This FAQ document answers the most common questions about Cassandra, its functionalities, integrations, and best practices.
General Questions
Q: Which MMM is Cassandra based on?
A: Cassandra AI is a proprietary model developed by our team of experts in Marketing Mix Modeling. Approach-wise, we follow the same approach and use similar techniques as Meta's open-source Project Robyn.
Q: How can I validate the insights from the Cassandra platform?
A: Validating insights ensures their accuracy and reliability. Here’s how:
- Data Analysis – Compare Cassandra’s insights with historical data trends.
- Incrementality Experiments – Conduct geo-experiments or audience-based experiments to validate impact.
- Expert Review – Consult Cassandra’s team for additional insights.
Q: Where can I find all my saved models?
A: You can find saved models in the 'Models Overview' section. If a model is still in training, check the 'Models List' page.
Q: What is the difference between a frequentist and Bayesian MMM model?
A: Frequentist MMM (e.g., Facebook Robyn) uses ridge regression to estimate the impact of each factor. Bayesian MMM (e.g., Uber Orbit) incorporates priors, making it more influenced by user-defined expectations.
Q: How does the Model Refresh work? Does it create a new model?
A: No, Model Refresh updates your existing model with new data, fine-tuning contributions, adstocks, and saturation curves. You can also remove or add variables during the refresh.
Q: How should I choose my modeling window?
A: Consider business context, data sparsity, and ensuring a recommended ratio of 1 input variable to 10 rows of data.
Data & Integration
Q: What type of analysis can I do on the Model’s Details page?
A: You can perform:
- Share of Spend vs. Share of Effect – Compare marketing investment vs. impact.
- Decomposition Over Time – Analyze how variables contribute over time.
- Spend Over Time – See how investment fluctuations affect performance.
- KPI Over Time per Week – Identify campaign performance trends.
Q: For marketing variables, do I need both spending and engagement metrics (impressions, clicks)?
A: No, spending data is sufficient. Click and impression data are optional but can enhance model accuracy.
Q: How does Cassandra handle seasonality?
A: Cassandra identifies recurring trends in your data (e.g., Black Friday sales) and accounts for them in the model.
Optimization & Budgeting
Q: How do I use the Budget Allocator?
A: Define a time period and budget, set constraints (if necessary), and Cassandra will optimize the allocation based on diminishing returns and uncertainty.
Q: Why is the Budget Allocator so conservative?
A: Cassandra considers uncertainty and historical budget stability. You can override constraints manually to increase flexibility.
Q: How should I read the Saturation Plot?
A: The X-axis represents weekly/daily budget, while the Y-axis shows expected KPI output. The blue line represents historical average spend.
Q: Why does my model have low accuracy?
A: Common reasons include:
- Missing Variables – The model lacks explanatory factors.
- Multicollinearity – Strong internal correlation between variables.
- Low Spend Volumes & Historical Data – Limited data reduces model reliability.
Advanced Features & Customization
Q: Is it possible to export the data from the Diminishing Returns chart?
A: Yes, but only for premium users. Click the export/download button to retrieve the data.
Q: How and why should I manually edit hyperparameters?
A: You can customize hyperparameters in the 'Advanced Model Builder' to better reflect known adstock and saturation values.
Q: Why is the ROI in Cassandra different from Google Analytics 4 (GA4)?
A: Cassandra uses incremental attribution, while GA4 typically uses last-click attribution, leading to different ROI estimates.
Q: How do I interpret Confidence Intervals?
A: Confidence intervals indicate uncertainty in attributions. A wider interval means higher uncertainty, and such channels may benefit from additional calibration or experiments.
Q: How do I add more people to my project?
A: Go to 'Invite Team Members' and assign permissions as needed. There is no limit on team members.
Final Notes
This FAQ is continuously updated as new features and methodologies are introduced in Cassandra.