Skip to main content


ChannelAttribution Pro is a machine learning library for data-driven marketing attribution and budget allocation. It is an R package and a Python library available for the main Operative Systems (Linux, Windows and Mac). We can also provide a preconfigured Docker container with RStudio or Jupyter and ChannelAttribution Pro installed.

ChannelAttribution Pro is installed locally and all the elaborations are made on the local system where it is installed. It means that no data are transferred outside your local environment.

ChannelAttribution Pro is the professional version of ChannelAttribution which improves ChannelAttribution by offering the following extra features:

Transaction-level attribution with heuristic models, Markov model and Shapley value
  • Monitor ROI for each channel at path-level and for aggregation of paths at time intervals
Perform attribution with the Hidden Attribution Touch model to overcome Media-Mix model limitations
  • Perform attribution when customer journeys are not available
  • Mix digital and traditional channels
  • Perform attribution when only short time-series are available
  • Perform attribution in case of low variability for one or more involved variables
  • Overcome the main limitations of a classical Media-mix model which requires more assumptions and fails in cases of short time series, sparsity of observations, and low variability.
Real-time attribution with Markov model and Shapley value
  • Save computational time. Train the model on huge amount of customer journeys, store the model parameters and then use it for performing attribution on new customer-journeys
Markov model and Shapley value with odds
  • More accurate attribution at path-level
Combine results from Media-mix Model and Multi-touch attribution models at path-level
  • More accurate attribution at path-level bringing results from a Media-mix model at path-level
Out-of-sample validation algorithm for choosing the best Markov model order
  • More accurate attribution with Markov models
  • Choose the best order also for highly inbalanced data using precision-recall curve instead of roc curve
Simplified Shapley value formula
  • Classical Shapley value formula limit the use of Shapley value to problems with less than 10 channels, while simplified Shapley value can be used also with thousands of channels
  • Faster execution of Markov model when huge amounts of customer journeys are elaborated
Read customer journeys directly from CSV files
  • Process huge amount of customer journeys avoiding out-of-memory issues


Student / Researcher
  • Free