Predict campaign performance. Optimize your media mix. Maximize ROAS.
An AI-powered tool that takes the guesswork out of media planning. Tell it your budget and goals — it tells you exactly where to spend and what to expect.
In plain English: you give it a campaign budget, and it tells you the smartest way to split that money across six media channels to get the best return.
It does three things:
-
Predicts performance — Enter your campaign details (industry, budget, audience, timing) and it forecasts impressions, CPM, click-through rate, conversions, and ROAS before you spend a dollar.
-
Analyzes your portfolio — Upload past campaign data or explore the built-in dataset. See which campaigns crushed it and which underperformed, with interactive charts that make patterns obvious.
-
Optimizes your media mix — Give it a total budget and any constraints ("at least 20% must go to Linear TV") and it mathematically finds the channel split that maximizes your return on ad spend.
Media sales teams face the same question on every deal:
"We have $2M to spend across six channels. Where should the money go?"
Today, that answer usually comes from gut instinct, last quarter's plan, or a spreadsheet someone built three years ago. The result? Suboptimal allocation that leaves money on the table.
This dashboard replaces guesswork with machine learning. It trains on thousands of campaign outcomes to learn what actually drives performance — then uses constrained optimization to find the best possible media mix for any scenario.
The optimizer consistently finds allocations that deliver 49% better ROAS than splitting budget equally across channels.
flowchart LR
A["📋 You Enter\nBudget + Goals"] --> B["🤖 ML Model\nPredicts KPIs"]
B --> C["🔄 Optimizer\nTests Allocations"]
C --> D["✅ Best Mix\nFound"]
D --> E["💡 Recommendations\nChannel Shifts"]
style A fill:#4A90D9,stroke:#2C5F8A,color:#fff
style B fill:#7B68EE,stroke:#5A4CB5,color:#fff
style C fill:#F5A623,stroke:#C4841A,color:#fff
style D fill:#50C878,stroke:#3A9A5B,color:#fff
style E fill:#FF6B6B,stroke:#CC5555,color:#fff
Step by step:
-
You configure a campaign — Pick an industry, set a budget ($25K–$5M), choose your audience segment and market tier, then set channel allocation sliders for all six channels.
-
The model predicts outcomes — A Random Forest model (trained on 3,000 campaign outcomes) predicts five KPIs simultaneously: impressions, CPM, CTR, conversion rate, and ROAS.
-
The optimizer searches for the best split — Using constrained mathematical optimization (SLSQP), it tests thousands of possible channel allocations to find the one that maximizes predicted ROAS while respecting your constraints.
-
You get specific recommendations — Not just "spend more on digital" — exact percentages and dollar amounts for each channel, with before/after comparisons showing the projected improvement.
| Metric | Value | What It Means |
|---|---|---|
| Model Accuracy (R²) | 0.85 | The model explains 85% of the variance in ROAS — strong predictive power |
| ROAS Improvement | +49% | Optimized mix vs. equal allocation across channels |
| Channels Covered | 6 | Linear TV, Digital Video, Social, Podcast, DOOH, Programmatic Display |
| Industries | 8 | Auto, CPG, Finance, Healthcare, Retail, Tech, Telecom, Travel |
| Audience Segments | 4 | Adults 18-34, Adults 25-54, Adults 35-64, HHI $100K+ |
| Market Tiers | 4 | Top 10 DMA, Top 25, Top 50, National |
| Predictions Per Campaign | 5 | Impressions, CPM, CTR, Conversion Rate, ROAS |
| Training Dataset | 3,000 | Synthetic campaigns with realistic performance correlations |
- Configure any campaign scenario with interactive sliders
- Real-time ML predictions update as you adjust channel mix
- AI recommendation engine suggests specific channel reallocation moves
- Metric cards show predicted impressions, CPM, CTR, conversions, and ROAS
- Upload your own campaign CSV or explore the built-in sample dataset
- Budget vs. ROAS scatter plots, industry box plots, channel efficiency correlations
- Top and bottom performer breakdowns
- Interactive Plotly charts — hover, zoom, filter
- Set a fixed budget and optional per-channel minimum constraints
