Our sampled firms were digital start-ups, which are defined as venture companies that create value through various digital technologies, with the internet start-up being one of the most common (Nambisan, 2017). Both business model research and demand-side perspective emphasize the central role of consumers in value creation. With a population of 1.4 billion, demand-driven business models are flourishing in China's digital economy. For example, the business models of Alibaba, Tencent, and TikTok are all demand-driven but have different features. Therefore, China represents an ideal context for this study.

We began our data collection by distributing a questionnaire to digital start-ups in China. The following procedures were used to create the questionnaire. First, all items were developed based on a systematic literature review, and the scale was revised according to the characteristics of Chinese digital start-ups. Second, we conducted several rounds of interviews with entrepreneurs to revise the questionnaire, ensuring that all scales and items were accurate and easily understood. Finally, we performed a pilot study on ten entrepreneurs from digital start-ups for further feedback.

The survey was conducted in 20172. We collected data from five provinces or municipalities of China—Beijing, Zhejiang, Shandong, Hunan, and Anhui. To improve response rates and avoid potential misunderstandings, we chose to conduct an on-site survey over online-based or mail-based survey (Sheng, Zhou & Li, 2011). Five hundred digital start-ups located in 34 incubators or start-up parks were randomly selected as samples. The telephone numbers and e-mail addresses of the selected firms were obtained from government agencies. With this information, we contacted these firms and invited them to participate in the survey. First, with the help of local governments and management committees of incubators, an interviewer contacted the start-up founders seeking consent and appointment for the interview. Second, in the face-to-face field survey, the interviewer explained the study purpose and the questionnaire to the respondents to avoid possible confusions. Finally, the interviewer thoroughly reviewed the completed questionnaires.

The questionnaire was filled by 389 firms, resulting in a response rate of 77.8%. The date and total time spent on the questionnaire were recorded for each respondent. Furthermore, 104 responses were excluded from the study in any of the following cases: (i) one or two respondents completed the questionnaire in less than 10 minutes, implying that they were distracted and the questionnaire might have been of poor quality; (ii) the questionnaire was incomplete; (iii) the company was not entirely a digital start-up according to our definition. Thus, the final sample size was 285 firms, leading to a response rate of 57%. To avoid non-response bias, we compared key attributes, such as firm size, firm age, and firm ownership of the responding and non-responding firms. The t-tests for them were all insignificant, indicating a low possibility of non-response bias.

The survey measures were developed based on established scales. All items for continuous variables were measured on a 5-point Likert-type scale ranging from “1” (strongly disagree or very bad) to “5” (strongly agree or very good). Table 1 lists survey items for our key constructs along with the results of the validity assessment.

The measures for the three elements of business model innovation were adapted from Clauss (2017). We revised the scale to better fit China's digital entrepreneurship context based on the current state of China's digital economy and feedback from respondents during the preliminary research. Notably, the revised scale passed the reliability and validity tests.

Value proposition innovation: We asked each respondent to evaluate the extent to which their firm had engaged in distinct types of value proposition innovation activities. These activities included providing new offerings, targeting new consumers and markets, and building new consumer relationships. Exploratory factor analysis yielded a single factor with an eigenvalue of 3.390, with factor loadings greater than 0.65. The scale reliability of Cronbach's α was 0.84, and the composite reliability was also 0.84.

Value creation innovation: The value creation innovation activities were measured using a nine-item scale that included four major aspects—new capabilities, new technologies or equipment, new partnerships, and new processes or structures. The items of value creation innovation loaded on a single factor with an eigenvalue of 5.403, with factor loadings greater than 0.72. The scale reliability of Cronbach's α was 0.916, and the composite reliability was 0.92.

Value capture innovation: This variable was measured using three items, considering two important dimensions—a new revenue model and a new cost structure. The two items loaded on a single factor with an eigenvalue of 2.089. Factor loadings ranged from 0.82 to 0.86 with high scale reliability (Cronbach's α = 0.78), and the composite reliability was 0.77.

