MMM Workflow Guide

Overview

This guide outlines the complete workflow for building marketing mix models in MixModeler, from initial data preparation through final insights delivery. Following this structured approach ensures reliable results and efficient analysis.

The Complete MMM Workflow

Phase 1: Planning & Data Gathering (1-2 weeks)

Define Objectives

Clarify what you want to learn:

• Which channels drive the most incremental sales?

• How should we allocate next quarter's budget?

• What's the ROI of each marketing activity?

• How do channels interact or complement each other?

Identify KPI

Choose your dependent variable:

• Revenue (most common)

• Sales units

• Conversions

• Leads

• Store visits

Inventory Data Sources

List all available marketing and business data:

• Marketing spend by channel (weekly or monthly)

• Media impressions or GRPs

• Digital metrics (clicks, impressions)

• Promotional activities

• Pricing data

• Competitive activity

• External factors (weather, events, holidays)

Determine Time Granularity

• Weekly data (recommended): 52-156 weeks (1-3 years)

• Monthly data: 24-60 months (2-5 years)

• Daily data (advanced): Requires aggregation to weekly

Set Project Timeline

• Week 1-2: Data collection and preparation

• Week 3: Initial model building and testing

• Week 4: Model refinement and validation

• Week 5: Analysis and insights development

• Week 6: Presentation and recommendation

Phase 2: Data Preparation (3-5 days)

Collect and Consolidate

1. Export data from all source systems

2. Combine into single Excel file

3. Ensure consistent date formats

4. Align all data to same time periods

Clean and Validate

1. Check for missing values (fill or flag)

2. Identify and investigate outliers

3. Verify data ranges make sense

4. Ensure spend matches financial records

5. Validate KPI totals

Structure for MixModeler

1. First column: Date (YYYY-MM-DD format)

2. Second column: KPI variable

3. Remaining columns: Marketing variables

4. One row per time period

5. Column headers: Clear, descriptive names

Create Excel File

Save as .xlsx with:

• Clean column names (no special characters except underscore)

• Numeric values only (no text in data cells)

• No blank rows or columns

• No merged cells

• Single sheet with all data

Phase 3: Initial Upload & Exploration (1 day)

Upload Data

1. Open MixModeler

2. Navigate to Data Upload

3. Select your prepared Excel file

4. Review upload summary

5. Verify variable count and date range

Explore in Variable Charts

1. Go to Variable Charts

2. Plot KPI over time (check for trends, seasonality)

3. Plot each marketing variable (identify patterns)

4. Create scatter plots (KPI vs each channel)

5. Generate correlation heatmap

Document Initial Observations

Note in separate document:

• Seasonal patterns in KPI

• Obvious relationships between variables

• Potential data quality issues

• Variables that might need transformation

Phase 4: Variable Engineering (2-3 days)

Apply Transformations

1. Open Variable Workshop

2. For media variables: Consider adstock transformation

3. For spending: Consider log or saturation curves

4. For seasonality: Create month or quarter dummies

5. For campaigns: Split by date if effects differ by period

Create Adstock Variables

1. Select media variables (TV, Radio, Digital, etc.)

2. Apply adstock transformation

3. Start with 50% decay rate

4. Test multiple rates (30%, 50%, 70%) in Variable Testing

Build Interaction Variables

For suspected synergies:

1. Create multiply variables (TV × Digital, etc.)

2. Test significance in Variable Testing

3. Keep only meaningful interactions

Organize Contribution Groups

Assign variables to business-relevant categories:

• Base (intercept, trends)

• TV

• Digital

• Print

• Promotions

• Seasonality

• Other

Phase 5: Initial Model Building (1 day)

Start Simple

Build baseline model:

1. Go to Model Builder

2. Create new model: "Baseline_v1"

3. Select KPI

4. Add top 5-10 most important variables

5. Choose OLS (faster for initial testing)

6. Run model

Review Initial Results

Check:

• R²: Aim for >0.70 (good), >0.80 (excellent)

• Coefficients: Do signs make business sense?

• P-values: Are key variables significant?

• Overall model fit

Iterate Quickly

1. Add variables one at a time

2. Remove non-significant variables

3. Test different transformations

4. Document what works and what doesn't

Phase 6: Model Refinement (3-5 days)

Add Complexity Gradually

1. Start with baseline model

2. Add control variables (seasonality, trends)

3. Add adstock transformations

4. Test saturation curves for media

5. Explore interactions

Run Diagnostics

For each model candidate:

1. Navigate to Model Diagnostics

2. Run all standard tests

3. Check multicollinearity (VIF <5 ideal, <10 acceptable)

4. Review residual plots

5. Identify influential points

Address Issues

• High VIF: Remove correlated variables

• Autocorrelation: Add lagged KPI or trends

• Non-normal residuals: Transform KPI or add variables

• Heteroscedasticity: Consider log transformation

Compare Models

1. Build 3-5 candidate models

2. Compare R², AIC, BIC

3. Assess coefficient stability

4. Evaluate business reasonableness

5. Select best performing model

Phase 7: Advanced Testing (2-3 days)

Optimize Adstock

1. Go to Variable Testing

2. Test adstock rates: 10%, 30%, 50%, 70%, 90%

3. Select rate with highest t-statistic

4. Apply to final model

Test Saturation Curves

1. For high-spend channels

2. Test S-curve vs concave curve

3. Compare fit vs linear specification

4. Use curve if significantly better

Bayesian Analysis (Optional)

1. Clone best OLS model

2. Switch to Bayesian

3. Set weakly informative priors

4. Run MCMC (standard settings)

5. Check convergence diagnostics

6. Compare credible intervals to OLS confidence intervals

Phase 8: Final Validation (1-2 days)

Comprehensive Diagnostics

Run full diagnostic suite on final model:

• Normality tests: PASS

• Autocorrelation: PASS

• Heteroscedasticity: PASS

• Multicollinearity: VIF <10 for all variables

• Influential points: Investigate outliers

Generate Documentation

1. Export model to Excel

2. Generate PDF diagnostic reports

3. Create decomposition analysis

4. Document all transformations

5. Record model specifications

Validate Results

Sanity checks:

• Do coefficients have expected signs?

