Multi-Touch Attribution

A cooperative game theory approach that calculates the marginal contribution of each pricing touchpoint by averaging its impact across all possible sequences. The Shapley Value ensures fairness by satisfying axioms of efficiency, symmetry, and additivity.

• Computes the weighted average of a touchpoint's incremental lift when added to every possible subset of other touchpoints
• Eliminates positional bias inherent in heuristic models like first-touch or last-touch
• Computationally intensive for large touchpoint sets; often approximated via sampling methods

A probabilistic approach that models the customer journey as a state machine, where each touchpoint represents a state and transitions carry probabilities. The removal effect quantifies a channel's importance by measuring the drop in conversion probability when that state is eliminated.

• Captures sequential dependencies between touchpoints rather than treating them independently
• Naturally handles non-linear customer paths with loops and revisits
• Outputs a transition probability matrix that can be visualized as a directed graph

An algorithmic variant that assigns exponentially increasing weight to touchpoints closer to the conversion event. The decay rate is a tunable hyperparameter calibrated against historical conversion data.

• Uses a half-life parameter to control how quickly older touchpoints lose attribution credit
• Effective for short consideration cycles where recency strongly correlates with influence
• Simpler to implement than game-theoretic methods but introduces recency bias

The core output mechanism where a single conversion's value is partitioned across multiple touchpoints based on their calculated contribution weights. This contrasts with binary attribution where one channel receives 100% credit.

• Enables ROAS calculation at the individual touchpoint level rather than campaign level
• Supports budget optimization by revealing which mid-funnel interactions create downstream lift
• Requires robust identity resolution to stitch cross-device and cross-session journeys

The practice of validating MTA model outputs against randomized controlled trials that hold out specific touchpoints to measure true causal lift. This addresses the fundamental limitation that observational attribution confuses correlation with causation.

• Uses ghost ads or geo-matched control groups to establish counterfactual baselines
• Calibrates algorithmic weights against empirically measured incremental revenue
• Critical for distinguishing cannibalization from genuine contribution in promotional touchpoints

Advanced MTA implementations that optimize attribution weights against multiple business constraints simultaneously, not just conversion volume. Objectives include customer lifetime value preservation, margin thresholds, and budget pacing.

• Uses constrained optimization to prevent over-crediting high-discount touchpoints that erode margin
• Incorporates saturation curves to model diminishing returns from repeated exposures
• Balances short-term conversion attribution with long-term retention signals

A cooperative game theory solution concept used to calculate the marginal contribution of each pricing touchpoint to the final conversion outcome. It evaluates every possible coalition of channels to determine a fair, additive credit distribution.

• Mechanism: Computes the weighted average of a touchpoint's marginal contribution across all possible channel permutations
• Key Property: Satisfies efficiency, symmetry, dummy, and additivity axioms, ensuring mathematically fair allocation
• Computational Challenge: Requires 2^n model evaluations for n channels, often mitigated via sampling approximations
• Use Case: Ideal for boardroom reporting where defensibility and fairness of budget allocation are paramount

A probabilistic approach that models the customer journey as a sequence of states, where each touchpoint represents a node and transitions carry a probability of moving toward conversion. The removal effect quantifies a channel's importance by measuring the drop in conversion probability when that node is deleted from the graph.

• Output: Transition probability matrix and steady-state conversion likelihood
• Advantage: Naturally captures the sequential order and dependency between touchpoints
• Limitation: Assumes the Markov property—that the next state depends only on the current state, not the full history
• Application: Effective for understanding how a mid-funnel price comparison page influences final purchase probability

A rule-based heuristic that assigns exponentially increasing credit to touchpoints closer in time to the conversion event. It operates on the assumption that recent interactions have a stronger causal influence on the purchase decision.

• Half-Life Parameter: Configurable decay rate determines how quickly older touchpoints lose relevance
• Simplicity: Computationally trivial and easily explainable to non-technical stakeholders
• Weakness: Ignores the critical role of upper-funnel awareness-building touchpoints like initial brand discovery
• Common Use: Frequently applied as a baseline model before deploying more sophisticated algorithmic attribution

A causal inference technique that directly estimates the incremental impact of a specific pricing or promotional intervention on an individual's conversion probability. Unlike correlational attribution, it isolates the true treatment effect by comparing exposed and control groups.

• Core Segments: Classifies users into Persuadables, Sure Things, Lost Causes, and Sleeping Dogs
• Methodology: Employs two-model or class-transformation approaches to predict conditional average treatment effects
• Distinction: Answers 'did the discount cause the purchase?' rather than 'was the discount present during the journey?'
• Integration: Complements multi-touch attribution by validating whether credited touchpoints actually drove incremental behavior

Google's proprietary algorithmic model that uses actual conversion path data from your account to distribute credit across touchpoints based on their observed contribution to conversions. It applies statistical analysis of consumer behavior patterns rather than fixed rules.

• Requirements: Minimum conversion volume threshold (typically 600 conversions in 30 days) for statistical validity
• Methodology: Compares observed paths with conversion to paths without conversion to isolate contributing factors
• Limitation: Operates as a black box with limited transparency into the exact weighting algorithm
• Contrast: Unlike Shapley values, DDA does not guarantee axiomatic fairness properties but is optimized for predictive accuracy

A statistical framework for isolating the true incremental effect of a price change from confounding variables like seasonality or competitor actions. Techniques such as Difference-in-Differences and Propensity Score Matching are essential for attribution models that must distinguish correlation from causation.

• Difference-in-Differences: Compares the change in outcomes for a treated group against a control group over the same time period
• Synthetic Control: Constructs a weighted combination of untreated units to serve as a counterfactual for the treated unit
• Instrumental Variables: Uses an external factor that affects price but not demand directly to identify causal effects
• Critical Need: Without causal grounding, attribution models risk crediting random noise or external market forces as pricing effectiveness