Cross-Channel Multi-Touch Attribution Model Specification

Role: Quantitative Growth Marketing Architect

Context

• Active Paid Channels: {{tracked_ad_channels}}
• Lookback Conversion Window: {{conversion_window_days}}
• Ingestion Source: {{touchpoint_data_source}}
• Target Acquisition Cost: {{baseline_cac_target}}
• Deprecation Cutoff Threshold: {{underperforming_channel_threshold}}
• Total Portfolio Ad Spend: {{total_ad_spend}}

Task

Design a data-driven multi-touch attribution specification that replaces heuristic models with a Markov chain removal-effect and Shapley-value framework to quantify the true marginal impact of {{tracked_ad_channels}}.

Method

1. Establish state transition rules and path sequence mapping across {{tracked_ad_channels}} over {{conversion_window_days}}.
2. Define the stochastic matrix calculation protocol for multi-touch path journeys ingested from {{touchpoint_data_source}}.
3. Formulate the removal effect algorithm to compute individual channel marginal contribution weights.
4. Integrate cooperative game theory (Shapley value) to evaluate position-weighted interaction effects.
5. Model cost-per-attributed-acquisition curves against {{baseline_cac_target}} and total spend {{total_ad_spend}}.
6. Construct the programmatic threshold logic for identifying channels below {{underperforming_channel_threshold}}.
7. Detail data hygiene requirements, including identity resolution rules and conversion deduplication.
8. Produce an optimization rebalancing formula that shifts media spend dynamically toward high-removal-effect channels.

Constraints

• MUST NOT include single-touch heuristic models (first-click, last-click, linear) as primary allocation drivers.
• MUST provide explicit matrix transition formulas and removal effect equations.
• Algorithmic definitions MUST account for user privacy constraints and cookie loss.
• Budget reallocation outputs MUST not exceed {{total_ad_spend}}.

Output format

• Model Architecture Overview: High-level mathematical logic and state definitions.
• Transition Matrix & Removal Spec: Exact formulas for Markov chain transition and Shapley distribution.
• Data Pipeline & Deduplication Requirements: Ingestion schemas, lookback limits, and join keys.
• Budget Optimization Rules: Decision matrix for capital shift based on attributed marginal ROAS.

Self-review

• Verify that transition probability formulas sum to 1.0 across absorbing and non-absorbing states.
• Ensure lookback window constraints properly prevent cross-cycle path distortion.
• Validate that attribution weighting directly ties back to {{baseline_cac_target}} thresholds.