The article examines the dynamic allocation of outbound call launches in conversational AI voice-calling systems, where voice bots qualify customers before transferring them to human operators. The study aims to reduce customer hold time and operator idleness by replacing fixed-concurrency settings with a receding-horizon integer-programming controller. The relevance of the work stems from the growing use of AI voice agents in outbound contact centers, where stochastic answer behavior, delayed transfers, and scarce operator capacity create a control problem that static dialing rules cannot resolve. The novelty lies in modeling the bot-induced transfer delay as an empirical lag convolution and embedding this representation within a closed-loop optimization framework, validated through discrete-event simulation calibrated against production logs. The main results show that the proposed controller captures most of the throughput from aggressive dialing while reducing abandonment and hold times, and it unlocks unused operator capacity at higher staffing levels. The article will be useful to researchers, contact center engineers, operations managers, and developers of AI voice-calling platforms.