Electronic Commerce Research and Applications 59 (2023) 101266

Contents lists available at ScienceDirect

Electronic Commerce Research and Applications

journal homepage: www.elsevier.com/locate/elerap

Understanding the role of live streamers in live-streaming e-commerce☆

Hailiang Chen a, Yifan Dou b, *, Yongbo Xiao c
a
Faculty of Business and Economics, The University of Hong Kong, Hong Kong
b
School of Management, Fudan University, China
c
School of Economics and Management, Tsinghua University, China

A R T I C L E I N F O A B S T R A C T

Keywords: The rise of live-streaming e-commerce has attracted the wide participation of online influencers, brands, and
Live-streaming e-commerce retailers. Live streamers offer a fresh shopping experience to consumers through broadcasting product demon­
Clustering strations and communicating with them. This study characterizes the streamers’ behavior and explores the key
Determinants
drivers of live-streaming e-commerce success as measured by gross merchandise value (GMV) and fan growth.
Gross merchandise value
Fan growth
We employ both machine learning and econometric methods to analyze a unique dataset of 55,096 shows by the
top 1,000 live streamers on Alibaba’s live streaming platform. We identify three distinct clusters. The most
important differentiating features include a live streamer’s platform affiliation and product category. Selling
more products and spending more time on each product in a live-streaming show are two factors driving both
GMV and fan growth. We also discover that a large fan base does not always help, as the positive effect of fan
base only exists conditionally.

1. Introduction KKW perfume was sold out in merely a few minutes (Bu et al., 2019).
Besides, brands and retailers also found it useful to engage consumers
Live-streaming e-commerce is an interactive form of online shopping during lockdowns and social distancing amid the COVID-19 pandemic,
that integrates video live-streaming with e-commerce, creating a which further contributes to the growth of live-streaming shopping
communication channel for live streamers (or anchors)1 to broadcast worldwide. Top technology companies, including Google, Amazon, and
product demonstrations and promotions and address consumers’ ques­ Facebook, have all launched live-streaming shopping platforms (Teng,
tions in real time. Live-streaming e-commerce is believed to appear first 2020). Shopify, the global leader in e-commerce services, also intro­
in 2016 on MOGU, an NYSE-listed online shopping platform from China duced livestreaming features (The Drum, 2020). Brick-and-mortar re­
(Li and Peng, 2021). Since 2017, the sales of live-streaming e-commerce tailers like Walmart partnered with TikTok to host live-streaming events
have been growing quickly, with a yearly average growth of more than during the 2020 holiday season (Walmart, 2021).
150% (Yang and Tung, 2020). The current leader, Alibaba’s Taobao The primitive form of live-streaming e-commerce dates back to home
Live, also reports a rapid growth of the daily active users and live- shopping TV programs in the 1980s (McDonald, 1995, Stephens et al.,
streamers in recent years (Zhang, 2021). 1996, Cortese and Rubin, 2010). On one hand, the similarity between
This trend has been expanding globally, with retailers and brands home shopping TV and live-streaming e-commerce lies in the hosting
joining the scramble to engage consumers through live-streaming e- anchors who deliver the product demonstrations visually. Therefore, in
commerce (WSJ, 2020, CNBC, 2021). During the Singles’ Day shopping both scenarios, the anchors’ product knowledge, personal usage expe­
festival in 2019, a live-streaming show hosted by Viya, one of China’s rience, and promotional skills are all instrumental for a successful show.
most famous key opinion leaders (KOLs), attracted 13 million viewers to On the other hand, live-streaming e-commerce is more ubiquitous, as
the store of Kim Kardashian West (KKW) on Alibaba’s Tmall, and the consumers face a stunning variety of choices because each platform can

☆
Hailiang Chen’s work is supported by the National Natural Science Foundation of China (91746101); Yifan Dou’s work is supported by the Shuguang Program of
Shanghai Education Development Foundation and Shanghai Municipal Education Commission; Yongbo Xiao’s work is supported by the National Natural Science
Foundation of China (72125002, 72293561).
* Corresponding author at: 670 Guoshun Rd. Yangpu District, Shanghai 200433, China.
E-mail address: yfdou@fudan.edu.cn (Y. Dou).
1
In this paper, we use the three phrases “live stream account”, “live streamer”, and “anchor” interchangeably.

https://doi.org/10.1016/j.elerap.2023.101266
Received 24 December 2021; Received in revised form 24 February 2023; Accepted 24 April 2023
Available online 29 April 2023
1567-4223/© 2023 Elsevier B.V. All rights reserved.
H. Chen et al. Electronic Commerce Research and Applications 59 (2023) 101266

host tens of thousands of live-streaming shows simultaneously. In streamers on Taobao Live, and all their 55,096 live-streaming shows in a
addition, the role of the live streamer is much more flexible than the two-month window, from October 10th to December 12th, 2020. We
home shopping programs because brands/sellers can either run live- construct 64 features from this dataset for profiling the live streamers,
streaming shopping shows (e.g., Tmall Supermarket in Fig. 1a and based on which our clustering analysis identifies three distinct clusters
Huawei in Fig. 1b) or collaborate with independent live-streaming of live streamers. Consistent with Luo et al. (2021), our result suggests
influencers (Borak, 2020). Austin Jiaqi Li (Fig. 1c), a former L’Oreal that the live streamers in the clothing/apparel category differ signifi­
salesman and now a famous live streamer, is a household name in China cantly from others because they sell more products in each show, and the
with 64.4 million followers on Taobao Live and another 44.9 million products are also more expensive averagely. We then incorporate this
followers on TikTok as of June 2022. information to measure the performance of each show by the gross
In contrast to its popularity in practice, live-streaming e-commerce merchandise value (GMV, a widely-used proxy for platform sales) and
remains largely understudied in the literature. Some recent works have fan growth, with a regression analysis at the show level. Among the
started to examine the live streaming platform design (e.g., Tong, 2017, many interesting insights yielded from this analysis, we show that
Sun et al., 2019), brands/product effects (e.g., Park and Lin, 2020), and selling more products in each show and spending more time on each
the role of live streamers. The last spectrum is most related to our study. product can help increase both GMV and fan growth. In addition, a
However, most literature in this spectrum is survey-based (e.g., Wong­ longer time gap between two shows is associated with a lower GMV,
kitrungrueng & Assarut, 2020, Lu and Chen, 2021, Gao et al., 2021). One which validates the current industry practice that live streamers run
exception is the recent work by Luo et al. (2021) which uses conversa­ shows almost on a daily basis. Surprisingly, on average, a live streamer’s
tion logs from live-streaming shopping shows to code the linguistic fan base (measured by the number of followers) does not have a sig­
persuasive styles of live streamers. They find that the linguistic style of nificant effect on GMV. By investigating its interaction with anchor
appealing to personality has the greatest positive impact on sales, but experience (measured by the number of prior shows), we find that fan
the linguistic style of appealing to logic has a negative impact. However, base has a positive effect on GMV only for a live streamer’s earlier
the linguistic pattern alone is still far from enough to characterize how shows. To explore the heterogeneity in the determinants of live
live streamers have reshaped e-commerce. For example, how do brands streaming success across different live streamers, we conduct a further
work with various live streamers, and how do live streamers plan their subsample analysis for each cluster. These findings offer rich managerial
shows? These are novel topics that do not appear in traditional e-com­ implications for practice.
merce settings. Our paper contributes to the e-commerce literature in three aspects.
These questions motivate our exploratory study in this paper, the First, to the best of our knowledge, our study is the first to empirically
starting point of which is to profile various live streamers by segmenting investigate the behavioral patterns of live streamers using a large and
them based on personal and operational characteristics. We believe it is representative sample. Second, our clustering analysis discovers that
a prerequisite for further investigations on the determinants of live- specializing in clothing/apparel products is one of the most important
streaming success. To see this, in traditional e-commerce (Fig. 2a), differentiating features among live streamer segments. Third, our study
buyers purchase from sellers through platforms, sometimes facilitated also identifies several factors that drive more sales and fan growth. Some
by reviews and ratings left by others in the purchase decisions (e.g., unique features in live streaming e-commerce, including product se­
Chevalier and Mayzlin, 2006). In contrast, in live-streaming e-commerce lection, show scheduling, fan base, and prior experience, could influence
(Fig. 2b), live video streaming enables consumer engagement in real- the GMV and fan growth. On one hand, these findings provide empirical
time with a visualized human touch which is far more influential than support for some common practices, such as hosting live-streaming
reading reviews from unknown buyers. The impact is further reinforced shows frequently on a daily basis. On the other hand, the regression
when the visualized interaction is from KOLs with an existing follower or analysis toward GMV also yields interesting results, such as the inter­
fan base. Therefore, the performance of live streaming e-commerce de­ action effects of live streamers’ prior experience and their fan base.
pends critically on the live streamer’s choice (e.g., sellers themselves or
KOLs) and how the chosen live streamer would organize the show. 2. Literature review
Therefore, our study takes two steps. First, we identify live streamer
segments and then use this information to explore the determinants that E-commerce is one of the most extensively studied research topics in
drive the success of live-streaming e-commerce shows. Specifically, we the information systems literature. Prior studies have examined various
obtain a unique dataset containing the records of the top 1,000 live forms of e-commerce, such as e-commerce sites accessed on desktop

