The disclosed technology describes on-demand adjusting of the quality of service (QoS) class identifier (QCI) relative weight settings for different QCI classes associated with user devices. A QoS controller, which can be implemented in a standalone or a cloud configuration, updates the QCI weight settings as requested or needed to deal with changing network environments, such as growing traffic, different RF conditions, new devices, ratio of different service classes of user devices, to improve the overall performance of one or more cell sites. In one implementation, the QoS controller collects actual network statistics, and runs simulations based on those statistics with different groups of candidate QCI weight settings to generate multiple result sets. The QoS controller evaluates the result sets to determine which group of QCI weight settings meets desired performance objectives, and then applies those QCI weight settings to one or more NodeBs for use with actual data traffic.