Organizing Variables into Groups
Strategic Approach to Variable Grouping
Organizing variables into groups is a critical step that determines how you'll analyze and communicate your MMM results. The grouping structure should reflect your business priorities, decision-making processes, and reporting needs.
Understanding Variable Types
Before grouping, understand the types of variables in your model:
Marketing Variables:
Media spend (TV, Digital, Radio, etc.)
Promotional activities
Marketing events
Campaign-specific variables
Control Variables:
Price changes
Distribution metrics
Seasonality indicators
External factors
Base/Structural Variables:
Constant term (intercept)
Time trends
Baseline performance metrics
Grouping Frameworks
Framework 1: Marketing Channel Groups
This is the most common approach for Marketing Mix Models:
Group Structure:
When to Use:
Standard MMM analysis
Budget allocation decisions
Channel performance comparison
Executive reporting
Benefits:
Clear channel attribution
Matches budget structure
Familiar to stakeholders
Enables ROI calculation
Framework 2: Media Type Grouping
Separate digital from traditional media:
Group Structure:
When to Use:
Comparing digital vs. traditional effectiveness
Different teams manage digital vs. traditional
Budget shifting between online/offline
Benefits:
Strategic digital transformation insights
Compares fundamentally different media types
Highlights digital growth opportunities
Framework 3: Customer Funnel Grouping
Organize by marketing funnel stage:
Group Structure:
When to Use:
Customer journey optimization
Full-funnel marketing strategy
Attribution across touchpoints
Benefits:
Aligns with customer journey
Identifies funnel gaps
Optimizes customer acquisition cost
Framework 4: Brand vs. Performance
Separate brand-building from performance marketing:
Group Structure:
When to Use:
Balancing short-term and long-term goals
Different ROI expectations for brand vs. performance
Separate budget owners
Benefits:
Strategic brand investment insights
Different time horizons considered
Balances immediate and future returns
Framework 5: Controllable vs. Uncontrollable
Separate variables you control from external factors:
Group Structure:
When to Use:
Focusing on actionable insights
Isolating marketing contribution
Understanding market dynamics
Benefits:
Separates what you can influence
Highlights marketing effectiveness
Contextualizes external impacts
Practical Grouping Guidelines
How Many Groups?
Too Few Groups (< 3):
❌ Loses granularity
❌ Can't identify specific drivers
❌ Limited actionability
Optimal Groups (5-10):
✓ Balanced detail and clarity
✓ Easy to visualize
✓ Actionable insights
✓ Executive-friendly
Too Many Groups (> 15):
❌ Defeats purpose of grouping
❌ Cluttered visualizations
❌ Difficult to interpret
❌ Better to use drill-down instead
Naming Conventions
Good Group Names:
Clear and descriptive
Business-relevant terminology
Consistent across models
No abbreviations or jargon
Appropriate length (1-3 words)
Examples:
✓ "TV Advertising"
✓ "Digital Marketing"
✓ "Promotions"
✓ "Seasonality"
Poor Group Names:
❌ "Grp1"
❌ "TV_ADS_SPEND_FINAL"
❌ "MKT"
❌ "Variables Related to Media Advertising and Marketing Communications"
Handling Special Cases
Intercept/Constant:
Always assign to "Base" group
Represents baseline KPI level
Should be its own group or combined only with time trends
Interaction Terms:
Group with the primary variable
Example: TV × Promotion → assign to TV or Promotions based on analysis goal
Document interaction effects
Transformed Variables:
Group by original variable meaning, not transformation type
TV_Ad70 (adstocked TV) → assign to "TV" or "Media"
Don't create groups like "Adstocked Variables"
Lagged Variables:
Group with contemporaneous variable
Example: Sales_Lag1 → same group as other sales variables
Maintain logical relationships
Step-by-Step Grouping Process
Step 1: Identify Variable Categories
List all variables and categorize them:
Marketing spend variables
Which channels?
Which campaigns?
Pricing and promotion variables
Price changes?
Promotional indicators?
Seasonal and calendar variables
Holidays?
Monthly effects?