Digital start-up performance: Unlike established firms, start-ups place a greater emphasis on growth rather than profit (Zott & Amit, 2007). The incubator—as an entrepreneurship-supporting ecosystem—increases the survival rate and growth potential of start-ups by compensating for gaps in knowledge, competencies, and resources (Grimaldi & Grandi, 2005; Kakabadse et al., 2020). In the context of business incubators, prior research has identified sales growth, employee growth, and assets growth as key indicators for assessing start-up success (Baum & Bird, 2010; Marlow & McAdam, 2012; Sullivan, Marvel, & Wolfe, 2021). By further referring to entrepreneurship literature (Chandler, McKelvie, & Davidsson, 2009; Brinckmann, Salomo, & Gemuenden, 2011), we used three items—sales growth, employee growth, and asset growth—to measure digital start-up performance. Sales growth reflects the market potential of a start-up, a growth in the number of employees implies organizational development, and an increase in total assets denotes an increase in organizational capabilities. The loadings of all items were above 0.82. The Cronbach's alpha for the scale was 0.82, and the composite reliability was 0.79.

Seven variables were controlled in the study. First, given the critical role played by a founder in a venture's success, founder education was controlled and measured using five ordered educational degrees (high school and below, junior college, bachelor's, master's, and doctoral). Second, founder age was controlled and measured using six age intervals (20 years and below, 21–30, 31–40, 41–50, 51–60, and 61 years and above). At the firm level, we controlled for firm size and firm age. Firm size was measured by the logarithm of the number of employees. Firm age was measured by calculating the number of years since the firm was established. At the industrial level, uncertainty has become a key feature of the external environment and a key determinant of venture growth in the digital economy (Nambisan, 2017). Following existing literature (Child, 1972; Jaworski & Kohli, 1993; McCarthy, Lawrence, Wixted, & Gordon, 2010), we developed scales for three types of environmental uncertainties in the context of China's digital economy. Specifically, we used three items to measure demand uncertainty, two key items to reflect technological uncertainty, and three items to measure institutional uncertainty. Loadings of all items were significantly above 0.60. The Cronbach's alphas for the three scales varied between 0.65 and 0.93, and the composite reliabilities were between 0.72 and 0.85.

Following Podsakoff, MacKenzie, Lee, and Podsakoff (2003), procedural and statistical remedies were conducted to minimize the magnitude of common method variance (CMV). Furthermore, to reduce social desirability bias, we phrased all the questions with neutral words, informed all respondents of the project's academic purpose, and assured them that their answers were confidential. Subsequently, for data collection, we used two informants rather than one from each firm. One informant responded to questions about business model innovation, while another responded to questions concerning digital start-up performance.

As statistical remedies, we first employed Harman's single-factor test to address potential CMV. Consequently, both the screen plot and Kaiser Criterion yielded four distinct factors with eigenvalues greater than one. No single factor was dominant, with the first factor explaining 23.37% of the variance. Second, the latent variable control method was adopted for testing (Anderson & Williams, 1992), and two measurement models were created, with one model linking all items to a common method factor and another model loading items onto their theoretically assigned latent variables. The results demonstrated that the model's fitting indexes—after incorporating the deviation latent variables of the common method—were satisfactory (RMSEA = 0.059, RMR 0.120, CFI = 0.941, TLI 0.932, IFI = 0.942, PGFI = 0.703, PNFI = 0.770). However, the t-test between each fitting index and the original model (RMSEA = 0.068, RMR 0.043, CFI = 0.920, TLI 0.909, IFI = 0.921, PGFI = 0.692, PNFI = 0.757) revealed that there was no significant difference between the fitting degrees of the model with latent variable and the original model. This suggested that CMV was insignificant. Third, the marker variable method was adopted (Lindell & Whitney, 2001). We added a marker variable in the test whose correlation coefficient with other variables was significantly low. The results reveal that the χ2 of the marker variable model, at 468.106 (df=203), is significantly higher than that of the original model (χ2 = 448.132, df =204). Thus, CMV is unlikely to influence the results of this study.

Table 2 summarizes the descriptive statistics and correlations between all the constructs. Although the correlations are generally low, some constructs, such as value proposition innovation, value creation innovation, and value capture innovation, may be conceptually related and empirically correlated. To deal with this issue, we compared our measurement models with alternative measurement models (Tanriverdi & Venkatraman, 2005). As shown in Table 3, Models 1, 3, and 5 are baseline models that treat two constructs as two distinct factors, respectively. Conversely, Models 2, 4, and 6 are one-factor models with two constructs combined into one single factor. The results clearly demonstrate that the three baseline two-factor models have consistently better model fits, indicating that the three constructs are distinct from each other.

