Exponential Time-Decaying Attribution Model Using Half-Life

The E-Commerce platform users are growing rapidly which has become critical for companies to analyze the user's journey towards buying their products. In this journey the user may have used/been exposed to various devices and touch-points/channels. So now to identify which of those touch-points resulted in a positive conversion, the companies use various types of attribution models. By identifying this, the company then allocates its marketing spend accordingly in the future on the chosen channels to further optimize their sales and customized user experience. Over the period of time, various types of attribution models have been developed (broadly categorized into 3 sections: Single-Touch, Multi-Touch and Probabilistic Attribution Model) with each model having its own pros and cons. Here, a new model is proposed, Exponential Time- Decaying Attribution Model using Half-Life (TD-HLED). And then a comparative analysis of 6 Models is provided: First-touch, Last-touch, Linear-Touch, Markov Chain Model and our proposed model, TD-HLED. Furthermore, extensive experiments have proved that this model's attributions have more accurately reflected on how the users have interacted with the marketing than the existing simplistic rule-based models.