Deftpower's platform allows EV drivers to earn rewards through smart charging. While these incentives are nice, they are usually not compelling enough for drivers to indicate their departure time, which is essential for optimizing a smart charging session. Therefore, Deftpower set out to automate EV departure times using AI predictions.

Deftpower and Datacation formed a blended team of data scientists. The goal: develop a state-of-the-art model for predicting EV departure times using patterns in the charging history of users. The case proved to be challenging, since the behavioural patterns of drivers highly fluctuate and drivers themselves often do not their exact departure time. To manage this uncertainty, we introduced a model-native confidence score that indicates how reliable a prediction is. This enables Deftpower to show only high-certainty predictions and automatically disregard cases where no clear behavioural pattern exists. 

AI-generated departure times have now been integrated into the charging apps of Deftpower’s clients. Predictions are shown to drivers when the model is sufficiently confident, and users retain full control to enter or override their departure time. This has increased the number of successful smart charging sessions, leading to higher rewards for drivers and better grid-friendly charging behaviour. 

Our collaboration continues with the development of forecasting models for the energy market. These models will help Deftpower trade energy flexibility by treating aggregated charging sessions as a virtual power plant. By shifting EV charging at scale, Deftpower contributes to a more balanced and resilient energy grid.