Fig. 1. Screenshots of Live Streamers Demonstrating Products in Live-Streaming Shows.

2
H. Chen et al. Electronic Commerce Research and Applications 59 (2023) 101266

Fig. 2. Comparison between Traditional E-commerce and Live-streaming E-commerce.

computers (e.g., Ba and Pavlou, 2002, Brynjolfsson and Smith, 2000), and Lee (2002) also shows that television shopping channels offer easy
mobile commerce (e.g., Xu et al., 2017), and social commerce (e.g., access for customers to impulse purchase opportunities. When consid­
Huang et al., 2020). Although live-streaming e-commerce has received ering multiple purchasing channels, Urbany et al. (1988) find that the
increasing awareness among e-commerce practitioners, related research TV shopping experience can help consumers form reference prices that
is still developing in the literature. A few recent studies have started to eventually affect their price search process in other channels. In contrast
examine the decision-making of different existing players in live- to these studies on TV shopping programs, our paper puts the emphasis
streaming e-commerce, but no prior study has specifically focused on on the role of live streamers, which is similar but much more complex
the role of live streamers in e-commerce. Chen et al. (2021b) demon­ than the hosting salesperson in a TV shopping program. For example, a
strate how a live-streaming e-commerce platform (e.g., Amazon Live, single live-streaming show in our dataset may run for more than 10h,
Taobao Live) can leverage information provision to improve its adver­ carrying more than 100 products, which is unprecedented in traditional
tising revenue. Qi et al. (2021) study the capacity investment strategy of TV shopping programs. Thus, our machine learning and regression an­
a manufacturer who sells products on a live-streaming e-commerce alyses on the live streamers help extend the understanding of hosting
platform that charges a commission. Chen et al. (2021a) examine how anchors.
online retailers’ adoption of live video streaming affects their product
sales. They find an increase in product sales only when the live video 3. Data
streaming is product-oriented. Fei et al. (2020) conduct an eye-tracking
experiment to examine how social cues (herding message and interac­ Our data is acquired from a leading analytics company specializing
tion text) affect e-commerce live-streaming viewers’ attention allocation in live-streaming e-commerce in China. This company obtains and
and purchase intention. Our study differs from these live-streaming e- merges data from multiple resources affiliated with Alibaba, including
commerce studies in that we primarily focus on the behavioral patterns Alibaba’s Brand Databank, Open Data API, and multiple analytical tools
of live streamers in e-commerce and the determinants of live-streaming from Alibaba’s seller service market (fuwu.taobao.com). The sample
e-commerce success. used in our study covers 55,096 live-streaming shopping shows hosted
Our study is also related to a growing literature on live video by the top 1,000 live streamers2 from October 10th to December 12th,
streaming of video games and talent shows. Unlike live-streaming e- 2020, on Taobao Live. As one of the fastest-growing businesses, Taobao
commerce, game-playing or talent shows do not directly involve product Live has reached a milestone of over 500 billion CNY (approximately US
promotions. Instead, the audience follows a pay-what-you-want $76.3 billion) of GMV in the fiscal year 2021, representing 6.3% of the
(PWYW) mechanism to “send gifts to” or “tip” the live streamers (e.g., overall GMV from the Alibaba ecosystem (Businesswire, 2021).
Li et al. 2018, Lin et al., 2021). For example, Lin et al. (2021) find that a For each live streamer, we first collect information on individual
happier live streamer makes the audience happier and begets intensified characteristics, including the account ID, account nickname, whether
viewer activities, particularly tips. These effects are pronounced only the live streamer is supported by an MCN agency, whether the account
after a live stream has been active for a while, and they manifest only in has at least one digital storefront on Taobao or Tmall (i.e., the two
streams by broadcasters who have more experience or are more popular largest marketplaces affiliated with Alibaba), whether the account is a
in past live streams. This is consistent with our findings from the live- retailer, and its specialty areas of merchandise. Many independent live
streaming e-commerce context, suggesting that the role of the streamers are supported by an MCN agency. Live streamer accounts
streamer experience is vital across various forms of live-streaming created by companies are more likely to have a storefront on Tmall. Only
shows. a small group of top live streamers, particularly the most popular and
Finally, our study is also related to the literature on TV shopping
programs. For example, Warden et al. (2008) suggest that in localized
television home shopping programs, the intensive use of communica­ 2
The ranking is measured based on the total GMV on the Singles’ Day
tions between salespersons and customers helps build customer re­ (November 11, or Double 11) shopping holiday in 2020. The total GMV pro­
lationships for marketing campaigns. The survey conducted by Kacen duced by one live streamer on this day ranged from 1.29 million to over 841
million CNY.

