Multichannel Marketing Data Correlation with Revenue Attribution

Build a correlation analysis and multi-touch attribution model for [COMPANY NAME], which invests $[AMOUNT]/month in [CHANNEL LIST].\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n**Required Data:**\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n- Conversions table with: date, last-click channel, revenue, customer_id\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n- Touchpoints table: customer_id, channel, interaction_date, interaction_type\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n- Analysis period: [NUMBER] months\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n- Attribution window: [NUMBER] days\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n**Part 1 — Channel Correlation Analysis:**\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\`\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\`\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\`python\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\nimport pandas as pd\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\nimport numpy as np\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\nimport matplotlib.pyplot as plt\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\nfrom scipy import stats\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n# Group spending and revenue by channel weekly\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\nweekly_df = df.groupby(['week', 'channel']).agg(\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n    spend=('spend', 'sum'),\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n    conversions=('conversion', 'sum'),\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n    revenue=('revenue', 'sum')\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n).reset_index()\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n# Pearson correlation: spend per channel vs total revenue\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\nchannels = ['google_ads', 'meta_ads', 'email', 'organic']\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\nfor channel in channels:\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n    channel_df = weekly_df[weekly_df['channel'] == channel]\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n    r, p_value = stats.pearsonr(channel_df['spend'], channel_df['revenue'])\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n    lag_correlations = []\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n    for lag in range(0, 5):\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n        r_lag, _ = stats.pearsonr(\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n            channel_df['spend'], channel_df['revenue'].shift(-lag).fillna(0)\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n        )\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n        lag_correlations.append((lag, r_lag))\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n    print(f'{channel}: r={r:.3f}, p={p_value:.4f}, best_lag={max(lag_correlations, key=lambda x: x[1])}')\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\`\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\`\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\`\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n**Part 2 — Attribution Models:**\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n**A) Last-click (baseline for comparison):**\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n- All revenue attributed to the last channel before conversion\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n**B) Linear (equal):**\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n- Revenue divided equally across all touchpoints\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n**C) Time-decay (recency):**\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n- Touchpoints closer to conversion receive more credit\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n- Weight = e^(-lambda * days_until_conversion), lambda = 0.7\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n**D) Position (U-shape):**\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n- First touch: 40% | Last touch: 40% | Middle: 20% split\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n**E) Data-driven (Shapley Values — most accurate):**\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\`\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\`\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\`python\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\nfrom itertools import combinations\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\ndef shapley_attribution(touchpoints_df, revenue_df):\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\n    \\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\