How does the hottest Ctrip improve call efficiency

  • Detail

How can Ctrip improve call efficiency with the help of "predictive outbound call"

Introduction to the author

the author of this paper is the call center team of Ctrip basic business research and development department. Based on the traditional call center, the team combines soft switch, intelligent distribution, automatic voice semantic processing and other technologies to provide Ctrip users with humanized, human-computer interaction and convenient voice services

Ctrip currently has 15000 + seats, with an average daily call volume of about 30W. In such a large-scale call center outbound service, we continue to try to reduce the waiting time of seats and improve the outbound call efficiency, so as to provide more reliable call center services for the business

as a pure soft platform, the softpbx system independently developed by Ctrip has strong functional flexibility and scalability, which also provides technical feasibility for improving the efficiency of agent outbound calls: predict the number of agents to be idle, the current number of callable and other information according to the algorithm, and automatically make outbound calls to the tasks to be performed. When the call is connected, it will be transferred to the agent, and the agent will continue to complete the subsequent work

Figure 1 predictive outbound workflow

principle introduction

predictive outbound platform is a system that automatically executes calls according to business customized outbound tasks. Its execution feature is that the outbound call strategy and outbound call number list need to be defined in advance. Execute the call in sequence according to the number list, and transfer the call to the agent or IVR according to the policy after the call is connected; If the call is unsuccessful, the detailed call result will be automatically recorded (such as shutdown, empty number, not in the service area, no response from the user, etc.)

the core principle can be seen at the door of the market is that it is predicted that there will be x seats free in N seconds, and Y seats will be initiated in advance. As soon as the seats hang up, there will be new customer assignments, and as soon as the customers answer, there will be seats free to answer

prediction algorithm

the accuracy of predictive outbound calls mainly depends on two aspects: the answering ability of agents and the called connection rate. The main parameters affecting these algorithms are shown in Figure 2:

Figure 2 predictive outbound call algorithm influence parameters

for pre range × 2、 × 5. Let's show the performance measurement algorithm:

outbound call volume = (current number of idle seats - number of outbound calls * outbound call success rate - current number of queues * proportion of users not giving up +number of idle employees hanging up after N seconds + current number of post-processing * configuration coefficient)/(outbound call success rate * proportion of users not giving up) * manually adjust the coefficient

the prediction algorithm takes the queue behavior as the statistical object, ignoring the individual behavior of the seats. If the behavior difference of seats is large, The prediction effect is poor

the fluctuation of off hook rate cannot be controlled, and the sudden change of off hook behavior of customers will cause too many queues in a short time

the task starts and calls according to the idle agent. Generally, the calling speed is relatively slow. The number of seats that can be served and the fluctuation of seat behavior will affect the prediction effect

the average call duration is an important parameter of the prediction algorithm. If it fluctuates too much, it will affect the queuing duration

the length of customer queuing will affect the final call loss. The call loss caused by non queuing hang up cannot be controlled by the prediction algorithm

therefore, the prediction algorithm is not fixed, and the weight ratio of factors affecting different services is also different. When each service is online, it needs to continuously iterate and optimize the algorithm according to the historical data of agents and users until the working efficiency and call loss rate of agents stabilize

another point is that the business must be allowed to manually force the intervention of the predicted value. According to the real-time data, the on-site supervision can also control the call loss rate at any time in case of emergencies, and ensure the work efficiency of the seats

platform design and construction

after the algorithm model is completed, start to design and build the outbound call platform. After many investigations, comparisons and attempts, the team independently developed a new softivr outbound call platform. The establishment of this platform perfectly combines Ctrip softpbx system (Ctrip's new generation soft switch platform based on SIP)

at present, the softpbx system has gradually covered the seats of many Bu, such as vacations, hotels, air tickets, etc., and carries about 50000 calls a day, ensuring the stability and efficiency of seat calls. Softpbx can realize traffic soft switching without relying on the switch, which is easy to deploy, smooth expansion, transparent business data and functions, provides flexible data support for the prediction formula, and greatly reduces the operation and maintenance cost