3
H. Chen et al. Electronic Commerce Research and Applications 59 (2023) 101266

successful ones, such as Austin Jiaqi Li and Viya, may operate their shows hosted by a live streamer before the show. We also take the log­
storefronts in Tmall as resellers. A live streamer is registered in the data arithmic transformation of the aforementioned independent variables
to be specialized in at least one of the following 15 product categories: because they are right-skewed. Double11 and Double12 indicate whether
Auto, Babycare, Beauty, Clothing, Digital, Entertainment, Flowers and a live streaming show is hosted during the time windows from
Pets, Food, Game, Health, Home, Jewelry, Men, Sports, and Travel. For November 1st to 11th, 2020, and December 5th to 12th, 2020, respec­
each show, we have the information of showroom ID, showroom host (i. tively.3 Morning, Afternoon, and Evening are indicators denoting whether
e., one of the top 1,000 live streamer accounts), showroom title, start a live streaming show is conducted from 6 am to noon, from noon to 6
time, end time, duration, number of products sold in the show, average pm, and from 6 pm to midnight in a day, respectively. The six hours from
transaction value, the host’s numbers of fans (i.e., followers) at both the midnight to 6 am serve as the baseline group for comparison in the
beginning and end of the show, and GMV (i.e., the sum of the sales of all regression analysis. Monday, Tuesday, Wednesday, Thursday, Friday, and
products exhibited in each show). Saturday are indicators denoting whether a show is conducted on the
Table 1 presents the summary statistics of the variables at the show corresponding day within a week. In the regression analysis, shows
level. We are interested in two key performance measures that later conducted on Sunday serve as the baseline group for comparison. Our
serve as the dependent variables in our regression analysis. GMV mea­ choice of the baseline groups for both time of day and day of the week is
sures the total sales amount generated during a single live-streaming arbitrary and can easily be changed, but it does not affect the regression
shopping show. It is perhaps the most important measure that charac­ results for other independent variables.
terizes how successful a live-streaming shopping show is. This variable is
right-skewed, so we take logarithmic transformation to make its distri­ 4. Live streamer segmentation
bution more symmetric and improve model fit. The other measure,
FanGrowth, is calculated as the difference in the number of fans/fol­ We begin our exploratory analysis by examining the live streamer’s
lowers between the beginning and the end of the show divided by the role. To operationalize, we perform a clustering analysis at the live
number of fans/followers at the beginning (i.e., the percentage of streamer level using the K-Means clustering algorithm. For this purpose,
change). we construct 64 features in 13 categories to describe their characteristics
Given that the mean of this growth measure is often very small (i.e., who they are) and behavioral patterns (i.e., how they operate and
during the time window of a single show, we multiply it by 100 and perform) based on the live streamer and show the data described above.
express FanGrowth in percentage numbers. Intuitively, FanGrowth rep­ For individual characteristics, we consider the types (e.g., MCN sup­
resents the streamer’s incremental influence gained during a single live- ported or not, selling platforms), specialties (i.e., product specializa­
streaming show. FanGrowth can be negative if the number of fans drops tion), fan base, and experience. For behavioral patterns, we consider the
after a show. The minimum fan growth in our sample is − 1.594%, while intensity and scheduling preferences regarding the time of day, day of
the maximum is 37.948%. week, duration, time gaps, product selection strategies, and perfor­
The rest in Table 1 describe each show’s characteristics, which serve mance. Table 2 presents the summary statistics of these 64 features.
as the independent variables in our regression analysis. Products is the In the category of anchor type, MCN denotes whether a live streamer
number of unique items promoted by the live streamer in a show. Avg­ is supported by an MCN agency company, Retailer denotes whether a live
TransactionValue is the average transaction value across all products sold streamer is a retailer, Taobao denotes whether a live streamer runs at
in a live-streaming show. The variable is supposed to capture generally least one shop on Taobao, and Tmall denotes whether a live streamer
how expensive the products are in a live-streaming show. Product se­ runs at least one shop on Tmall. A live streamer can belong to more than
lection is an important decision for the live streamer and her/his agency one of these types simultaneously. For instance, many live streamers
company (if any). AvgProductDuration is the average time spent on a backed by an MCN company also have a shop on Taobao. In the anchor
product in seconds. In our sample, live streamers spend an average of specialty category, we include 15 indicator variables denoting whether a
585 s promoting each product in a show. HoursSinceLastShow is the live streamer specializes in each of the 15 product categories, respec­
number of hours since the same live streamer’s last show. Both the mean tively. Again, a live streamer can specialize in more than one product
and median of this variable are<24, implying that most live streamers category. In the anchor fans category, we include four features (mean,
run shows on a daily basis, especially during the shopping season in standard deviation, min, and max) to describe the distribution of a live
November and December. FanBase is the number of fans a live streamer streamer’s number of fans at the beginning of each show in the study
has before the show. Show# is the number of previous live-streaming period. In the anchor fans category, given that fan growth for every
single show is, on average, very small, we only include features that
Table 1 describe the distribution of the number of fans at the beginning of each
Summary Statistics of Show-Level Variables (N = 55,096). show. In the anchor experience category, we have one feature: the
logarithmic transformation of the number of prior shows hosted by a live
Variable Mean Stdev Min Median Max
streamer at the beginning of the study period.
Log(GMV) 11.851 2.507 − 3.912 11.854 23.760
In the show number category, we have the logarithmic trans­
FanGrowth (%) 0.059 0.325 − 1.594 0.011 37.948
Log(Products) 4.173 0.744 0 4.174 6.211 formation of the total number of shows hosted by a live streamer in the
Log(AvgTransactionValue) 5.263 1.244 − 4.605 5.144 13.698 study period. In the two categories of show start time and show end
Log(AvgProductDuration) 5.959 0.885 − 0.297 5.901 10.900 time, we have four features to capture the proportion of live streaming
Log(HoursSinceLastShow) 2.711 0.827 − 7.496 3.046 6.624 shows that either start or end in the four periods of six hours within a
Log(FanBase) 14.978 0.850 12.876 14.841 17.845
day, respectively. Each day is split into four periods of equal length: from
Log(Show#) 6.388 0.637 0.693 6.485 7.857
Double11 0.235 0.424 0 0 1 midnight to 6 am, 6 am to noon, noon to 6 pm, and 6 pm to midnight.
Double12 0.121 0.326 0 0 1 The maximum value of some features (e.g., Start_0_6) is<1, indicating
Morning 0.463 0.499 0 0 1 that the live streamer at the maximum point does not start or end all of
Afternoon 0.240 0.427 0 0 1
his/her shows in the corresponding 6-hour time period. In the category
Evening 0.263 0.440 0 0 1
Monday 0.147 0.354 0 0 1
of show day of the week, we have seven features corresponding to the
Tuesday 0.155 0.362 0 0 1
Wednesday 0.157 0.364 0 0 1
Thursday 0.122 0.328 0 0 1 3
These two time periods were chosen by various retailers in China, partic­
Friday 0.138 0.345 0 0 1
ularly those in Alibaba’s marketplaces, as two of the most prominent shopping
Saturday 0.144 0.351 0 0 1
festivals in a year.

4
H. Chen et al. Electronic Commerce Research and Applications 59 (2023) 101266

Table 2
Summary Statistics of 64 Features Used in Clustering (N = 1,000).
Category Variable Mean Stdev Min Median Max

