Description
Multi-touch attribution: MMM regression, Shapley values, saturation curves, reallocation with expected lift.
Most marketing allocates by historical share or last-click (overvalues final touchpoint, undervalues funnel top). This framework builds defensible allocation. 1. CHANNEL INVENTORY: List every channel with spend, impressions, clicks, conversions, last-click %. Flag >20% spend on last-click-only (likely undervalued). 2. MMM REGRESSION: conversions = β₀ + β₁·TV + β₂·Search + β₃·Social + β₄·Email + ε. Which β has highest ROI? Lowest CI? Negative βs suggest cannibalization—remove and re-run. 3. SHAPLEY VALUE: Average each channel's marginal contribution across all possible subsets. Compare last-click vs first-click vs Shapley. LinkedIn typically gains most under Shapley (contributes to early funnel even without final click). 4. SATURATION CURVE: For highest-spend channel, plot spend × conversions with log trend: conversions ≈ a×ln(spend)+b. Saturation point where derivative (a/spend) = 1. Reallocate excess to next-best marginal return channel. OUTPUT: Attribution comparison table (last-click, first-click, Shapley), reallocation with expected conversion lift ±CI.
No comments yet. Be the first!