Design and Implementation of Ctrip's Predictive Outbound Call Platform

The authors, from Ctrip's Basic Business R&D Call Center team, explain how they built a fully software‑based SoftPBX system that leverages predictive algorithms to reduce agent wait time and increase outbound efficiency for over 15,000 seats handling roughly 300,000 daily outbound calls.

The predictive outbound platform automatically executes outbound tasks based on predefined strategies and number lists, predicting the number of agents that will become free in the next N seconds and pre‑dialing Y calls so that a new customer call is ready as soon as an agent finishes the previous one.

Figure 1: Predictive Outbound Workflow

The core prediction formula combines current idle agents, ongoing outbound calls, queue length, expected agent hang‑ups after N seconds, and configurable coefficients to estimate the number of calls to place, adjusting for success rates and user abandonment.

Key algorithm considerations include the impact of agent behavior variance, pickup rate fluctuations, average call duration, and queue waiting time on overall call loss, requiring continuous tuning based on historical data and manual overrides for unexpected situations.

The platform is built on the SoftIVR system, which integrates with Ctrip's SoftPBX (a SIP‑based softswitch). SoftPBX supports up to 50,000 daily calls across multiple business units, offering flexible data for the prediction model and reducing operational costs.

Figure 2: Parameters Influencing Predictive Outbound Algorithm

Key platform features include a floating‑IP primary‑backup SoftPBX architecture with an OutCallService that dynamically acquires SoftPBX IPs for horizontal scaling, and a concurrent task generator that partitions outbound tasks across multiple instances to improve scheduling real‑time performance.

Figure 4: Communication Structure Diagram

Production deployment began with the hotel business, covering over 200 agents across two locations, achieving a 7% increase in outbound success rate and a more than 20% boost in per‑agent daily call capacity.

Results to date:

200+ hotel agents using predictive outbound.

Outbound success rate improved by ~7% compared to manual dialing.

Per‑agent daily outbound volume increased by over 20%.

As SoftPBX adoption expands, the predictive outbound platform will continue to enhance efficiency for a growing number of agents, serving as one of many extensible features for modern call‑center operations.