Anchor Type MCN 0.046 0.210 0 0 1

Retailer 0.027 0.162 0 0 1
Taobao 0.199 0.399 0 0 1
Tmall 0.803 0.398 0 1 1
Anchor Specialty Auto 0.001 0.032 0 0 1
Babycare 0.059 0.236 0 0 1
Beauty 0.077 0.267 0 0 1
Clothing 0.243 0.429 0 0 1
Digital 0.040 0.196 0 0 1
Entertainment 0.012 0.109 0 0 1
FlowersPets 0.005 0.071 0 0 1
Food 0.035 0.184 0 0 1
Game 0.003 0.055 0 0 1
Health 0.011 0.104 0 0 1
Home 0.056 0.230 0 0 1
Jewelry 0.009 0.094 0 0 1
Men 0.038 0.191 0 0 1
Sports 0.018 0.133 0 0 1
Travel 0.003 0.055 0 0 1
Anchor Fans Avg_Log_Start_Fan 14.798 0.796 13.196 14.668 17.843
Stdev_Log_Start_Fan 0.018 0.027 0 0.010 0.394
Min_Log_Start_Fan 14.767 0.800 12.876 14.645 17.841
Max_Log_Start_Fan 14.823 0.790 13.652 14.687 17.845
Anchor Exp Log_Prior_Shows 6.095 0.807 0.693 6.239 7.770
Show Num Log_Shows 3.834 0.640 0.693 3.932 5.521
Show Start Time Start_0_6 0.023 0.066 0 0 0.950
Start_6_12 0.415 0.312 0 0.400 1
Start_12_18 0.287 0.238 0 0.239 1
Start_18_24 0.275 0.287 0 0.174 1
Show End Time End_0_6 0.497 0.292 0 0.481 1
End_6_12 0.034 0.077 0 0 0.787
End_12_18 0.139 0.159 0 0.067 0.926
End_18_24 0.330 0.275 0 0.273 1
Show Day of Week Monday 0.147 0.031 0 0.148 0.333
Tuesday 0.161 0.036 0 0.156 0.500
Wednesday 0.165 0.043 0.073 0.158 0.667
Thursday 0.116 0.035 0 0.119 0.333
Friday 0.134 0.036 0 0.138 0.471
Saturday 0.140 0.036 0 0.143 0.333
Sunday 0.136 0.037 0 0.138 0.400
Show Duration Avg_Log_Duration 10.171 0.451 8.887 10.191 11.318
Stdev_Log_Duration 0.443 0.198 0.004 0.425 1.978
Min_Log_Duration 8.805 1.050 4.394 8.932 11.010
Max_Log_Duration 10.923 0.389 9.346 10.976 12.209
Time Between Shows Avg_Log_Time_Gap 11.089 0.414 9.379 11.184 12.846
Stdev_Log_Time_Gap 0.602 0.267 0.000 0.594 1.885
Min_Log_Time_Gap 9.098 1.725 0.693 9.477 11.387
Max_Log_Time_Gap 12.388 0.725 10.759 12.214 14.813
Product Number Avg_Log_Products 4.180 0.688 1.875 4.154 5.791
Stdev_Log_Products 0.255 0.175 0.006 0.211 1.348
Min_Log_Products 3.490 1.005 0 3.555 5.529
Max_Log_Products 4.627 0.695 2.197 4.682 6.211
Product Price Avg_Log_ATV 5.268 1.076 1.686 5.153 8.753
Stdev_Log_ATV 0.479 0.390 0.009 0.380 3.178
Min_Log_ATV 3.786 1.755 − 4.605 3.945 8.149
Max_Log_ATV 6.261 1.374 2.265 6.166 13.698
Show Performance Avg_Log_GMV 11.915 1.239 7.687 11.802 18.810
Stdev_Log_GMV 2.047 0.630 0.491 1.988 4.414
Min_Log_GMV 7.143 2.389 − 3.912 7.157 16.659
Max_Log_GMV 16.145 1.390 13.297 15.985 23.760
Avg_Fan_Growth 0.068 0.196 − 0.070 0.032 4.807
Stdev_Fan_Growth 0.114 0.401 0 0.046 8.200
Min_Fan_Growth − 0.025 0.064 − 1.594 − 0.016 0.384
Max_Fan_Growth 0.575 1.926 − 0.005 0.194 37.948

proportion of shows hosted on each of the seven days in a week. In each shows in the study period for each streamer, we calculate the mean,
of the following four categories, we have four features (mean, standard standard deviation, minimum, and maximum of each performance
deviation, min, and max) to describe the distribution of four aspects of measure and thus have eight features in this category.
live streaming show behaviors, including show duration, the time gap We set the number of clusters (k) as 1 to 10 and then run the K-Means
between shows, product number, and product price. clustering algorithm on the sample of 1,000 live streamers, each with 64
In the category of show performance, we focus on two key perfor­ features. To avoid overweighting features with a larger mean in clus­
mance measures, log-transformed GMV and fan growth. To evaluate the tering, we normalize the features so that they fall within the range be­
distribution of these two performance measures across all live streaming tween 0 and 1. We adopt the elbow method to select the optimal number

5
H. Chen et al. Electronic Commerce Research and Applications 59 (2023) 101266

of clusters based on the sum of squared error (SSE) scores, and the 5. Determinants of live streaming success
optimal k turns out to be 3.4 Fig. 3 visualizes the three clusters in a 3-D
space after applying the principal component analysis (PCA). The PCA In this section, based on the clustering results, we move one step
helps reduce the dimensionality of the data from 64 features to three further to connect it to exploring the determinants of live streaming
principal components, which are uncorrelated with each other and success. As mentioned, the show performance measures include GMV
simultaneously maximize the variance in the data. The live streamers in and fan growth. GMV is widely employed by retailing platforms in
the same cluster are labeled with the same color (blue, brown, and practice,5 including the hosting platform in our study, Taobao Live.6
yellow, respectively) in the figure. The three clusters are well separated, Besides, an increasing number of recent studies have been using GMV as
implying that the clustering algorithm’s outcome is satisfactory. the dependent variable (e.g., Lyu et al., 2022, Ye et al., 2022, Baltescu
To understand how live streamers’ behaviors differ among three et al., 2022). Meanwhile, we also choose fan growth as the measure of
clusters, we calculate the variance of the three mean values for each of live-streaming success because the fan base, as the social capital for the
the 64 features and rank these variances in descending order. The live streamer, is documented to be significantly associated with the
intuition is that if the variance is substantially large across the mean purchases (e.g., Addo et al., 2021, Liu and Wang, 2023). In addition, the
values of a feature for the three clusters, this feature helps differentiate live streamers, as the influencer in the live-streaming context, are ex­
live streamers’ characteristics or behavior patterns in different clusters. pected to be strongly motivated to grow the fan base and maintain a
Conversely, a small variance indicates that live streamers from different long-term reputation (Haenlein et al., 2020).
clusters are hardly differentiable in the corresponding feature. We classify the determinants of live streaming success into four
Table 3 presents the results of this analysis. We only include the top groups: product selection (including the number of products and the
15 features with the largest variance across mean values. Based on the average price of products sold in a show), show scheduling (average
top 5 features, which include three features in the anchor type category time spent on a product and the time gap between two shows), fan base
and one feature in the anchor specialty category, we assign a label to (number of fans), and control variables (live streaming experience,
each of the three clusters, “Taobao + MCN Players” for Cluster 1, “Tmall shopping season, time of day, and day of week). Specifically, little has
+ Non-Clothing Players” for Cluster 2, and “Tmall + Clothing Players” been documented in the literature on the role of show scheduling. As an
for Cluster 3. There are 203, 641, and 156 live streamers in each cluster, e-commerce live streaming show in our data averagely runs for more
respectively. Below we analyze the behavior differences across these than 10 h and carries more than 60 products, the impact of scheduling
three clusters and derive several interesting insights. First, Cluster-1 arrangement is nontrivial but remains largely unknown in the literature.
streamers (Taobao + MCN Players) start 48% of their live-streaming Some recent work has started to examine the issue of mitigating
shows in the evening between 6 pm and midnight and 26.8% of the streamer fatigue, using artificial intelligence to assist live streamers
shows in the morning between 6 am and noon. By contrast, live (Wang et al., 2021). The show arrangement also connects to the litera­
streamers in Clusters 2 and 3, who typically have shops on Tmall, are ture on shelf layout management in brick-and-mortar stores (e.g., Gu
more likely to start live-streaming shows in the morning but less likely to and Liu, 2013). The configurations of the other three groups are more
start the shows in the evening. Cluster 1 ends 40.2% of the shows in the widely studied, such as product fit (Park and Lin, 2020, Song et al.,
evening from 6 pm to midnight, while Clusters 2 and 3 end only 31.8% 2021), fan interactions (Liu et al., 2021, Wang et al., 2021, Song et al.,
and 28.7% of their shows in the same evening period, respectively. 2021), and live-streamer popularity (Zhao et al., 2021). Fig. 4 provides a
These behavioral patterns suggest that Taobao + MCN players differ summary of our empirical framework to approach this problem.
significantly from Tmall players in their scheduling choice for shows. To account for potential correlations between observations associ­
The tendency to utilize more evening time in scheduling by Taobao + ated with the same live streamer, we cluster the standard errors by the
MCN players is perhaps because they are individuals or small companies anchor in the fixed-effect model. Tables 4 and 5 present the regression
and thus can be more flexible than Tmall players, typically larger analysis results for the two performance measures, respectively. In each
companies. Second, on average, live streamers selling clothing products table, we first analyze the whole sample (Columns 1 to 2) and then
in Cluster 3 sell more products in each show or almost double the explore how the determinants may have a different effect on the success
average number of products sold by non-clothing sellers. This pattern is of live streamers in each of the three clusters (Columns 3 to 8).
also roughly consistent for both the minimum and maximum numbers of When success is measured by GMV in Table 4, we have the following
products sold in a show. The average and minimum product prices are interesting findings from the analysis of the whole sample in Columns
higher for Cluster 3 than the other two clusters, suggesting that Cluster 3 (1) and (2). First, both the number of products and average transaction
players sell more expensive products. Live streamers in Cluster 3 also are value are positively associated with GMV, and the coefficients are sta­
more likely to specialize in Men products than those in Clusters 1 and 2. tistically significant at the 1% level. In terms of economic significance, a
Third, the shows hosted by Tmall sellers in Clusters 2 and 3 are longer 1% increase in the number of products promoted in a live-streaming
than those by Taobao or MCN sellers in Cluster 1, as indicated by the show is associated with a 2.196% increase in GMV. A 1% increase in
average and maximum show durations. The difference is around 2 h on the average transaction value is associated with a 0.808% increase in
average and 4 h at maximum. GMV. These results imply that selling more products and more expensive
An important observation from Table 3 is that none of the eight ones helps increase the overall GMV of a live-streaming show.
features in the show performance category (e.g., Avg_Log_GMV, Avg_­ Second, the coefficients of Log(AvgProductDuration) are positive and
Fan_Growth) is among the top features that differentiate the three clus­ statistically significant at the 1% level, suggesting that spending more
ters. This implies that live streamers from different clusters, on average, time on each product in a live-streaming show can help improve GMV.
are similar when measured by show performance metrics such as GMV Specifically, a 1% increase in the average product duration is associated
and fan growth, indicating that further analysis is necessary to identify with a 1.391% increase in GMV. However, a long time gap between two
the determinants of live streaming show success in terms of GMV and fan live-streaming shows is harmful to GMV, as the coefficient estimates of
growth. We explore these directions with a regression analysis in the Log(HoursSinceLastShow) are negative and statistically significant at the
next section. 1% level. This result supports the industry’s standard practice of running
live-streaming shows frequently on a daily basis.

