Past seismicity is the key to forecasting the future. However, despite decades of research on how to best use seismicity observations for long- medium- and short-term forecasting, there is limited consensus on the optimal strategies to extract and exploit seismicity parameters. The key challenges are: 1) Seismicity parameters vary as a function of space, time and magnitude range, and unraveling predictive patterns is challenging; 2) The data completeness and data quality available is likewise highly variable and typically spans only a fraction of a seismic cycle; 3) The underlaying physical mechanisms are at best partially understood and cannot be readily modelled. This talk will review some of the challenges we are facing as a community when trying to enhance probabilistic seismic hazard assessment as well as operational earthquake forecasting. I will also show promising case studies and recent advances from around the world that suggest future directions in the never-ending quest to enhance our earthquake forecasting abilities.