Creating New Models

Step-by-Step Guide to Building Your First Marketing Mix Model

Overview

Creating a new model in MixModeler is the starting point of your MMM journey. A model begins with just a dependent variable (KPI) and a constant term, then you add independent variables (marketing channels, control factors) to explain variations in your KPI.

What happens when you create a model:

1. You select a KPI (dependent variable)

2. Model initializes with constant term only (baseline)

3. Empty model with R² ≈ 0%

4. Ready for you to add variables in Model Builder

Prerequisites

Before Creating a Model

✅ Data uploaded: Your Excel file must be successfully uploaded with date column and numeric variables

✅ Data validated: At least 26 observations (weeks), no missing values in critical columns

✅ KPI identified: Know which variable you want to predict (Sales, Revenue, Conversions, Brand Awareness, etc.)

✅ Subscription active: Free tier allows 2 models, Professional/Enterprise allow unlimited

Step-by-Step: Creating Your First Model

Step 1: Navigate to Model Library

Path: Main Menu → Model Library

The Model Library page displays all existing models (if any) and the Create New Model button.

Step 2: Click "Create New Model"

Click the "+ Create New Model" button in the top-left area.

A creation dialog or section appears with two required fields:

• Model Name

• KPI Variable

Step 3: Enter Model Name

What makes a good model name:

✅ Descriptive:

• Includes KPI: Sales_Model, Revenue_MMM

• Includes time period: Q4_2024, Full_Year_2024

• Includes focus: Digital_Focus, TV_Heavy

✅ Consistent:

• Use underscores or hyphens consistently

• Capitalize consistently

• Follow your naming convention

✅ Unique:

• Each model needs a unique name

• Cannot duplicate existing model names

Examples of good names:

• Sales_MMM_Q1_2024

• Revenue_Model_Full_Year

• Conversions_Digital_Campaign

• Brand_Awareness_National_TV

• Leads_Model_v1

Examples of poor names:

• Model1 (not descriptive)

• Test (too generic)

• Untitled (meaningless)

• New Model (default, not specific)

Step 4: Select KPI Variable

KPI (Key Performance Indicator) is your dependent variable - what you want to predict or explain.

How to select:

MixModeler shows a searchable dropdown of all numeric variables from your uploaded data.

For large datasets (100+ variables):

• Use the search box to filter by typing variable name

• Dropdown filters in real-time

• Much faster than scrolling

Common KPI variables:

• Sales: Revenue, Units_Sold, GMV

• Conversions: Leads, Sign_Ups, Purchases

• Traffic: Website_Visits, App_Opens

• Brand: Brand_Awareness_Score, Consideration_Index

• Engagement: Video_Views, Time_On_Site

What makes a good KPI:

✅ Business-relevant: Tied to company goals ✅ Numeric: Continuous or count variable ✅ Sufficient variation: Not constant or near-constant ✅ Available consistently: Present in all time periods ✅ Reliable: Accurately measured

❌ Avoid as KPI:

• Binary variables (0/1)

• Categorical variables (text)

• Marketing spend variables (these are independent variables)

• Variables with too many zeros

Step 5: Click "Create Model"

After entering name and selecting KPI, click "Create Model".

What happens:

1. Model object created in backend

2. Initial regression run with constant term only

3. Model appears in Model Library table

4. Shows R² (will be very low, ~0-5%)

5. Shows 0 variables (no independent variables yet)

Step 6: Add Variables in Model Builder

Your model now exists but is empty (only constant term). Next steps:

1. Click on your new model name in the table

2. Navigate to Model Builder

3. Add independent variables (marketing channels, control factors)

4. Run model to see coefficients and R²

5. Iterate until satisfied

Understanding the Initial Model

What is the "Constant Term"?

When you create a model, it starts with just the constant (intercept):

Mathematical representation: KPI = β₀ + ε

Where:

• KPI = your dependent variable

• β₀ = constant term (intercept)

• ε = error term

Business interpretation: The constant represents baseline KPI level when all other factors are zero.

Example: If KPI is Sales and constant = 50,000, this means baseline sales with no marketing is $50,000 per week.

Why R² is ~0% Initially

R² measures explained variance. With only a constant term:

• Model predicts the same value (mean of KPI) for every observation

• Doesn't explain variation in KPI

• R² ≈ 0%

This is normal and expected. R² will increase as you add meaningful variables.