5
https://rechargepayments.com/blog/understanding-gmv-in-ecommerce/.
6
https://www.nasdaq.com/press-release/taobao-live-enriches-its-
4
The plot of the SSE scores for different k’s is available upon request. ecosystem-to-fuel-further-success-for-partners-2021–04-28.

6
H. Chen et al. Electronic Commerce Research and Applications 59 (2023) 101266

Fig. 3. Visualization of the Clusters based on the Dimensions Identified by PCA.

Third, to our surprise, the effect of the fan base is statistically

Table 3
insignificant on GMV in Column (1). To explain this counterintuitive
Characteristics and Behavioral Patterns of Live Streamers in Each Cluster.
result, we examine how anchor experience, which is measured by the
Cluster 1 Cluster 2 Cluster 3 number of prior live-streaming shows (Show#), may moderate the effect
Taobao + MCN Tmall + Non- Tmall + Clothing
of the fan base on GMV. In Column (2), we find that the coefficient of the
Players Clothing Players Players
interaction term Log(FanBase) × Log(Show#) is negative, while the main
Tmall 3.4% 99.8% 100%
effect of Log(FanBase) becomes positive; both coefficients are statisti­
Taobao 95.1% 0.3% 2.6%
Clothing 42.9% 0 100%
cally significant at the 1% level. This finding indicates that the effect of
Start_18_24 48% 22.9% 19.4% the fan base on GMV is larger for new live streamers, but it decreases as a
MCN 22.7% 0 0 live streamer hosts more shows and becomes more experienced.
Start_6_12 26.8% 44.2% 49.5% Fourth, as to the results for control variables, we first find that, on
Avg_Products 63.82 57.05 117.68
average, the number of prior shows hosted by a live streamer does not
Max_Products 118.87 85.46 174.34
Min_products 24.09 31.34 58.85 affect the GMV of a future show after controlling for other factors in
Men 2.0% 1.7% 14.7% Column (1). This is a bit surprising, as experienced anchors are generally
Avg_duration 5.97 7.47 8.37 expected to produce more sales. When we consider the interaction be­
(hour)
tween the fan base and anchor experience in Column (2), the result
Max_duration 12.04 16.45 16.11
(hour)
shows that a live streamer can generate more GMV as she hosts more
End_18_24 40.2% 31.8% 28.7% shows only when the fan base is low. GMV is on average higher for shows
Avg_ATV (CNY) 151.11 190.57 288.88 hosted during the two shopping festivals of Double 11 and Double 12.
Min_ATV (CNY) 34.16 38.55 106.48 Compared with shows that start within the period of from midnight to 6
# Live Streamers 203 641 156
am, shows that start in the morning, afternoon, or evening all generate a
Note: features are ranked by the variance of the mean values of the three clusters higher GMV, and based on the relative magnitudes of the coefficient
in descending order. estimates, we can infer that shows that start in the evening on average
generate the highest GMV. Among the days in a week, shows hosted on

Fig. 4. Determinants of Live Streaming Success: Empirical Framework.

7
H. Chen et al. Electronic Commerce Research and Applications 59 (2023) 101266

Table 4
Determinants of Live Streaming E-commerce Success – GMV.
Log(GMV)

All Cluster 1 Cluster 2 Cluster 3

Taobao + MCN Players Tmall + Non-Clothing Players Tmall + Clothing Players

(1) (2) (3) (4) (5) (6) (7) (8)

Log(Products) 2.200*** 2.196*** 1.739*** 1.741*** 2.395*** 2.393*** 2.166*** 2.165***

(0.063) (0.063) (0.126) (0.127) (0.072) (0.072) (0.160) (0.159)
Log 0.808*** 0.808*** 0.869*** 0.869*** 0.760*** 0.760*** 0.663*** 0.665***
(AvgTransactionValue) (0.029) (0.029) (0.042) (0.042) (0.035) (0.035) (0.123) (0.123)
Log(AvgProductDuration) 1.391*** 1.392*** 1.563*** 1.564*** 1.331*** 1.332*** 1.229*** 1.225***
(0.031) (0.031) (0.093) (0.094) (0.033) (0.033) (0.094) (0.094)
Log(HoursSinceLastShow) − 0.140*** − 0.140*** − 0.179*** − 0.178*** − 0.092*** − 0.092*** − 0.192*** − 0.194***
(0.016) (0.016) (0.034) (0.034) (0.018) (0.018) (0.037) (0.036)
Log(FanBase) − 0.820 (0.671) 3.370*** − 0.024 (0.774) − 0.565 (1.263) 0.475 (0.667) 2.166 (1.467) 2.632 (2.975) 15.711***
(1.205) (5.682)
Log(FanBase) × Log − 0.768*** 0.119 (0.283) − 0.295 (0.246) − 2.368***
(Show#) (0.201) (0.866)
Log(Show#) 0.433 (0.350) 11.813*** 0.004 (0.471) − 1.686 (4.089) 0.586 (0.397) 5.047 (3.818) − 1.633 (1.517) 35.627**
(2.984) (14.156)
Double11 1.674*** 1.672*** 1.766*** 1.766*** 1.581*** 1.579*** 1.976*** 1.966***
(0.037) (0.037) (0.077) (0.077) (0.046) (0.046) (0.100) (0.100)
Double12 0.573*** 0.572*** − 0.640*** − 0.640*** 1.105*** 1.104*** 0.310** 0.317**
(0.062) (0.062) (0.086) (0.086) (0.075) (0.075) (0.145) (0.145)
Morning 0.740*** 0.740*** − 0.390 (0.484) − 0.389 (0.484) 0.782*** 0.781*** 0.699*** 0.696***
(0.106) (0.106) (0.120) (0.120) (0.140) (0.138)
Afternoon 0.648*** 0.650*** − 0.474 (0.496) − 0.473 (0.496) 0.683*** 0.683*** 0.835*** 0.834***
(0.110) (0.111) (0.123) (0.123) (0.157) (0.155)
Evening 1.199*** 1.203*** 0.647 (0.502) 0.648 (0.502) 1.017*** 1.018*** 1.294*** 1.296***
(0.115) (0.115) (0.126) (0.126) (0.190) (0.190)
Monday − 0.272*** − 0.270*** − 0.221*** − 0.221*** − 0.263*** − 0.263*** − 0.384*** − 0.386***
(0.019) (0.019) (0.041) (0.041) (0.024) (0.024) (0.046) (0.047)
Tuesday 0.603*** 0.605*** 0.358*** 0.357*** 0.674*** 0.674*** 0.670*** 0.666***
(0.025) (0.025) (0.049) (0.049) (0.033)
(0.033) (0.064) (0.064)
Wednesday 0.249*** 0.244*** 0.200*** 0.201*** 0.273*** 0.271*** 0.170*** 0.173***

