Abakus uses the value of a set of players (each player set can have a different value empirically observed and measured in the data) and computes the incremental value of each player over all possible orderings (permutations) of players. A key requirement of computing the Shapley value for attributions is to have sufficient statistics (occurrences of actual players set combinations) in the data. The resulting Shapley value for a player is the average incremental value associated with the player over all combinations and sequences of players. The Shapley value has a mathematical proof of being unique for marketing attribution.
The question asked was whether we use the different values for a player in a player set based on the sequence the players in the set occur in the data. The answer is no. There are several issues:
- The number of sequences is infinite, where a player can appear in more than one in a sequence. If we limit to the sequences we observe in the data, even then the statistics associated with each sequence will be very small
- The problem of incremental value for a player that appears multiple times in a sequence is not well defined
- The order of players is not well defined for multiple occurrences in a sequence (e.g. in sequence A-B-A-B which ad A or B appears before the other).
Note that Abakus can solve for certain restricted cases of sequence analysis such as first Display ad before first Search ad. However, we view this as secondary analysis that we will provide to clients only on request as a custom analysis.