Closing the knowing-doing gap in AI: From model interpretability to better decisions (sponsored by Teradata)

Description

Businesses can use AI techniques to make accurate predictions but still not act effectively on this knowledge. Since business value derives from actions rather than knowledge, it’s essential to identify a clear path from model predictions to effective actions. The need for such clarity is particularly strong in enterprise settings where models have an overwhelming diversity of input predictors and dizzying array of choice options. Yet the nonlinear “black box” nature of many modern predictive models obscures the link between predictions and actions, since the models provide little reason for a particular score or guidance for the best action to take given a score. You need to bridge the gap between knowing what is likely to happen and what to do about it.

Nachum Shacham reviews model interpretability from the perspective of decision support and outlines approaches to extending model interpretability to action guidance, including multistage modeling where a prescriptive model step is added to leverage both the predictive model scores and model interpretation output. Nachum explains how to implement this approach to predictive decision support at scale, leveraging the Teradata Analytics Platform, and illustrates the techniques using a business example of customer churn prediction and mitigation.

This session is sponsored by Teradata.