(0.024) (0.024) (0.047) (0.047) (0.032) (0.032) (0.062) (0.063)

Thursday 0.276*** 0.272*** 0.193*** 0.194*** 0.293*** 0.291*** 0.320*** 0.321***

(0.026) (0.026) (0.044) (0.045) (0.036) (0.036) (0.072) (0.072)

Friday − 0.070*** − 0.070*** − 0.074* − 0.074* − 0.074*** − 0.074*** − 0.052 − 0.049

(0.021) (0.021) (0.042) (0.042) (0.027) (0.027) (0.053) (0.054)

Saturday 0.457*** 0.455*** 0.391*** 0.391*** 0.470*** 0.469*** 0.535*** 0.535***

(0.018) (0.018) (0.037) (0.037) (0.023) (0.023) (0.043) (0.043)

Week Dummies Yes Yes Yes Yes Yes Yes Yes Yes
Anchor Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes
#Observations 55,096 55,096 11,997 11,997 35,072 35,072 8,027 8,027
#Anchors 1,000 1,000 203 203 641 641 156 156
Adj. R2 0.575 0.575 0.674 0.674 0.598 0.598 0.479 0.480

Notes: (1) robust standard errors clustered by anchor are in parentheses; (2) * p < 0.10, ** p < 0.05, *** p < 0.01.

Tuesday, Saturday, Thursday, and Wednesday generate a higher GMV implies that Taobao and MCN players pay much less attention to the
than shows hosted on Sunday, which serves as the baseline group, while Double 12 shopping festival than the Tmall players. In addition, the
shows hosted on Monday and Friday generate a lower GMV. coefficients of Morning, Afternoon, and Evening are all statistically
In Columns (3) to (8) of Table 4, we explore the determinants of GMV insignificant at the 10% level for Cluster 1, indicating that there is no
for each cluster and investigate the heterogeneity across different live time of day effect for Taobao and MCN players.
streamers. While most coefficient estimates remain similar across the
three clusters, two exceptions are worth noting. The first one is that 6. Conclusion
there is no significant effect of fan base on GMV for live streamers in
Clusters 1 and 2, even among their early shows, as the interaction term is Live-streaming e-commerce has gained increasing attention from
also insignificant. The effect of fan base on GMV only exists for live academia and practice in recent years. Understanding the new role of
streamers in Cluster 3 or Tmall sellers in the Clothing product category. live streamers is an important step to fully leveraging the potential of
The effect size is much larger for Cluster 3 than in the whole sample. This live video streaming in e-commerce. This paper provides the first set of
finding implies that the benefit of having a large fan base in live empirical evidence to characterize the behaviors of live streamers and
streaming is limited to the early shows hosted by certain live streamers identify the success factors of this new role in e-commerce.
only but not all. The second difference is related to the unique behavior Our study offers important managerial implications for practitioners
of live streamers in Cluster 1. The coefficient of Double12 is negative and in the booming live-streaming e-commerce industry. First, before busi­
statistically significant at the 1% level for Cluster 1, while the result is nesses decide whether they should collaborate with independent live
the opposite for the other two clusters and the whole sample. This streamers or become live streamers themselves, it is crucial to learn how

8
H. Chen et al. Electronic Commerce Research and Applications 59 (2023) 101266

Table 5
Determinants of Live Streaming E-commerce Success – Fan Growth.
FanGrowth (%)

All Cluster 1 Cluster 2 Cluster 3

Taobao + MCN Players Tmall + Non-Clothing Players Tmall + Clothing Players

(1) (2) (3) (4) (5) (6) (7) (8)

Log(Products) 0.073*** 0.073** 0.098*** 0.101*** 0.064*** 0.062*** 0.032*** 0.032***

(0.009) (0.009) (0.027) (0.028) (0.005) (0.005) (0.004) (0.004)
Log 0.004* (0.003) 0.004 (0.003) 0.001 0.010) 0.002 0.003 (0.002) 0.004 (0.003) 0.000 (0.001) 0.000 (0.001)
(AvgTransactionValue) (0.010)
Log(AvgProductDuration) 0.050*** 0.050*** 0.042*** 0.042*** 0.053*** 0.054*** 0.025*** 0.025***
(0.003) (0.003) (0.008) (0.008) (0.003) (0.003) (0.003) (0.003)
Log(HoursSinceLastShow) − 0.001v − 0.001 0.003 (0.006) 0.003 0.007) − 0.002 − 0.002 − 0.003*** − 0.004***
(0.003) (0.003) (0.004) (0.005) (0.001) (0.001)
Log(FanBase) − 1.220*** − 0.908 − 3.893*** − 5.146*** − 0.508* 1.919 (1.979) − 0.412 (0.344) 0.118 (0.286)
(0.462) (1.664) (1.271) (1.679) (0.300)
Log(FanBase)×Log 0.125 (0.236) 0.973 (3.792) 0.275 (0.232) − 0.424 − 0.096***
(Show#) (0.306) (0.033)
Log(Show#) − 0.057 0.211 (0.415) − 3.699 (3.408) 0.384(0.293) 6.788(4.843) 0.007(0.062) 1.518***
(0.242) (0.533)
Double11 0.011(0.009) 0.011(0.008) − 0.030(0.032) − 0.030(0.032) 0.022*** 0.019*** 0.020*(0.012) 0.020*(0.011)
(0.008) (0.007)
Double12 − 0.007(0.005) − 0.007 − 0.017(0.014) − 0.018(0.014) 0.000(0.005) − 0.002 − 0.011(0.009) − 0.011(0.009)
(0.005) (0.004)
− 0.001(0.008) − 0.001 − 0.023(0.029) − 0.022(0.031) 0.011(0.008) 0.011(0.008) 0.003(0.005) 0.003
(0.008) (0.005)
Afternoon − 0.005(0.007) − 0.004 − 0.027(0.030) − 0.025(0.031) 0.006(0.008) 0.007(0.008) 0.004(0.007) 0.004(0.007)
(0.008)
0.003(0.007) 0.003(0.007) − 0.009(0.029) − 0.007(0.031) 0.011(0.007) 0.013(0.008) 0.004(0.006) 0.004(0.006)
Monday 0.001(0.003) 0.001(0.003) − 0.010(0.007) − 0.011(0.007) 0.004(0.004) 0.004(0.004) 0.000(0.005) 0.000(0.005)
Tuesday 0.030*** 0.030*** 0.019**(0.008) 0.017**(0.008) 0.034*** 0.034*** 0.017*** 0.017***
(0.005) (0.005) (0.007) (0.007) (0.005) (0.005)
Wednesday 0.015**(0.007) 0.015** 0.006(0.014) 0.009(0.015) 0.026*** 0.023*** 0.003(0.002) 0.003(0.002)
(0.006) (0.009) (0.008)
Tursday 0.000(0.006) 0.000(0.005) − 0.022*(0.013) − 0.021(0.013) 0.013** 0.011** 0.001(0.004) 0.001(0.004)
(0.006) (0.004)
Friday − 0.006(0.004) − 0.006* − 0.018** − 0.017** − 0.001 − 0.001 − 0.003(0.004) − 0.003(0.004)
(0.004) (0.009) (0.008) (0.004) (0.004)
Saturday 0.013*** 0.013*** 0.024(0.016) 0.025(0.017) 0.014*** 0.013*** 0.005*(0.003) 0.005*(0.003)
(0.004) (0.004) (0.005) (0.004)
Week Dummies Yes Yes Yes Yes Yes Yes Yes Yes
Anchor Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes
#Observations 55,096 55,096 11,997 11,997 35,072 35,072 8,027 8,027
#Anchors 1,000 1,000 203 203 641 641 156 156
Adj. R2 0.180 0.180 0.300 0.302 0.091 0.100 0.136 0.137