- Optimizer finds the mathematically best channel allocation
- Before/after comparison with delta metrics for every KPI
- Clear dollar-amount recommendations for each channel
flowchart TB
subgraph Data["Data Layer"]
GEN["generate_data.py\n3,000 synthetic campaigns"]
CSV[("campaigns.csv")]
GEN --> CSV
end
subgraph Model["ML Layer"]
TRAIN["train.py\nRandom Forest Pipeline"]
JOBLIB[("model.joblib\n+ scaler + features")]
CSV --> TRAIN
TRAIN --> JOBLIB
end
subgraph App["Application Layer"]
STREAMLIT["app.py — Streamlit Dashboard"]
TAB1["Campaign Planner\nPredict KPIs"]
TAB2["Portfolio Analyzer\nVisualize History"]
TAB3["Media Mix Optimizer\nMaximize ROAS"]
STREAMLIT --> TAB1
STREAMLIT --> TAB2
STREAMLIT --> TAB3
end
JOBLIB --> STREAMLIT
CSV --> TAB2
style Data fill:#E8F4FD,stroke:#4A90D9,color:#333
style Model fill:#F0E8FD,stroke:#7B68EE,color:#333
style App fill:#E8FDE8,stroke:#50C878,color:#333
Scenario: $1.5M Auto campaign targeting Adults 25-54 in Top 25 DMAs
| Channel | Equal Split | Optimized Mix | Change |
|---|---|---|---|
| Linear TV | 16.7% ($250K) | 28.0% ($420K) | ▲ +11.3% |
| Digital Video | 16.7% ($250K) | 24.5% ($368K) | ▲ +7.8% |
| Social | 16.7% ($250K) | 22.0% ($330K) | ▲ +5.3% |
| Podcast | 16.7% ($250K) | 12.5% ($188K) | ▼ -4.2% |
| DOOH | 16.7% ($250K) | 8.0% ($120K) | ▼ -8.7% |
| Programmatic Display | 16.7% ($250K) | 5.0% ($75K) | ▼ -11.7% |
| KPI | Equal Split | Optimized | Improvement |
|---|---|---|---|
| ROAS | 3.2x | 4.8x | +49% |
| Predicted Impressions | 18.4M | 22.1M | +20% |
| Predicted CTR | 1.8% | 2.4% | +33% |
The optimizer shifts budget toward channels with the strongest predicted performance for this specific campaign profile — not a one-size-fits-all answer, but a tailored recommendation.
For Account Executives: Walk into a pitch with data-backed media plans instead of recycled decks. Show clients exactly how their budget should be allocated and what performance to expect.
For Sales Managers: Benchmark your team's campaign recommendations against the optimizer. Identify which reps are consistently over-allocating to underperforming channels.
For VPs of Sales: Make strategic channel investment decisions grounded in predictive analytics. Know which verticals and audience segments have the highest ROAS potential before setting quotas.
For Revenue Operations: Automate the "what should we recommend?" question. Instead of hours in spreadsheets, get optimized media plans in seconds.
# One-command setup: creates venv, installs deps, generates data, trains model, launches app
chmod +x run.sh && ./run.shOr step by step:
python3 -m venv venv && source venv/bin/activate
pip install -r requirements.txt
python data/generate_data.py # Generate 3,000 synthetic campaigns
python model/train.py # Train Random Forest model
streamlit run app.py # Launch dashboard at localhost:8501media-sales-dashboard/
├── app.py # Streamlit application (3 tabs)
├── data/
│ ├── generate_data.py # Synthetic campaign data generator (~3,000 records)
│ ├── campaigns.csv # Generated dataset
│ └── __init__.py
├── model/
│ ├── train.py # Random Forest training pipeline
│ ├── model.joblib # Trained model artifact
│ ├── scaler.joblib # Fitted StandardScaler
│ ├── feature_columns.json # Feature column order for prediction
│ └── __init__.py
├── requirements.txt
├── run.sh # One-command setup & launch script
└── README.md
| Component | Technology | Purpose |
|---|---|---|
| Frontend | Streamlit | Interactive dashboard UI |
| Visualization | Plotly | Interactive charts and graphs |
| ML Model | scikit-learn (Random Forest) | Multi-target KPI prediction |
| Optimization | SciPy (SLSQP) | Constrained media mix optimization |
| Data Processing | pandas, NumPy | Feature engineering and data manipulation |
CJ Fleming — Media sales leader with 15+ years of experience across linear TV, digital video, and emerging platforms. Columbia University AI certification. This project demonstrates the intersection of deep media sales domain expertise and applied machine learning.
This isn't a theoretical exercise — it's the tool I wish I'd had during a decade and a half of building media plans, negotiating upfronts, and optimizing cross-platform campaigns.
