The impact of AI-based conversational agent on the firms’ operational performance: Empirical evidence from a call center

Figures & data

Figure 1. The service flow chart of call center.

Table 1. Descriptive statistics.

Variables	Data type	Explanations	Min	Max	Mean	Median
Age	Integer	Actual age calculated based on ID card information	16	70	40.02	37.9
Gender	Binary	l = male, 0 = female	0	1	0.37	0
Open year	Integer	Length of time from registration to now	0	21.67	6.83	6.2

Table 2. Randomization check.

Group	Age	Gender	Open year
Group 1 (last_digit=1,7)	39.89	0.36	6.74
Group 2 (last_digit=3,5,9)	39.97	0.36	6.77
T-value	0.99	−0.19	1.10

Table 3. The description of variables.

Variables	Description
Average daily call lengtht,g (ALt,g)	Average daily call length of group 1 or group 2 on day t
Average call length differencet (ALDt)	The difference of average daily call length between group 1 and group 2 on day t
Abnormal daily incoming callst (ADCt)	The number of abnormal daily incoming calls of group 1 on day t
Abnormal average call length difference of each dayt (AALDt)	The difference of average abnormal daily incoming calls between group 1 and group 2 on day t
Predicted daily incoming callst (PDCt)	The predicting daily incoming of group 1 on day t
The mean of abnormal daily call lengtht (M1)	The abnormal daily call length of group 1 on the time period of post-event
The mean of abnormal average call lengtht (M2)	The average abnormal call length of group 1 on the time period of post-event

Table 4. A table of function sequences.

Equations	Description
Equation 1: $A{L}_{t,g}=\frac{\sum _{i=1}^N{L}_{i,t,g}}{DI{C}_{t,g}}$	This equation is used to calculate the average daily call length. i represents the i incoming call, t represents the t day, and g represents the group of the caller.
Equation 2:$AL{D}_t=A{L}_{t,1}-A{L}_{t,2}$	This equation is used to calculate the average call length difference. t represents the t day.
Equation 3: $AD{C}_t=PD{C}_{t,1}-PD{C}_{t,2}$	This equation is used to calculate the abnormal daily incoming calls. t represents the t day.
Equation 4:$AAL{D}_t=PA{L}_{t,1}-OA{L}_{t,2}$	This equation is used to calculate the abnormal average call length difference of each day. t represents the t day.
Equation 5: $OD{C}_t=\alpha +\beta X_t+{\epsilon }_t$	This equation is used to calculate the $\stackrel{⌢}{\beta }$. t represents the t day.
Equation 6: $PD{C}_t=\alpha +\hat{\beta }{X}_t+{\epsilon }_t$	This equation is used to calculate the predicted daily incoming calls. t represents the t day.
Equation 7:$M_1=\sum _{t=1}^{\hspace{0.17em}}\frac{AD{C}_t}{N_1}$	This equation is used to calculate the mean of the abnormal average daily incoming calls. t represents the t day. N represents the total number of days.
Equation 8: $M_2=\sum _{t=1}^{\hspace{0.17em}}\frac{AAL{D}_t}{N_1}$	This equation is used to calculate the mean of the abnormal average call length difference of total days. t represents the t day. N represents the total number of days.

Figure 2. Timeline of event study.

Table 5. Results.

	Mean	T value
M1	−19.63	−0.56
M2	−7.23**	−4.53

Table 6. Abnormal call length and daily incoming calls by time blocks.

Time block	Mean of abnormal daily incoming calls	Mean of abnormal call length difference
8:00 A.M.-8:59 A.M	−2.12(−0.54）	−7.81 (−0.96)
9:00 A.M.-9:59 A.M	8.05 (1.70)	−5.62 (−1.63)
10:00 A.M.-10:59 A.M	1.10 (0.19)	−0.14 (−0.02)
11:00 A.M.-11:59 A.M	3.54 (0.46)	1.88 (0.38)
12:00 P.M.-12:59 P.M	4.40 (0.71)	−16.84**(- 3.039)
1:00 P.M.-1:59 P.M	2.06 (0.46)	−3.51 (−0.76)
2:00 P.M.-2:59 P.M	5.85 (1.74)	−3.82 (−0.34)
3:00 P.M.-3:59 P.M	12.40**(2.12)	1.50 (0.26)
4:00 P.M.-4:59 P.M	5.37 (1.07)	−15.89**(- 2.33)
5:00 P.M.-5:59 P.M	2.65(0.43)	−16.76**(- 2.01)
6:00 P.M.-6:59 P.M	−6.64**(−2.61)	−4.73 (−0.74)
7:00 P.M.-7:59 P.M	−3.80 (−0.69)	−1.24 (−0.08)
8:00 P.M.-8:59 P.M	2.21 (1.30)	−26.05**(- 2.52)

Table 7. Falsification test.

	Mean	T value
abnormal daily incoming calls	−12.46	−0.88
abnormal call length difference	−.91	−0.57

***, ** and * represent significance at the 1%, 5%, and 10%level, respectively