Notes: (1) robust standard errors clustered by anchor are in parentheses; (2) * p < 0.10, ** p < 0.05, *** p < 0.01.

independent live streamers supported by MCN agencies may exhibit is conducted at the show level. We do not have more fine-grained data
different behaviors from those affiliated with well-established brands or about how live streamers run each of their shows, such as how they
businesses of different sizes. Live streamers selling clothing products are describe different products and their promotional techniques. These
particularly distinct and differ from live streamers operating in other micro-level behaviors could potentially have a significant effect on the
product categories. Second, there are several ways for live streamers to success of live streamers. Third, we also have limited information about
increase GMV or fan growth. Selling more products and spending more other players in live-streaming e-commerce. For instance, we do not
time on each product in a live-streaming show help improve both per­ observe detailed information about the viewers and buyers participating
formance metrics. Running the live streaming show at different hours in each live-streaming show. Thus we cannot characterize what kind of
within a day and on different days within a week could also affect the viewers a live streamer attracts and how successful the live streamer is in
GMV to some extent. Third, a high fan base does not necessarily translate converting them into buyers. We also do not have information about the
to more GMV. The positive effect of the fan base on GMV only exists for specific products sold in these shows, such as their prices, discounts, and
the earlier shows, and this positive fan base effect applies only to Tmall available quantities. Measuring live streamers’ success at selling
live streamers in the clothing product category. This finding could imply different types of products could also yield additional insights.
that experienced live streamers should focus more on the quality of their Live-streaming e-commerce is an exciting new area for information
fans instead of the size of the fan base. If the quality of live-streaming systems researchers, and we believe there are several interesting di­
show fans is higher, even a smaller fan base could potentially yield a rections for future research. The first promising direction is to produce
higher GMV. fine-grained measures of live streamers’ behaviors by indexing and la­
Although the dataset analyzed in this study is already quite rich and beling the streaming videos, possibly through visual and audio deep
unique, there are still several limitations. The first limitation is with the learning technologies. Much information can be extracted from the
sample selection. Our dataset comes from one of the leading live- videos, such as the live streamer’s facial expressions, verbal transcripts,
streaming e-commerce platforms in China. Our sample of the top voice tones, body gestures, sales techniques, presentation styles, and
1,000 live streamers also includes the relatively more popular ones in even the design of the live-streaming showroom. The second promising
this industry. Therefore, it is unclear whether our findings generalize to direction is product selection choice for each show. Popular live
other platforms or niche live streamers. Second, our regression analysis streamers often have a large pool of products to select from for their

