🎯 Multi-Touch Attribution (MTA) Model

An end-to-end Multi-Touch Attribution pipeline implementing 7 attribution models — from simple rule-based approaches to data-driven Markov Chain and Shapley Value models — with a full MTA vs MMM reconciliation framework. Built on 50,000 synthetic user journeys across 6 marketing channels.

---

🎯 Business Questions

• Which marketing channels actually drive conversions — not just which ones get the last click?
• How does attribution change when we use data-driven vs rule-based methods?
• Where do MTA and MMM disagree, and what does that tell us about measurement gaps?
• What is the true ROI of each channel after removing last-touch bias?

---

📁 Project Structure

mta_model/
├── outputs/                        # Charts, CSVs, executive summary
├── src/
│   ├── data_prep.py                # 50,000 synthetic user journeys (6 channels)
│   ├── rule_based.py               # 5 rule-based models
│   ├── markov_chain.py             # Markov Chain removal effect model
│   ├── shapley.py                  # Shapley Value attribution (game theory)
│   ├── mmm_reconciliation.py       # MTA vs MMM reconciliation framework
│   └── visualizations.py          # 8 charts + summary dashboard
├── main.py                         # Run the full pipeline
├── requirements.txt
└── README.md

---

📐 Models Implemented

Rule-Based Models (5)

Model	Logic	Best For
Last Touch	100% credit to final touchpoint	Direct response; ad platform default
First Touch	100% credit to first touchpoint	Awareness and acquisition focus
Linear	Equal credit across all touches	Long complex sales cycles
Time Decay	More credit to recent touches	Short sales cycles
Position-Based	40% first / 40% last / 20% middle	Balanced acquisition + conversion

Data-Driven Models (2)

Markov Chain (Removal Effect)

• Builds a transition probability matrix from channel → channel → conversion
• For each channel, measures how much overall conversion rate drops when that channel is removed
• This "removal effect" is the channel's true causal contribution

Shapley Values (Game Theory)

• Treats the customer journey as a cooperative game where channels are players
• Computes the average marginal contribution of each channel across all possible channel orderings
• Satisfies 4 mathematical fairness axioms (efficiency, symmetry, dummy, additivity)
• Foundation of SHAP values used in modern ML explainability

MMM Reconciliation Framework

Compares MTA outputs against Marketing Mix Model (MMM) channel attribution to identify:

• Strong agreement (high confidence in both signals)
• MTA inflation (last-touch bias overstating lower-funnel channels)
• MMM premium (brand/awareness effects that MTA's digital tracking misses)

Decision Framework:
  MMM → Annual budget allocation, CFO conversations, offline channels
  MTA → Weekly campaign optimization, bid strategy, creative testing
  Reconciliation → Where signals diverge most = where to investigate

---

📈 Output Files

File	Description
01_journey_distribution.png	Journey length by conversion outcome
02_channel_positions.png	First touch, last touch, overall share by channel
03_rule_based_comparison.png	All 5 rule-based models side by side
04_transition_heatmap.png	Markov Chain transition probability matrix
05_data_driven_vs_rule_based.png	Markov + Shapley vs Last Touch baseline
06_mta_mmm_reconciliation.png	Channel share divergence between MTA and MMM
07_roi_comparison.png	MTA ROI vs MMM ROI per channel
08_summary_dashboard.png	Full summary dashboard
rule_based_results.csv	All rule-based model outputs
markov_results.csv	Markov Chain attribution
shapley_results.csv	Shapley Value attribution
mta_mmm_reconciliation.csv	Full reconciliation table
mta_executive_summary.txt	Business-ready summary with recommendations

---

🚀 Getting Started

git clone https://github.com/yourusername/mta_model.git
cd mta_model
pip install -r requirements.txt
python main.py

Note: The Shapley model evaluates 64 channel subsets across 50,000 users — allow 2-3 minutes for this step.

No data download required — data is generated synthetically with realistic multi-channel journey patterns.

---

🛠 Tech Stack

• Python 3.14+
• pandas / numpy — journey data processing
• scipy — statistical utilities
• matplotlib — all visualizations
• Pure Python — Markov Chain matrix operations and Shapley computation

---

💡 Key Concepts Demonstrated

• Multi-channel customer journey simulation
• Rule-based attribution (Last Touch, First Touch, Linear, Time Decay, Position-Based)
• Markov Chain transition matrices and removal effect calculation
• Shapley Value computation from cooperative game theory
• MTA vs MMM reconciliation framework
• Channel-level ROI measurement and comparison
• Attribution bias identification (last-touch inflation, brand effect gaps)
• Executive-ready measurement reporting

---

📚 Further Reading

• Shapley (1953) — A Value for n-Person Games — original game theory paper
• Markov Chain MTA — standard in production measurement at Google, Meta, Amazon
• MTA vs MMM reconciliation — growing best practice in sophisticated measurement teams

---

👤 Author

Built as part of a data science / marketing analytics portfolio.
Background: 15+ years in Marketing Analytics | SQL | Python | Statistical Modeling | MMM | MTA