adding it to the model.
   • Run GeoLift or Conversion Lift experiments to validate contributions.
3. Baseline Contribution Seems Too High or Too Low
   • The baseline represents organic sales and brand equity—a high baseline suggests strong non-paid revenue sources.
   • If baseline seems too high, check if some paid channels are under-attributed and require calibration.
   • Increase the historical data window to improve trend analysis.

Step 3: Troubleshooting Budget Allocator & Predictions

1. Budget Allocator Suggestions Seem Off
   • Check if recent budget changes were reflected in the model before running the allocator.
   • If channels are over-allocated, look at their saturation curves—some may be hitting diminishing returns.
   • If channels are under-allocated, ensure they have enough historical spend to be accurately modeled.
2. Predicted ROI Doesn’t Match Real Performance
   • ROI predictions assume stable conditions—if external factors change (e.g., economic shifts, new competitors), adjustments may be needed.
   • Run incrementality experiments to validate the model’s attributions.
   • Compare past model forecasts with actuals to assess calibration accuracy.

Step 4: Technical Issues & Platform Errors

1. Model Refresh Fails
   • Check if data sync was completed before running a refresh.
   • Ensure the date range is correct—it should align with existing model boundaries.
   • If warnings appear about missing variables, verify the latest dataset includes all columns.
2. Platform is Slow or Unresponsive
   • Try clearing your browser cache and reloading the page.
   • Large datasets may cause delays—limit query ranges or use filters to refine results.
   • If issues persist, contact Cassandra support.

Summary & Next Steps

• Validate data formatting before uploading.
• Use incrementality tests to validate contributions.
• Adjust budget allocation based on saturation curves.
• Check historical data length if model accuracy is low.