Model Creation Options

Standard Creation

Process described above:

• Manual name entry

• KPI selection from dropdown

• One model at a time

Best for: Most use cases, deliberate model building

Rapid Creation (Multiple Models)

When you need several models quickly:

Create models with different KPIs efficiently:

1. Create Sales_Model with Sales KPI

2. Clone and rename to Revenue_Model

3. Go to Model Builder, change dependent variable... (Not directly supported - requires recreation)

Note: Currently, changing KPI requires creating a new model. Cannot change KPI of existing model.

Common Model Types

Full-Attribution Model

Purpose: Explain overall KPI with all available marketing and control factors

Structure:

• KPI: Sales, Revenue, Conversions

• Variables: All marketing channels + seasonality + trends + controls

• Time period: Full dataset

Use when: You want comprehensive attribution across all channels

Channel-Specific Model

Purpose: Deep-dive into specific channel effectiveness

Structure:

• KPI: Sales

• Variables: Only digital channels (Search, Display, Social) + necessary controls

• Time period: Full or filtered

Use when: Optimizing budget within a specific channel category

Seasonal Model

Purpose: Understand seasonal patterns and effects

Structure:

• KPI: Sales

• Variables: Marketing channels + seasonal indicators (holiday flags, month dummies)

• Time period: Full year to capture seasonality

Use when: Planning seasonal campaigns or budgeting

Campaign-Specific Model

Purpose: Evaluate specific campaign impact

Structure:

• KPI: Sales or campaign-specific metric

• Variables: Campaign channels + baseline + controls

• Time period: Campaign period only (using observation filtering)

Use when: Measuring ROI of specific initiative

Best Practices

Model Naming Strategy

Use hierarchical naming:

[PROJECT]_[KPI]_[FOCUS]_[VERSION]

Examples:

• ClientA_Sales_Digital_v1

• Q4Campaign_Conversions_TV_v2

• Brand_Awareness_National_Final

Benefits:

• Easy sorting and grouping

• Clear purpose identification

• Version tracking

• Project organization

KPI Selection Guidelines

Choose KPI based on business objective:

If goal is revenue growth: KPI = Revenue, Sales, GMV

If goal is customer acquisition: KPI = New_Customers, Sign_Ups, Leads

If goal is engagement: KPI = Website_Visits, Time_On_Site, App_Opens

If goal is brand building: KPI = Brand_Awareness_Score, Consideration_Index

Avoid:

• Using intermediate metrics when final outcomes available

• KPIs with insufficient variation

• KPIs that are sums/aggregates of your independent variables

Model Organization

Create focused models:

• One primary model per KPI

• Experimental variants clearly labeled

• Archive old/superseded models

Version control through cloning:

• Start with v1

• Clone for major changes → v2

• Keep best performing version

• Delete failed experiments

After Model Creation

Immediate Next Steps

1. Navigate to Model Builder

2. Add your first variable (usually strongest marketing channel)

3. Run model to see initial results

4. Check diagnostics (even with one variable)

5. Iterate: Add more variables, test transformations

Building Model Incrementally

Recommended approach:

Phase 1: Core variables (R² target: 40-60%)

• Add strongest marketing channels first

• Include trend/seasonality if obvious

• Get basic model working

Phase 2: Additional marketing (R² target: 60-75%)

• Add remaining marketing channels

• Apply adstock transformations

• Test saturation curves

Phase 3: Control variables (R² target: 70-85%)

• Add price, promotions

• Add external factors (weather, events)

• Add competitive activity if available

Phase 4: Refinement (R² target: 75-90%)

• Optimize transformations

• Test variable interactions

• Remove non-significant variables

• Validate diagnostics

Troubleshooting

"KPI variable not found"

Cause: Selected variable doesn't exist in data

Solution:

• Refresh page

• Re-upload data

• Verify variable name matches data exactly

"Model name already exists"

Cause: Duplicate model name

Solution:

• Choose a unique name

• Add version number or date

• Delete old model if no longer needed

"Insufficient observations"

Cause: KPI variable has too many missing values or data has <26 observations

Solution:

• Check data quality

• Fill missing values

• Upload more data if needed

"Model creation failed"

Generic error - possible causes:

• Data not properly loaded

• KPI variable is non-numeric

• System error

Solutions:

• Reload data

• Verify KPI is numeric

• Try again

• Contact support if persists

Key Takeaways

• Models start with just KPI + constant term (R² ≈ 0%)

• Use descriptive, unique names following consistent conventions

• KPI selection is permanent - choose carefully

• Initial model is a starting point - add variables in Model Builder

• Build models incrementally: core → marketing → controls → refinement

• One model per primary KPI/objective

• Clone models for experimentation

• Clean up failed experiments regularly