Abstract: Aiming at the shortcomings of current predictive outbound algorithm, this paper proposes an algorithm that divides predictive outbound algorithm into two parts, predictive outbound algorithm and dynamic statistic. This paper provides optimal methods for both parts, avoiding the bottleneck caused by only optimizing one of them. Dynamic statistics allows to employ various kinds of strategies according to the current state and historical data without concern of how the algorithm is implemented or whether it's useful, it can also generate random numbers in accordance with common distributions, like exponential distribution, Poisson distribution, normal distribution, Chi-square distribution and gamma distribution etc., based on the theoretical model built by the algorithm. Predictive outbound algorithm provides an effective link-theory-and-practice algorithm based on queuing theory model, increasing agent utilization and decreasing call lose rate at the same time.

Key words: predictive outbound algorithm, dynamic statistics, queuing theory, birth-and-death process, agent utilization, call lose rate

摘要： 针对目前预测式外呼算法存在的不足，提出了一种把预测式外呼算法分成预测算法和动态统计两个模块的方法，可以从两个模块方向进行优化，两个方向都有许多优化方案，避免了只优化算法出现的瓶颈问题。统计模块根据当前状态和历史数据选择各种统计策略进行统计，而不用考虑算法如何实现和对算法是否有用，还能根据算法建立的理论模型生成常用分布如指数分布、泊松分布、正态分布、卡方分布和伽马分布等的模拟随机数提供给算法使用。预测算法模块则基于排队论等理论模型提供了一种理论联系实际的高效算法，在提高坐席利用率的同时有效降低了电话呼损率。

关键词: 预测式外呼算法, 动态统计, 排队论, 生灭过程, 坐席利用率, 电话呼损率