Figure 3 system logic structure diagram

feature 1: the outgoing call system and softpbx multipoint communication architecture

softivr outgoing call system adopts the active and standby mode of floating IP. In the actual process, only one softpbx handles outbound calls, and the outbound call volume that a single softpbx can carry is limited, so the number of service outbound calls that the outbound call system can carry is limited. Therefore, softivr chooses to adopt the one-to-one connection mode between the outgoing call communication service

outcallservice (responsible for sending outgoing call data to softpbx) and softpbx. Multiple outcallservice instances obtain their corresponding softpbxip address through preemption, so that the outgoing call platform has horizontal expansion capability and improves business access capability

Figure 4 communication structure diagram

feature 2: outbound trigger generato Qingdao Energy Institute has developed a new polyurethane material r concurrency improvement

outbound platform is designed as a multi instance and service concurrency mechanism, which mainly solves the following two problems:

1) there is only one example in traversing services and screening outbound call tasks. Microscopically, different services are executed in sequence. With the increase of the number of services, the traversal cycle is very long, The real-time performance of scheduling will become very weak. The concurrent scheduling control by business can realize the concurrent scheduling between different businesses and enhance the real-time scheduling

2) based on the task filtering process of a single task table, with the increase of the volume of business data, the length of outbound call data filtering will increase, and the probability of other operation timeouts will increase. The outbound call task filtering based on sub table can greatly reduce the time-consuming of data filtering and the impact on other concurrent operations

Figure 5 task scheduling structure

softivr platform replaces the original supplier platform through the server, and can be deployed through the standard virtual machine, without additional hardware support, and the cost is low. At the same time, the platform has fully considered the existing business scenarios and the future development direction of outbound call mode at the beginning of design to ensure its versatility and reliability. Externally, the platform provides standard API interfaces to facilitate Bu's rapid access. At present, it has access to train tickets and hotel outbound calls, with a maximum of 100000 outbound calls per day. Internally, the platform also has a perfect monitoring system to ensure overall stability

based on the prediction algorithm model and the establishment of softpbx system and softivr platform, the prediction outbound platform was successfully launched and officially verified in production in two months

production examples

at this stage, the application of softpbx and the load on the business have become increasingly mature, which has great advantages for predicting the official production of the outbound platform. The launch of the new platform will not have a great impact on the business, the overall process is transparent, and the business does not need to spend a lot of time to get familiar with the new operating habits. It is predicted that the first guest of the outbound call platform is the hotel. Within two weeks, it will complete the server deployment, firmware compilation and update, ccdesk release, and the deployment and on-site support of a total of 200 + seats in Shanghai and Xinyang

while deploying, the emergency plan is also prepared. On the one hand, when the outbound call platform breaks down, it can ensure that the seats can switch back to the original working mode, that is, manually outbound calls when temporarily backing back to the original preview, such as battery weight impact testing machine, battery ball impact testing machine, battery weight impact testing machine, battery impact testing machine, heavy object impact testing machine; On the other hand, since the work content of the business has changed from outbound to inbound, in order to prevent this new process from having a certain impact on the business personnel, communicate with the business personnel in advance, clarify the overall operation process and precautions, make it start as soon as possible, and avoid more problems

after the deployment of agents, in order to ensure the smooth transition between new and old modes and give business personnel sufficient time to adapt, the prediction outbound platform adopts the gray tangent method to gradually switch to production, which also effectively reduces the risk

up to now:

Hotel predictive outbound calls have covered 200 + seats

the success rate of outbound calls is about 7% higher than that of traditional manual outbound calls

the average daily outbound call capacity of a single seat has been increased by more than 20% from 100

with the migration of softpbx seats, predictive outbound call platform will bring convenience and work efficiency improvement to more and more seats. Predictive outbound call is only one of the expansion functions brought by softpbx to the call center business. More and more functions will be expanded to quickly respond to the needs of the call center business in the new era

Copyright © 2011 JIN SHI