MMM Budget Simulator

❔Business Problem

Marketers often struggle to allocate budgets effectively across digital channels. This project creates a tool that empowers marketers to simulate real-world scenarios for a small-to-medium e-commerce brand and understand how shifting spend across channels impacts sales.

---

📨Executive Summary

• Built a Media Mix Model (MMM) using weekly digital marketing spend and sales.
• Applied adstock to capture the carry-over effects of advertising.
• Fitted an Ordinary Least Squares (OLS) regression to estimate each channel’s contribution to sales.
• Produced coefficients that quantify sales lift per channel.
• Developed a Power BI budget simulator to test “what-if” scenarios and compare baseline vs adjusted sales predictions.

---

🧑🏻‍🍳Project Summary

This project delivers a budget simulator that helps marketers understand the potential lift in sales when reallocating spend across digital channels: Display, Email, Influencer, Search, and Social.

---

🗂️Repository Contents

1. Python Scripts

• MMM_Simulated_Data → generates simulated weekly spend + sales data.
• MMM_Budget_Simulator_V4 → applies adstock, fits the OLS regression, and outputs results for Power BI.

2. Excel Files

• MMM_Simulated_Data.xlsx → simulated dataset.
• MMM_Data_Dictionary_V2.xlsx → data dictionary for all fields.
• mmm_budget_sim_results.xlsx → baseline vs scenario predicted sales.
• mmm_sales_coefficients.xlsx → regression coefficients.
• mmm_sales_features_baseline.xlsx → baseline features for Power BI.

3. Power BI Tool

• MMM_Simulator_V3.pbix → interactive dashboard for exploring budget scenarios.

---

🧬Replication Instructions

1. Generate simulated data
   • Run MMM_Simulated_Data.py (or use the provided file MMM_Simulated_Data.xlsx).
2. Run the budget simulator script
   • Run MMM_Budget_Simulator_V4.py (or use the provided outputs).
   • This generates:
     • mmm_budget_sim_results.xlsx
     • mmm_sales_coefficients.xlsx
     • mmm_sales_features_baseline.xlsx
3. Open the Power BI workbook (MMM_Simulator_V3.pbix)
   • Requires Power BI Desktop.
   • Inputs:
     • Top-left slicer → set your budget (up to $10,000 in this version).
     • Left-side sliders → allocate percentages of the budget to each channel.
       • Percentages will normalize to 100% automatically.
   • Outputs:
     • A summary statement of the scenario’s predicted impact.
     • A channel-level breakdown table.
     • A waterfall chart showing sales lift or decline.
     • A cluster chart comparing actual vs scenario spend per channel.

---

📊Using Your Own Data

1. Match your dataset to the same structure as MMM_Simulated_Data.xlsx.
2. Run MMM_Budget_Simulator_V4 on your data — this will generate the same three output files (mmm_budget_sim_results, mmm_sales_coefficients, mmm_sales_features_baseline).
3. Plug these outputs into the Power BI simulator (MMM_Simulator_V3.pbix).
   • Adjust the budget slicer max to reflect your business’s maximum budget.

---

🖋️Methodology

This project builds a Media Mix Model (MMM) to estimate the impact of different marketing channels on weekly sales and to simulate “what-if” budget reallocations.

1. Data Preparation

• Weekly spend data aggregated by channel.
• Each channel’s spend was adstocked to capture carry-over effects.
• Non-marketing controls (e.g., price index, inventory, promotions, holidays) were included to isolate true marketing effects.

2. OLS Regression

We assume weekly sales can be explained as:

$$ y=β0+β1x1+β2x2+⋯+ϵ $$

• y = actual sales
• β0 = intercept (baseline sales if spend = 0)
• βi = coefficient for channel or control variable
• xi = adstocked spend or control variable
• ϵ = residual (unexplained variation)

OLS estimates the coefficients by minimizing the squared difference between actual and predicted sales.

3. Predictive Equation

Once fitted, the model produces estimated coefficients:

$$ \hat{y}=β^0+β^1x1+β^2x2+… $$

• $\hat{y}$​ = predicted sales (no residual term — purely systematic, explainable part).

4. Budget Simulator

• Apply multipliers to channel spends.
• Recalculate predicted sales under new scenarios.
• Compare baseline vs scenario predictions to measure sales lift/decline.

---

💡Results / Insights

This project uses simulated data. In one scenario with a $5,000 budget:

• +15% Social and +10% Search
• –10% Email and –15% Display

Produced an estimated +75.2% lift in predicted sales.

---

👷🏻Future Improvements

• Incorporate long-term brand-building metrics (e.g., organic social, YouTube).
• Experiment with alternative models (Ridge/Lasso regression, Bayesian MMM).
• Add confidence intervals to quantify uncertainty in predictions.

---

🔑 Why This Project Matters

• Demonstrates the end-to-end workflow: simulation → adstock → OLS regression → coefficients → Power BI dashboard.
• Bridges data science and business decision-making: marketers can experiment with real trade-offs in channel allocation.
• Recruiters/hiring managers can quickly see both the technical rigor and the practical business application of the project.

---

🤖 Use of AI in This Project

This project was conceived, scoped, and designed entirely by me:

• Defined the business problem.
• Designed the methodology and simulation approach.
• Built the Power BI simulator for budget allocation testing.

I also used AI as a coding assistant and learning partner:

• Helped accelerate Python scripting for data simulation, adstock transformations, and OLS modeling.
• Provided explanations of statistical concepts (e.g., adstock, intercepts, coefficients) to speed up my learning.
• Assisted with debugging and documentation formatting.

All product thinking, methodology choices, and dashboard design were my own. I designed the dish — what to bake and how to serve it — but AI was like a recipe book: offering techniques and instructions I could adapt to my own kitchen.