Contribution Groups Setup

What Are Contribution Groups?

Contribution groups organize your model variables into logical business categories for decomposition analysis. Instead of viewing the impact of dozens of individual variables, you group related variables together to understand contributions at a strategic level.

Purpose: Organize model variables into meaningful business categories (Media, Price, Promotions, etc.) to enable group-level decomposition analysis and insights.

Why Group Variables?

Strategic Clarity: See high-level contributions rather than getting lost in granular details

Executive Communication: Present results in business terms stakeholders understand

Simplify Analysis: Reduce dozens of variables into 5-10 meaningful groups

Enable Drill-Down: Start with group view, then drill into individual variables when needed

Consistent Reporting: Use standardized groups across all models and time periods

How Contribution Groups Work

The decomposition calculation works as follows:

Variable Contribution = Coefficient × Variable Value

For each time period, every variable's contribution is calculated by multiplying its coefficient by its value at that time.

Group Contribution = Sum of All Variables in Group

Variables assigned to the same group have their contributions summed together for each time period.

Total Predicted Value = Sum of All Group Contributions

All group contributions add up to equal the model's predicted value.

Accessing Contribution Groups

Navigation:

  1. Click "Contribution Groups" in the left sidebar

  2. Select your fitted model from the dropdown

  3. View all variables in your model with their properties

Prerequisites:

  • Model must be fitted (OLS or Bayesian)

  • Model must have at least one variable besides the constant

Understanding the Variable Table

The Contribution Groups page displays a comprehensive table with the following columns:

Column
Description
Example

Variable

Variable name from your model

TV_Spend_Ad70

Coefficient

Impact size (green=positive, red=negative)

2.45 (green)

T-Stat

Statistical significance (bold if |t| ≥ 1.96)

3.21

Transform

Data transformation applied

STA, SUB, MDV, ADBUG

Group

Business category you assign (editable)

Media

Adjustment

Optional mathematical adjustment

None, Min, Max

Color Coding:

  • Green coefficients: Positive impact on KPI

  • Red coefficients: Negative impact on KPI

  • Bold t-statistics: Statistically significant (|t| ≥ 1.96)

Variable Information Displayed

Coefficient:

  • Shows the impact size and direction

  • Positive values increase KPI

  • Negative values decrease KPI

  • Magnitude indicates strength of effect

T-Statistic:

  • Measures statistical significance

  • |t| ≥ 1.96 is significant at 95% confidence

  • |t| ≥ 2.58 is significant at 99% confidence

  • Bold formatting highlights significant variables

Transform:

  • Shows any transformations applied to the variable

  • STA: Standardized

  • SUB: Subset

  • MDV: Multiply/Divide

  • ADBUG: Adstock with Adbudg saturation curve

  • ADSUBICP: Adstock with SUB and ICP saturation

  • Multiple transformations may be chained

Organizing Variables into Groups

Step 1: Assign Group Names

Click the Group dropdown for each variable and select or type a group name:

Key Principles:

  • Use consistent naming across all variables

  • Choose business-relevant category names

  • Keep names short and descriptive

  • Use the same name for all related variables

Example Assignments:

TV_Spend → Media
Digital_Spend → Media
Search_Spend → Media
Price_Variable → Price
Holiday_Indicator → Seasonality
Promotion_Flag → Promotions

Step 2: Use Multi-Selection for Efficiency

Multi-select features allow bulk operations:

  1. Check boxes next to variables to select multiple

  2. Select All using the header checkbox

  3. Bulk assign group to all selected variables

  4. Bulk set adjustment for all selected variables

Use Case: Select all digital channels at once and assign them to "Digital Media" group

Step 3: Review Group Organization

After assigning groups, review the organization:

Check for completeness:

  • Every variable should be assigned to a group

  • No variables left unassigned

Verify consistency:

  • Similar variables use the same group name

  • No typos or variations (e.g., "Media" vs "media" vs "Marketing")

Validate business logic:

  • Groups make sense for your business

  • Grouping aligns with how you think about drivers

Common Group Categories

Use these standard categories as a starting point:

Base:

  • Constant term (intercept)

  • Baseline variables

  • Trend variables

  • Purpose: Represents the KPI level when all marketing is zero

Media:

  • All advertising and marketing channels

  • TV, Radio, Display, Video, OOH

  • Both online and offline media

  • Purpose: Total marketing contribution

Digital:

  • Online marketing channels

  • Search, Social, Display, Video, Email

  • Sometimes separated from traditional Media

  • Purpose: Digital-specific attribution

Price:

  • Pricing variables

  • Discount rates

  • Price indices

  • Purpose: Price elasticity and impact

Promotions:

  • Promotional indicators

  • Sale events

  • Coupon campaigns

  • Purpose: Promotional lift measurement

Seasonality:

  • Holiday indicators

  • Monthly/quarterly dummies

  • Seasonal indices

  • Purpose: Seasonal effects on KPI

External:

  • Competitor activity

  • Economic indicators

  • Weather data

  • Purpose: Factors beyond your control

Distribution:

  • Store count

  • Distribution expansion

  • Availability metrics

  • Purpose: Distribution effects

Group Categories & Best Practices

Best Practices for Grouping

1. Align with Business Structure

  • Mirror your organization's marketing structure

  • Match how budgets are allocated

  • Reflect decision-making hierarchies

2. Balance Granularity

  • Not too few groups (lose insights)

  • Not too many groups (defeats the purpose)

  • Typically 5-10 groups is optimal

3. Consider Analysis Goals

For Budget Allocation:

  • Group by controllable spend categories

  • Separate fixed vs. variable costs

  • Match budget line items

For Channel Performance:

  • Group by media type

  • Separate brand vs. performance channels

  • Group by customer funnel stage

For Executive Reporting:

