What are the factors that affect the call center traffic forecast?

1. Call centers at different stages apply different traffic forecasting methods.

Traffic forecasting belongs to the category of time series forecasting, and the forecasting methods vary from simple to difficult, but it needs to be reasonably selected according to the main business, scale, traffic incoming law and service level target of the customer service center. Otherwise, even if you choose the most complicated forecasting method, if it is not suitable for your own forecasting environment, it will only bring you more trouble and the result will be similar. The following are several common forecasting methods:

1, mean value prediction method

Average forecasting method is widely used in customer service centers with single business and stable traffic.

The specific formula is: predicted value = average value of all historical data.

2. Moving average forecasting method

The moving average forecasting method only averages the n historical data with the greatest correlation with the data in the forecasting time.

The specific formula is: predicted value = the average value of n historical data with great correlation.

3. Exponential smoothing prediction method

Exponential smoothing prediction method is an improvement of moving average method, which gives different weights to n historical data related to the current time period.

The specific formula is: predicted value = n 1 (historical data 1)+N2 (historical data 2)+…+ (historical data n).

4.ARIMA model

ARIMA is an automatic regression integral moving average model, which is mainly used for the analysis and prediction of time series with long-term trends and seasonal fluctuations. ARIMA's idea is simple. Firstly, the seasonal fluctuation is removed by difference, then the long-term trend is removed, then the sequence is smoothed, and finally the sequence is fitted with a linear function+white noise.

Second, the methods and procedures of incoming call prediction

(A) the collection and collation of historical materials

The processing of historical data is the most important step before forecasting. If the data processing is not clean, it will directly affect the accuracy of prediction. For the traffic situation of the customer service center, the situations that have an impact on the original traffic are mainly summarized as follows:

1, system failure. If a system failure occurs on a certain day, it is necessary to eliminate the telephone traffic with system failure on that day according to the time dimension of the failure and restore the original incoming traffic.

2. The concentrated calls of customers are caused by some sensitive short messages or public opinions.

3. If a sensitive short message is sent or a public opinion occurs on a certain day, resulting in concentrated calls from customers, it is necessary to eliminate the corresponding traffic according to the affected time dimension and restore the original incoming electricity.

4. Lack of manpower and low connection rate.

If it is not because of the above situation, but because of the shortage of manpower and low connection rate, resulting in more repeated calls, the call data of that day is not the real call demand of the original customer. We need to restore the call volume to the original call volume according to the level of repeated calls on that day.

(B) the initial establishment of the model

Different businesses have different influencing factors, so we need to mine historical data to find more important influencing factors. For example, credit card business mainly involves repayment date, bill SMS reminder date, deferred repayment period and other influencing factors. The comprehensive business mainly involves the date of repayment, the date of SMS reminder and other influencing factors. After finding the influencing factors of incoming electricity, we need to give different weights to different customer groups.

At this point, the model has been initially constructed, but with the constant changes in business, route adjustment and customer volume, we need to constantly adjust and improve our forecasting model. The forecasting model is artificially constructed, and there must be factors that we can't consider, so the model is not perfect, and the model value needs to be manually adjusted according to the experience, and the experience needs to be summarized by the forecaster in continuous study and work.

(3) Prediction points outside the model

There are many time periods that cannot be predicted by the model, and we need to make predictions manually according to historical data and experience. For example, during the Spring Festival, Golden Week and small holidays.