Reference:
Song, X., Fu, M., Fang, J., Cai, Z., Tan, C.-W., Lim, E. T. K., & Chong, A. Y. L. (2025). Turning the wheels of engagement: Evidence from entertainment live streaming. Journal of the Academy of Marketing Science: Official Publication of the Academy of Marketing Science, 53(4), 1055–1080. https://doi-org.devinci.idm.oclc.org/10.1007/s11747-024-01020-1
Keywords: Customer engagement,Emergent process,Engagement transition,live streaming,Gratuity,Scheduling strategy, Markov chain
Summary :
The article investigates how customer engagement evolves during entertainment live streaming and how influencers can strategically manage this engagement to increase monetization.
The study adopts a quantitative empirical methodology based on:
- a Markov chain model
- a Multilevel Linear Model (MLM)
Development :
Live streaming : form of real-time digital entertainment where influencers (streamers) interact directly with viewers through activities such as chatting, gaming, and talent performances
Markov chain model is used in this article to analyse customer’s engagement transitions during the live streams.
it tracks how viewers move from a specific engagement state to another one over time during the live streams.and then calculates the propability of the engagement fluctuation.
These engagement states are specified as 3 :
- Commenting (low)
- Nonmonetary gifting
- Paid gifting (high)
Multilevel Linear Model (MLM) analyzes the effects of scheduling strategies on engagement transitions and gratuities.
Data was collected from a major Chinese entertainment live-streaming platform throughout 3 months
he final dataset included:
- 91,148 engagement records
- 18,965 viewers
- 9,995 streamers
Key findings :
- Engagement fluctuates and isnt static : it goes through
- escalation
- de-escalation
- repeated behaviors.
Escalation Increases Gratuities
When viewers transitioned toward higher engagement states (e.g., commenting → paid gifting), gratuities increased.
Limitations:
The study focuses only on: entertainment live streaming and one Chinese platform so the results cannot be generalised.