In LTE and 5G mobile networks, radio resources are dynamically allocated. In the uplink, the Base Station (BS) needs to know how much data each User equipement (UE) has in its butter to properly share the resource among the different UEs. In this paper, using a simple scenario and measurements made on an operational network, we show that the BS works on systematically outdated information about the butter size, which generates delay spikes. We thus propose to add a predictive estimation of the butter size, called enhanced Buffer Status Report (eBSR). It works with off-the-shelf UE and is incorporated in BS at the Medium Access Control (MAC) layer without changes in scheduling algorithm. We evaluated it on a lab testbed with various traffic profiles. Results show a clear improvement on uplink latency and jitter, together with a limited impact on radio resource usage efficiency.