External/control variables
Competition?
Economic factors?
Structural variables
Constant term?
Time trends?
Step 2: Map to Business Structure
Align with how your business thinks about marketing:
Questions to Ask:
How are budgets allocated?
Who owns each channel?
How do stakeholders discuss performance?
What decisions need to be made?
Example Mapping:
Step 3: Consider Analytical Goals
Different analyses may require different groupings:
For Budget Optimization:
Group by budget line items
Keep channels you might shift budget between separate
Combine channels with fixed budget relationships
For Executive Reporting:
Use high-level categories
Match board presentation structure
Align with strategic priorities
For Operational Decisions:
More granular channel grouping
Separate test vs. always-on
Group by campaign type
Step 4: Apply Grouping in MixModeler
In the Contribution Groups page:
Select your model from the dropdown
Review all variables in the table
Start with major categories:
Assign all media variables first
Then pricing/promotion variables
Then seasonal variables
Finally base/structural variables
Use multi-select for efficiency:
Check multiple TV channels
Assign all to "TV" group at once
Refine and verify:
Check for consistent naming
Ensure no variables are unassigned
Validate grouping logic
Examples from Real Models
Example 1: CPG Brand Model
Model Variables:
TV_National
TV_Local
Digital_Display
Digital_Video
Search_Branded
Search_Generic
Social_Paid
Radio
OOH
Print_Magazine
Trade_Promotions
Consumer_Promotions
Price_Index
Holiday_Indicators
Trend
Grouping Structure:
Example 2: E-Commerce Model
Model Variables:
Search_Brand
Search_NonBrand
Social_Facebook
Social_Instagram
Display_Prospecting
Display_Retargeting
Video_YouTube
Email
Affiliates
Promotion_Discount
Shipping_Free
Month_Dummies
COVID_Period
Grouping Structure:
Example 3: B2B SaaS Model
Model Variables:
Paid_Search
LinkedIn_Ads
Display_Retargeting
Content_Syndication
Webinar_Spend
Event_Sponsorships
Email_Campaigns
Organic_Traffic
Free_Trial_Rate
Price_Changes
Comp_Activity
Grouping Structure:
Common Grouping Mistakes
Mistake 1: Over-Grouping
Putting all media in one "Marketing" group
Loses ability to compare channels
Can't make optimization decisions
Solution: Separate major channels or at least digital vs. traditional
Mistake 2: Inconsistent Naming
"Media" in one model, "Marketing" in another
"TV Ads" vs "Television" vs "TV"
Makes cross-model comparison difficult
Solution: Standardize group names across all models
Mistake 3: Mixing Levels
Combining "TV" group with specific channel "Facebook"
Different levels of aggregation
Confusing comparisons
Solution: Keep groups at the same hierarchical level
Mistake 4: Ignoring Business Structure
Grouping that doesn't match how business operates
Groups that don't align with decision-making
Solution: Involve stakeholders in grouping decisions
Mistake 5: Too Many Small Groups
20+ groups each with 1-2 variables
Cluttered visualizations
Defeats purpose of grouping
Solution: Combine related variables, use drill-down for details
Testing Your Grouping
After grouping, validate by:
1. Run Decomposition
Do the groups make visual sense?
Are contributions interpretable?
Can you tell the story?
2. Share with Stakeholders
Do they understand the groups?
Do groups match their mental model?
Can they make decisions based on it?
3. Check for Actionability
Can you optimize based on groups?
Do groups align with budget allocation?
Are insights clear?
4. Verify Completeness
All variables assigned?
No overlapping groups?
Clear group boundaries?
When to Regroup
Regroup variables when:
Business structure changes
New stakeholders with different needs
Analysis goals shift
Current grouping doesn't yield clear insights
Adding/removing variables
Visualizations are cluttered
Flexibility: Grouping is not permanent - adjust as needed to serve analytical goals
Next Steps
After organizing variables:
Assign colors to groups for visualization
Set adjustment parameters if needed
Save configuration
Run decomposition analysis
Use group drill-down for detailed insights
Last updated