--- title: Shapley Value Attribution Explained | Attribution Model Guide | RGM® url: https://realgrowthmatters.com/learn/measurement/shapley-attribution/ updated: 2026-06-10 source_html: https://realgrowthmatters.com/learn/measurement/shapley-attribution/ --- # Shapley Value Attribution: game-theoretic credit allocation that satisfies four fairness axioms. Shapley value attribution uses cooperative game theory to allocate conversion credit fairly across channels. Each channel's credit equals its average marginal contribution across all possible channel orderings. Part of the attribution model family that also includes first-touch, last-touch, linear, time-decay, position-based, Markov chain, and Shapley attribution. Modern teams use multiple models in parallel and validate against incrementality testing. By **David Schaefer** · [LinkedIn](https://www.linkedin.com/in/daschaefer/) · Updated May 2026 · 10 min read · [5 sources cited](#sources) ## Key takeaways - Shapley value attribution uses cooperative game theory to allocate conversion credit fairly across channels. - Lloyd Shapley won the 2012 Nobel Prize for the math behind it. - Each channel's credit equals its average marginal contribution across all possible channel orderings. - The model satisfies four fairness axioms: efficiency, symmetry, dummy, additivity. - Computational cost grows exponentially with channel count. Production implementations use Monte Carlo sampling. - Similar outputs to Markov chain attribution at higher cost. Most teams should use Markov. ## What Shapley value attribution is Shapley value attribution applies cooperative game theory to marketing channels. Each channel's credit equals its average marginal contribution across all possible orderings of channels in the journey. The method is the mathematically fair allocation: it is the only attribution rule that satisfies four fairness axioms (efficiency, symmetry, dummy, additivity). Lloyd Shapley won the 2012 Nobel Prize in Economics for the underlying mathematics. The approach is rare in production because the computational cost grows exponentially with the number of channels. The model treats each channel as a player in a cooperative game. The "payoff" is the conversion. Each subset of channels (each "coalition") has a known conversion probability from observed data. Each channel's Shapley value is the average of how much it adds to every possible coalition it could join. The math is elegant and the result is provably fair under the four axioms. ## How it works Six steps. Define the set of channels. For each subset of channels, compute the conversion probability from observed data. For each channel, calculate the marginal contribution it adds to every subset that excludes it. Average those marginal contributions across all orderings. The average is the channel's Shapley value. Sum across the period for total channel attribution. The computational cost is the trap. With N channels, the number of subsets is 2 to the power of N. For 5 channels, that is 32 subsets. For 10 channels, 1,024 subsets. For 20 channels, over 1 million. Production implementations approximate Shapley using Monte Carlo sampling rather than exact computation. ## When Shapley attribution fits Shapley is right for businesses that need provably fair attribution and have data engineering capacity to implement Monte Carlo Shapley. The model is rare in production because Markov chain attribution produces similar outputs at lower computational cost. Shapley is most useful as the academic gold standard against which simpler models are validated. ## Common failure modes Two failures. Implementing exact Shapley with too many channels, hitting computational explosion. Treating Shapley as causal truth when it is still correlational from observed data. ## Quick answers What is Shapley value attribution? : Cooperative-game-theory math that allocates conversion credit fairly across channels. Each channel's credit equals its average marginal contribution across all possible orderings. Why is it 'fair'? : Shapley value is the unique allocation that satisfies four fairness axioms: efficiency, symmetry, dummy, and additivity. Lloyd Shapley won the 2012 Nobel for the math. How is it different from Markov? : Markov uses removal effects; Shapley uses marginal contributions across all orderings. Both produce similar outputs for most journeys. Shapley is more computationally expensive. Why is it rare in production? : Computational cost. With N channels, the number of subsets is 2 to the power of N. Twenty channels = 1 million subsets. Most teams use Markov instead for similar outputs at lower cost. How do production implementations work? : Monte Carlo sampling instead of exact computation. Sample a subset of orderings, average marginal contributions, accept the approximation. Should I use Shapley or Markov? : Markov for most teams. Shapley is the academic gold standard but the computational cost rarely justifies the marginal accuracy gain over Markov. ## Frequently asked What is Shapley value attribution? An attribution model that uses cooperative game theory to allocate conversion credit fairly across marketing channels. Each channel's credit equals its average marginal contribution across all possible orderings of channels in the journey. Who developed Shapley value? Lloyd Shapley, an American mathematician at UCLA. He developed the value in 1953 and won the 2012 Nobel Prize in Economics jointly with Alvin Roth for related work in market design. Why is Shapley value 'fair'? It is the unique allocation rule that satisfies four fairness axioms: efficiency (total credit equals total conversion), symmetry (channels that contribute equally get equal credit), dummy (channels that contribute nothing get nothing), and additivity (credit aggregates linearly across conversion events). How is Shapley different from Markov chain attribution? Markov uses removal effects (drop in conversion probability when a channel is removed). Shapley uses marginal contributions across all possible orderings of channels. Both produce similar outputs for most journeys; Shapley is more computationally expensive. Why is Shapley rare in production? Computational cost. With N channels, the number of subsets is 2 to the power of N. Twenty channels means 1 million subsets to evaluate. Production implementations use Monte Carlo sampling to approximate exact Shapley. What is Monte Carlo Shapley? An approximation method that samples a subset of all possible channel orderings instead of computing all of them. The Monte Carlo approximation converges to true Shapley value with enough samples and is the practical way to compute Shapley at production scale. Is Shapley causal? Closer to causal than heuristic models but still correlational from observed data. The fairness axioms guarantee fair allocation given the observed data; they do not guarantee causal accuracy. Validate against incrementality tests. Should I use Shapley over Markov? Markov for most teams. Shapley produces similar outputs at substantially higher computational cost. Use Shapley when you specifically need provably fair allocation (academic work, legal-sensitive contexts) or when you have the engineering capacity for Monte Carlo implementation. ### Sources cited on this page 1. Google — [Attribution models documentation in GA4](https://support.google.com/analytics/). 2. Avinash Kaushik — [Occam's Razor blog on attribution models](https://www.kaushik.net/avinash/). 3. Anderl, Becker, von Wangenheim, Schumann — "Mapping the Customer Journey", International Journal of Research in Marketing (2014). 4. Anthropic / Google AI — Academic literature on attribution modeling. 5. Real Growth Matters Inc. — Internal audit data on attribution-model performance, 2024-2026.