Predict the financial return of a sports sponsorship deal before you sign it.
This is an AI-powered calculator that predicts three things about any sports sponsorship deal:
- Media Value -- How much earned media exposure your dollars will generate
- Brand Lift -- How much brand awareness will increase among fans
- ROI Score -- An overall return-on-investment rating from 1 to 100
You enter the deal details (sport, market, spend, activation channels), and the model tells you what to expect -- along with a plain-English explanation of why.
Sports sponsorship is a $65B+ global market, and most deals are still evaluated with spreadsheets and gut instinct.
| Pain Point | Current Approach | This Tool |
|---|---|---|
| "Will this deal deliver ROI?" | Gut feel + past experience | ML prediction based on 5,000 deal patterns |
| "What's driving the value?" | Opaque valuation models | SHAP explainability shows exactly which factors matter |
| "How does this compare?" | Manual comp research | Instant comparable deal lookup across 8 leagues and 10 markets |
| "How long does analysis take?" | Days of spreadsheet work | Seconds -- real-time interactive predictions |
flowchart LR
A["<b>Configure Deal</b><br/>Sport, market, spend,<br/>channels, brand category"] --> B["<b>Feature Engineering</b><br/>Log transforms,<br/>interaction features,<br/>one-hot encoding"]
B --> C["<b>XGBoost Models</b><br/>3 trained regressors<br/>with regularization"]
C --> D["<b>Predictions</b><br/>Media value ratio<br/>Brand lift %<br/>ROI score"]
C --> E["<b>SHAP Explainability</b><br/>Feature-level waterfall<br/>showing what drives<br/>each prediction"]
C --> F["<b>Comparable Deals</b><br/>Similar historical<br/>deals for<br/>benchmarking"]
style A fill:#4A90D9,stroke:#2C5F8A,color:#fff
style B fill:#6C757D,stroke:#495057,color:#fff
style C fill:#E8833A,stroke:#C06A2B,color:#fff
style D fill:#28A745,stroke:#1E7E34,color:#fff
style E fill:#7B68EE,stroke:#5B48CE,color:#fff
style F fill:#17A2B8,stroke:#117A8B,color:#fff
The model was trained on 5,000 synthetic deals calibrated to IEG and Nielsen Sports benchmarks, then evaluated on a held-out 20% test set.
| Metric | Media Value Ratio | Brand Lift % | ROI Score | What It Means |
|---|---|---|---|---|
| R-squared | 0.686 | 0.718 | 0.784 | The model explains 69-78% of the variance in outcomes |
| MAE | 0.25 | 0.82 | 5.65 | Average prediction error is small relative to each scale |
| MAPE | 9.5% | 11.9% | 10.4% | Predictions are within ~10% of actual values on average |
| Feature | Description |
|---|---|
| Real-Time Predictions | Adjust any deal parameter and see results update instantly |
| SHAP Waterfall Charts | Visual explanation of which factors push ROI up or down |
| Comparable Deal Finder | Automatically surfaces similar historical deals for context |
| 8 Leagues Covered | NFL, NBA, MLB, MLS, NCAA, Tennis, Golf, Boxing/MMA |
| 10 Major Markets | LA, New York, Chicago, Miami, Dallas, SF, Boston, Atlanta, Phoenix, Denver |
| 6 Deal Types | Jersey patch, venue signage, broadcast, digital/social, experiential, naming rights |
| 5 Activation Channels | On-air, digital, social, experiential, digital out-of-home |
| Model Diagnostics | Under-the-hood panel with residual analysis and feature importance |
| Layer | Technology | Why It Was Chosen |
|---|---|---|
| ML Model | XGBoost | Industry-standard gradient boosting -- handles mixed feature types and captures nonlinear relationships |
| Explainability | SHAP | Game-theory-based feature attribution -- shows why the model made each prediction, not just what it predicted |
| Frontend | Streamlit | Turns Python into an interactive web app with zero frontend code |
| Data Processing | pandas, NumPy, scikit-learn | Feature engineering pipeline with log transforms, interaction features, and one-hot encoding |
| Visualization | matplotlib | Publication-quality charts for SHAP waterfalls, feature importance, and residual analysis |
# 1. Install dependencies
pip install -r requirements.txt
# 2. Run the full pipeline (generate data, train model, launch app)
bash run.sh
# 3. Or skip to the app if data/model already exist
bash run.sh --appThe app opens at http://localhost:8501.
flowchart TB
subgraph Data["Data Layer"]
G["generate_data.py<br/><i>5,000 synthetic deals</i>"]
CSV[("sponsorship_deals.csv")]
G --> CSV
end
subgraph Model["ML Layer"]
T["train.py<br/><i>XGBoost pipeline</i>"]
ART[("trained_model.joblib")]
MET[("metrics.json")]
T --> ART
T --> MET
end
subgraph App["Application Layer"]
ST["app.py<br/><i>Streamlit interface</i>"]
end
CSV --> T
ART --> ST
MET --> ST
CSV --> ST
style Data fill:#E8F4FD,stroke:#4A90D9,color:#333
style Model fill:#FFF3E0,stroke:#E8833A,color:#333
style App fill:#E8F5E9,stroke:#28A745,color:#333
sponsorship-roi-calculator/
├── app.py # Streamlit web application
├── data/
│ ├── generate_data.py # Synthetic data generator (5,000 deals)
│ └── sponsorship_deals.csv # Generated dataset
├── model/
│ ├── train.py # XGBoost training pipeline
│ ├── trained_model.joblib # Saved model artifacts
│ └── metrics.json # Evaluation metrics
├── run.sh # One-command pipeline script
├── requirements.txt # Python dependencies
└── README.md
When you configure a deal, the calculator returns three prediction cards plus an explanation.
Example: NFL jersey patch deal in New York, $5M/year, 3-year term
| Prediction | Value | Interpretation |
|---|---|---|
| Media Value Ratio | 3.1x | Every $1 spent generates ~$3.10 in earned media value |
| Brand Lift | 7.2% | Expected brand awareness increase among the target audience |
| ROI Score | 72 / 100 | Excellent -- top quartile among comparable deals |
The SHAP waterfall chart then shows exactly why -- for instance, that NFL in New York drives the prediction up, while a shorter deal length or missing activation channels would pull it down.
For partnership sellers: Stop guessing. Walk into a pitch with AI-backed projections that show a prospect exactly what their deal will deliver -- media value, brand lift, and ROI -- before they commit a dollar.
For brand buyers: Compare deals across leagues, markets, and activation strategies using data, not decks. Know which deal structure maximizes your return before the negotiation starts.
For leadership: Replace subjective deal scoring with a quantitative framework. Evaluate pipeline quality with consistent, explainable metrics instead of one-off analyses.
CJ Fleming -- Revenue leader with 15+ years in media partnerships and a Columbia University AI certification. This project demonstrates what happens when deep sales domain expertise meets machine learning engineering: tools that solve real business problems, not just academic exercises.
The sponsorship economics in this model -- deal structures, media multipliers, activation strategies, league-by-market dynamics -- come from years of negotiating these deals firsthand.