• Are magnitudes reasonable?

• Does seasonal pattern match business knowledge?

• Are top channels aligned with expectations?

• Can you explain any surprises?

Phase 9: Insights & Decomposition (2-3 days)

Run Decomposition

1. Set up contribution groups (if not done)

2. Run decomposition analysis

3. Review group contributions over time

4. Analyze variable-level attribution

5. Calculate channel ROI

Calculate ROI

For each channel:

ROI = (Coefficient × Average Spend) / Average KPI

Or use decomposition:

ROI = Average Contribution / Average Spend

Identify Opportunities

• Channels with high ROI: Increase investment

• Channels with low ROI: Reduce or optimize

• Underutilized channels: Test increased spend

• Saturated channels: Optimize or maintain

Develop Scenarios

Test budget reallocation:

1. Clone model

2. Adjust spend variables manually

3. Predict new KPI

4. Calculate impact

5. Compare scenarios

Phase 10: Reporting & Recommendations (2-3 days)

Prepare Stakeholder Materials

1. Create executive summary (1 page)

2. Build detailed presentation (10-15 slides)

3. Prepare supporting Excel exports

4. Generate diagnostic PDFs for appendix

Structure Presentation

• Slide 1: Executive summary

• Slide 2-3: Methodology overview

• Slide 4-6: Channel performance

• Slide 7-8: Attribution and trends

• Slide 9-10: Recommendations

• Slide 11+: Appendix (technical details)

Deliver Insights

Present to stakeholders:

• Start with key findings

• Show channel rankings

• Present budget recommendations

• Quantify expected impact

• Address questions and concerns

Phase 11: Implementation & Monitoring (Ongoing)

Implement Changes

1. Get approval for budget reallocation

2. Plan gradual implementation

3. Set success metrics

4. Define monitoring schedule

Track Results

• Weekly: Monitor KPI performance

• Monthly: Compare actual vs predicted

• Quarterly: Update model with new data

• Annually: Full model rebuild

Iterate and Improve

• Incorporate learnings from implementation

• Refine model as new data arrives

• Test new variables and channels

• Adjust for market changes

Workflow Tips by Experience Level

Beginners (First MMM Project)

Start Simple:

• Use 6-12 weeks of data initially

• Include only 5-10 main variables

• Use OLS before attempting Bayesian

• Focus on learning the tool

Learn Incrementally:

• Week 1: Data upload and exploration

• Week 2: Basic model building

• Week 3: Diagnostics and validation

• Week 4: Interpretation and reporting

Get Help:

• Follow Quick Start Guide

• Watch tutorial videos

• Review example models

• Ask questions early

Intermediate (2-5 MMM Projects)

Add Sophistication:

• Test adstock transformations

• Explore saturation curves

• Build interaction variables

• Compare multiple model specifications

Systematic Testing:

• Document transformation decisions

• Track model iterations

• Compare candidate models rigorously

• Validate assumptions thoroughly

Efficiency:

• Develop reusable Excel templates

• Create standard variable lists

• Build model comparison frameworks

• Streamline reporting process

Advanced (5+ MMM Projects)

Advanced Techniques:

• Bayesian modeling with informative priors

• Complex interaction terms

• Time-varying coefficients

• Multi-level models (if multiple markets)

Automation:

• Standardized data pipelines

• Automated diagnostic checking

• Templated reporting

• Version control for models

Experimentation:

• Test novel variable transformations

• Explore advanced saturation curves

• Implement custom testing procedures

• Push methodological boundaries

Common Workflow Variations

Quick Analysis (1 week)

For rapid insights:

1. Day 1: Data prep and upload

2. Day 2: Quick model building (OLS only)

3. Day 3: Basic diagnostics

4. Day 4-5: Insights and recommendations

Trade-offs: Less rigorous, fewer tests, simpler model

Deep Dive (6-8 weeks)

For thorough analysis:

• Weeks 1-2: Extensive data exploration

• Weeks 3-4: Multiple model specifications

• Weeks 5-6: Bayesian analysis and validation

• Weeks 7-8: Scenario planning and detailed recommendations

Benefits: Higher confidence, deeper insights, robust recommendations

Ongoing Program (Quarterly Updates)

For continuous MMM:

• Month 1: Data collection

• Month 2: Model update and validation

• Month 3: Analysis and recommendations

• Repeat each quarter

Advantages: Trends over time, adaptive to market changes, continuous optimization

Success Metrics

Model Quality:

• R² > 0.75

• All diagnostic tests passing

• VIF < 5 for key variables

• Coefficients make business sense

Process Efficiency:

• Data to insights: <4 weeks

• Model iterations: 5-10

• Stakeholder alignment achieved

• Recommendations implemented

Business Impact:

• Marketing ROI improvement

• Budget allocation optimization

• Clear channel performance insights

• Data-driven decision making

---

Next Steps: Review Data Preparation Tips for detailed data quality guidance, or explore Model Building Strategy for advanced modeling techniques.