9
H. Chen et al. Electronic Commerce Research and Applications 59 (2023) 101266

shows. Their ability to select the most promising products that fit their The Drum. 2020. Shopify... for Livestreaming E-commerce: The Drum Editorial Team’s
Best of 2020. Available at https://www.thedrum.com/news/2020/12/31/shopify-
streaming styles is a key differentiator in competition. In addition, how
livestreaming-e-commerce-the-drum-editorial-team-s-best-2020 (accessed date June
to select the right mix of products to maximize the overall GMV of a live- 17th, 2021).
streaming show is also an unanswered question. The third promising Fei, M., Tan, H., Peng, X., Wang, Q., Wang, L., 2020. Promoting or attenuating? An eye-
direction is the contract design between independent live streamers and tracking study on the role of social cues in e-commerce livestreaming. Decision
Support Syst. 143 (3), 113466.
product sellers. There are ongoing disputes on the unexpectedly high Gao, X., Xu, X., Tayyab, S.M.U., Li, Q., 2021. How the live streaming commerce viewers
product returns in live-streaming e-commerce (e.g., Cheng, 2020). Live process the persuasive message: An ELM perspective and the moderating effect of
streamers often blame sellers for inferior-quality products. In contrast, mindfulness. Electron. Commer. Res. Appl. 49, 101087.
Haenlein, M., Anadol, E., Farnsworth, T., Hugo, H., Hunichen, J., Welte, D., 2020.
sellers criticize live streamers for faking the viewer counts and sales Navigating the new era of influencer marketing: How to be successful on Instagram,
numbers. This might give rise to a new context of incomplete contract TikTok, & Co. Calif. Manage. Rev. 63 (1), 5–25.
design, as no traditional contract form can effectively eliminate the Huang, S., Aral, S., Hu, Y.J., Brynjolfsson, E., 2020. Social advertising effectiveness
across products: A large-scale field experiment. Mark. Sci. 39 (6), 1142–1165.
conflict. The last promising direction is to look into the operations of Kacen, J.J., Lee, J.A., 2002. The influence of culture on consumer impulsive buying
new supporting roles in the live-streaming industry, such as how MCN behavior. J. Consumer Psychol. 12 (2), 163–176.
agencies and data analytics providers add value. For example, it is Li, R., S. Peng. 2021. Why Live Commerce? – A New Growth Point of Online Shopping.
Available at https://polymathv.com/why-live-commerce-is-the-future-of-online-
important to understand how MCN agencies contribute to the success of shopping/ (accessed date June 16th, 2021).
a live streamer, and there is potential knowledge spillover among live Lin, Y., Yao, D., Chen, X., 2021. Happiness begets money: Emotion and engagement in
streamers working with the same MCN agency. live streaming. J. Mark. Res. 58 (3), 417–438.
Liu F, Wang W, Huang W, 2021. Monetizing live videos: Do monetary behaviors affect
While our exploration efforts yield the practical insights above and
live social media engagement? Working Paper.
demonstrate the exciting research opportunities in the context of live- Liu, B., Wang, W., 2023. Live commerce retailing with online influencers: Two business
streaming e-commerce, our study is only the beginning of this line of models. Int. J. Prod. Econ. 255, 108715.
work. It can be further extended, such as pursuing causal inference Lu, B., Chen, Z., 2021. Live streaming commerce and consumers’ purchase intention: An
uncertainty reduction perspective. Inf. Manage. 58 (7), 103509.
based on more sophisticated empirical strategies and looking into the Luo, H., Cheng, S., Zhou, W., Yu, S., Lin, X., 2021. A study on the impact of linguistic
theoretical aspects of the consumers’ and live-streamers’ behaviors. persuasive styles on the sales volume of live streaming products in social e-
commerce environment. Mathematics 2021 (9), 1576.
Lyu, W., Qi, Y., Liu, J., 2022. Proliferation in Live Streaming Commerce, and Key
Declaration of Competing Interest Opinion Leader Selection. Electronic Commerce Research. Accepted.
McDonald, W.J., 1995. Home shopping channel customer segments: A cross-cultural
The authors declare that they have no known competing financial perspective. J. Direct Mark. 9 (4), 57–67.
Park, H.J., Lin, L.M., 2020. The effects of match-ups on the consumer attitudes toward
interests or personal relationships that could have appeared to influence internet celebrities and their live streaming contents in the context of product
the work reported in this paper. endorsement. J. Retail. Consumer Services 52 (1), 101934.
Qi, A., S. Sethi, L. Wei, J. Zhang. 2021. Strategic overcapacity in live-streaming platform
selling. Working paper.
Data availability Song D, Chen X, Guo Z, Gao X. 2021. What drives sales of e-commerce live streaming?
Evidence from Taobao. Working Paper.
The authors do not have permission to share data. Stephens, D.L., Hill, R.P., Bergman, K., 1996. Enhancing the consumer-product
relationship: Lessons from the QVC home shopping channel. J. Bus. Res. 37 (3),
193–200.
References Sun, Y., Shao, X., Li, X., Guo, Y., Nie, K., 2019. How live streaming influences purchase
intentions in social commerce: An IT affordance perspective. Electron. Commer. Res.
Addo, P., Fang, J., Asare, A., Kulbo, N., 2021. Customer engagement and purchase Appl. 37, 100886.
intention in live-streaming digital marketing platforms. Serv. Ind. J. 41 (11–12), Teng, B. 2020. China’s Betting Big on the Latest Live Streaming E-commerce Boom.
767–786. Available at https://english.ckgsb.edu.cn/blog/chinas-betting-big-on-the-latest-live-
Ba, S., Pavlou, P.A., 2002. Evidence of the effect of trust building technology in streaming-e-commerce-boom/ (accessed date June 17th, 2021).
electronic markets: Price premiums and buyer behavior. MIS Q. 26 (3), 243–268. Tong, J., 2017. A study on the effect of web live broadcast on consumers’ willingness to
Baltescu, P., H. Chen, N. Pancha, A. Zhai, J. Leskovec, and C. Rosenberg. 2022. ItemSage: purchase. Open J. Bus. Manage. 2017 (5), 280–289.
Learning product embeddings for shopping recommendations at Pinterest. In Urbany, J.E., Bearden, W.O., Weilbaker, D.C., 1988. The effect of plausible and
Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data exaggerated reference prices on consumer perceptions and price search. J. Consum.
Mining (KDD ’22). New York, NY, USA, 2703–2711. Res. 15 (1), 95–110.
Borak, M. 2020. Live Streaming E-Commerce Is Booming in China after Covid-19 Walmart. 2021. Walmart Doubles Down on TikTok Shopping, Hosts All-New Live Stream
Pandemic. South China Morning Post. Available at https://www.scmp.com/abacus/ Shopping Event. Available at https://corporate.walmart.com/newsroom/2021/03/
culture/article/3090201/live-streaming-e-commerce-booming-china-after-covid-19- 09/walmart-doubles-down-on-tiktok-shopping-hosts-all-new-live-stream-shopping-
pandemic (accessed date March 24th, 2021). event (accessed date June 17th, 2021).
Brynjolfsson, E., Smith, M.D., 2000. Frictionless commerce? A comparison of internet Wang L, He Y, Ding J, Huang N, Hong Y, Guo X, Liu D, Chen G. 2021. Effectiveness of AI
and conventional retailers. Manag. Sci. 46 (4), 563–585. assistance in live streaming. A randomized field experiment. ICIS 2021 Proceedings.
Bu, L., L. Xu, D. Zipser. 2019. The Innovations behind China’s Singles Day Shopping Warden, C.A., Huang, S.C.T., Liu, T.C., Wu, W.Y., 2008. Global media, local metaphor:
Phenomenon. McKinsey Report. Available at https://www.mckinsey.com/~/media/ Television shopping and marketing-as-relationship in America, Japan, and Taiwan.
McKinsey/Featured%20Insights/China/The%20innovations%20behind%20Chinas J. Retail. 84 (1), 119–129.
%20Singles%20Day%20shopping%20phenomenon/The-innovations-behind- Wongkitrungrueng, A., Assarut, N., 2020. The role of live streaming in building
Chinas-Singles-Day-shopping-phenomenon-vF.pdf (accessed date June 16th, 2021). consumer trust and engagement with social commerce sellers. J. Bus. Res. 117,
Businesswire. 2021. Alibaba Group Announces March Quarter and Full Fiscal Year 2021 543–556.
Results. Available at https://www.businesswire.com/news/home/ WSJ. 2020. Live-streaming Craze Turns into a Lifeline for China’s Stores. The Wall Street
20210513005533/en/Alibaba-Group-Announces-March-Quarter-and-Full-Fiscal- Journal. Available at https://www.wsj.com/articles/live-streaming-craze-turns-into-
Year-2021-Results (accessed date May 27th, 2021). a-lifeline-for-chinas-stores-11594287003 (accessed date June 16th, 2021).
Chen, C., Y. Hu, Y. Lu, Y. Hong. 2021. Everyone can be a star: Understanding the role of Xu, K., Chan, J., Ghose, A., Han, S.P., 2017. Battle of the channels: The impact of tablets
live video streaming in online retail. Working paper. on digital commerce. Manag. Sci. 63 (5), 1469–1492.
Chen, Y. J., G. Gallego, P. Gao, Y. Li. 2021. Position auctions with endogenous product Yang, R., H. Tung. 2020. Is Live Streaming the New Frontier of eCommerce? Available at
information: Why live-streaming advertising is thriving. Working paper. https://en.ggvc.com/insights/is-live-streaming-the-new-frontier-of-ecommerce/
Cheng, E. 2020. Livestreaming Is Taking off in China, But It is Not Driving Much Sales in (accessed date June 17th, 2021).
A Market Still Reeling from Coronavirus. Available at https://www.cnbc.com/2020/ Ye, B., Yang, S., Hu, B., Zhang, Z., He, Y., Huang, K., Zhou, J., Fang, Y., 2022. Gaia:
04/29/livestreaming-drives-fraction-of-sales-in-coronavirus-hit-china-market.html Graph neural network with temporal shift aware attention for gross merchandise
(accessed date June 10th, 2021). value forecast in e-commerce. In: Proceedings of the IEEE 38th International
Chevalier, J.A., Mayzlin, D., 2006. The effect of word of mouth on sales: Online book Conference on Data Engineering (ICDE). Kuala Lumpur, pp. 3320–3326.
reviews. J. Mark. Res. 43 (3), 345–354. Zhang, J. 2021. Alibaba’s Taobao Live Reports Over US$60 Billion in GMV in 2020 as
CNBC, 2021. US Retailers Scramble to Crack the Code on Livestream Shopping Events. China Intensifies Scrutiny of Live-stream E-commerce Industry. Available at https://
Available at https://www.cnbc.com/2021/05/03/retailers-from-bloomingdales-to- www.scmp.com/tech/big-tech/article/3131447/alibabas-taobao-live-reports-over-
petco-test-livestreaming-to-win-sales.html (accessed date June 17th, 2021). us60-billion-gmv-2020-china (accessed date June 17th, 2021).
Cortese, J., Rubin, A.M., 2010. Uses and gratifications of television home shopping. Zhao Y, Hu Y, Hong Y, Westland C. 2021. Understanding characteristics of popular
Atlantic J. Commun. 18 (2), 89–109. streamers on live streaming platforms: Evidence from Twitch.tv. Working Paper.

10