  • Use high-level business categories

  • Group by strategic initiatives

  • Match stakeholder mental models

4. Enable Drill-Down

  • Use group decomposition for detailed analysis

  • Keep individual variable names descriptive

  • Maintain hierarchical relationship

Strategic Grouping Patterns

Pattern 1: Media Type Grouping

Traditional Media: TV, Radio, Print, OOH
Digital Media: Search, Social, Display, Video

Pattern 2: Funnel Stage Grouping

Awareness: TV, Display, Video
Consideration: Content, Email, Social
Conversion: Search, Retargeting, Email

Pattern 3: Owned vs. Paid

Paid Media: All advertising spend channels
Owned Media: Email, Organic Social, SEO

Pattern 4: Brand vs. Performance

Brand Building: TV, Display, Sponsorships
Performance: Search, Affiliates, Retargeting

Color Assignment Strategy

Assigning Colors to Groups

Once variables are grouped, assign a distinct color to each group:

Access Color Picker:

  • Color picker appears below the variable table

  • Shows all unique groups

  • Click the color swatch to choose a color

Choosing Effective Colors:

Group Type
Recommended Color
Reason

Base

Gray (#808080)

Neutral, represents baseline

Media

Blue (#0078D4)

Professional, traditional for marketing

Digital

Teal (#00B7C3)

Modern, distinguishes from traditional

Price

Red (#E81123)

Alert color, represents impact

Promotions

Orange (#FF8C00)

Attention-grabbing, special events

Seasonality

Green (#107C10)

Natural cycles, recurring patterns

External

Purple (#5C2D91)

Distinct, outside control

Color Selection Principles:

1. Distinct Colors

  • Choose colors that are easily distinguishable

  • Avoid similar shades (e.g., two blues)

  • Test visibility in charts

2. Consistent Meaning

  • Use same colors across all models

  • Match company branding when appropriate

  • Align with industry conventions

3. Accessibility

  • Consider colorblind-friendly palettes

  • Ensure sufficient contrast

  • Avoid red-green combinations alone

4. Professional Appearance

  • Use muted, professional tones

  • Avoid overly bright/neon colors

  • Match corporate style guides

Color Persistence

Colors are saved with the model:

  • Consistency: Same colors used in all decomposition charts

  • Saved: Colors persist across sessions

  • Editable: Can be changed anytime

  • Exported: Colors included in Excel exports

Adjustment Parameters

Understanding Adjustments

Adjustments modify how variable contributions are calculated in decomposition:

None (Default):

  • Use the variable contribution as calculated

  • Contribution = Coefficient × Value

  • Most common setting

Min Adjustment:

  • Subtract the minimum contribution across all time periods

  • Shows only incremental effects above the minimum

  • Formula: Adjusted = Original - Min(Original)

Max Adjustment:

  • Subtract the maximum contribution across all time periods

  • Rarely used

  • Formula: Adjusted = Original - Max(Original)

When to Use Adjustments

None Adjustment (Recommended for Most Variables):

  • Default setting for all variables

  • Shows actual contribution values

  • Maintains interpretability

Min Adjustment Use Cases:

1. Baseline Removal

  • When you want to show only incremental effects

  • For variables with consistent baseline contribution

  • To highlight variability rather than absolute levels

2. Comparing Variable Contributions

  • Makes it easier to compare variables with different baselines

  • Focuses on changes rather than levels

Example:

TV Original Contribution: [100, 120, 110, 140, 130]
TV Min Adjusted: [0, 20, 10, 40, 30]
Interpretation: Shows incremental TV impact above minimum

Max Adjustment:

  • Rarely used in practice

  • Special analytical scenarios

  • Typically not needed for standard MMM

Adjustment Impact on Base Group

Important: When adjustments are applied, the subtracted values are added to the Base group to maintain the total predicted value.

Example:

  • If Min adjustment subtracts 100 from TV each period

  • Base group contribution increases by 100 each period

  • Total contribution remains unchanged

Saving Your Configuration

Save Button:

  • Click "Save Groups" when finished

  • Saves group assignments for all variables

  • Saves color selections

  • Saves adjustment settings

Confirmation:

  • Success message appears when saved

  • Configuration persists across sessions

  • Used automatically in decomposition analysis

When to Save:

  • After assigning all groups

  • After choosing colors

  • After setting adjustments

  • Before running decomposition

Workflow Summary

Follow this workflow to set up contribution groups:

  1. Select Model: Choose fitted model from dropdown

  2. Review Variables: Examine all variables and their properties

  3. Assign Groups: Use dropdowns or multi-select to categorize variables

  4. Choose Colors: Select distinct, professional colors for each group

  5. Set Adjustments: Apply adjustments if needed (usually keep as None)

  6. Save Configuration: Click "Save Groups" to persist settings

  7. Run Decomposition: Navigate to Decomposition page to analyze

Tips for Success

Start Simple:

  • Begin with broad categories (Media, Price, Seasonality)

  • Refine grouping as analysis needs evolve

  • Don't overthink initial grouping

Test and Iterate:

  • Run decomposition to see if groups make sense

  • Adjust grouping based on chart readability

  • Refine until insights are clear

Document Decisions:

  • Keep notes on grouping logic

  • Explain choices to stakeholders

  • Maintain consistency across models

Consider Audience:

  • Group for your primary stakeholders

  • Use terminology they understand

  • Enable the conversations you need to have

Next Steps

After setting up contribution groups:

  • Navigate to Decomposition page

  • Run group-level decomposition analysis

  • Drill down into specific groups

  • Calculate ROI by group

  • Export results for reporting

PreviousTroubleshooting Failed TestsNextOrganizing Variables into Groups

Last updated