Two approaches were used to test for and avoid the problem of multicollinearity. First, all variables were mean-centered before regression (Aiken & West, 1991). Second, the variance inflation factor (VIF) was calculated. The test results presented in Table 3 reveal that the largest VIF is 2.525, which is significantly below the marginal value of 10 (Neter, Wasserman, & Kutner, 1990). Therefore, multicollinearity is not a significant issue in our analysis.

Table 4 presents the ordinary least squares (OLS) results. Models 1, 3, and 5 are baseline models with only control variables. Model 6 strongly supports H1—value proposition innovation (β = 0.214, p < 0.01) is positively and significantly related to digital start-up performance. H2 and H3 predicted that value creation and value capture innovation mediate the relationship between value proposition innovation and digital start-up performance, respectively. The three-step procedure of Baron and Kenny (1986) was adopted to test the two mediating effects. The first step of the process requires the independent variable to be significantly related to the dependent variable and the suggested mediating variable. This is satisfied by Model 6, which reveals a positive link between value proposition innovation and digital start-up performance. Second, as shown by Models 2 and 4, value proposition innovation is significantly and positively related to value creation innovation (β = 0.603, p < 0.001) and value capture innovation (β = 0.810, p < 0.001). The third step requires the mediating variable to be significantly related to the dependent variable when controlling the independent variable. As shown by Models 7 and 8, value creation innovation (β = 0.258, p < 0.01) and value capture innovation (β = 0.115, p < 0.1, p = 0.062) are both significantly linked to digital start-up performance. The effect of value proposition innovation on digital start-up performance is eliminated in Model 7 when value creation innovation is added (β reduces from 0.214 to 0.059, p = 0.523) and Model 8 when value capture innovation is added (β reduces from 0.214 to 0.121, p = 0.168). These results strongly support the existence of mediating effects. Similarly, when value proposition, value creation, and value capture innovation are all included in the regression (see Model 9), the results continue to support the mediating effects strongly.

Several statistical methods were used to test the robustness of our results. First, following Preacher and Hayes (2008), we used the bootstrapping approach to further test the mediating effects. The bootstrapped overall indirect effect of value creation innovation on digital start-up performance across 1,000 bootstrapped samples is 0.165 (the 95% confidence interval for the mediated effect ranges from 0.047 to 0.311 and does not straddle zero). This indicates that the mediating effect is significant (p < 0.05). Similarly, the overall indirect effect of value capture innovation on digital start-up performance across 1,000 bootstrapped samples is 0.137 (the 95% confidence interval for the mediated effect ranges from 0.031 to 0.275 and does not straddle zero). This indicates that the mediating effect is significant (p < 0.05). Considered together, both the aforementioned methods evidence the mediating roles played by value creation and value capture innovation.

Second, AMOS software was employed to perform maximum structural equation modeling (SEM) to test the mediating effects. Table 5 presents a sequence of nested structural models. Here, Model 1 presents the null model that constraints relationships among all latent variables to zero, and Model 2 adds value proposition innovation and firm performance relationship to Model 1. Further, Model 3 links value creation and value capture innovation to firm performance, and Model 4 adds value proposition innovation to value creation and value capture innovation to obtain the fully loaded mediation model. We find that each subsequent model has a better model fit than the earlier one. Model 4 shows a reliable and the strongest model fit (Chi-square = 393.246, degrees of freedom = 184, p < 0.001; RMSEA = 0.064, CFI = 0.930, TLI = 0.920, IFI = 0.931). Furthermore, the results indicate that value creation and value capture innovation fully mediate the relationship between value proposition innovation and digital start-up performance (β = 0.282, p = 0.297 (not significant)). Fig. 2 shows the results of SEM analysis.

Table 5.

Summary of fit indices for SEM analyses

Model	Model description	Chi-square	DF	RMSEA	CFI	TLI	IFI
Model 1	Null model	721.662	189	0.100	0.824	0.804	0.825
Model 2	Value proposition innovation → Digital start-up performance	709.403	188	0.099	0.828	0.807	0.829
Model 3		696.503	186	0.098	0.831	0.809	0.832
Model 4	The suggested model	393.246	183	0.064	0.930	0.920	0.931

Note: DF = degree of freedom; RMSEA = root mean square error of approximation; CFI = comparative fit index; TLI = Tucker-Lewis index; IFI = incremental fit index.

Fig. 2.

Results of SEM analyses. Note: †p < 0.1, *p < 0.05, ⁎⁎p < 0.01, ⁎⁎⁎p < 0.001; n.s. = not significant. Model fit: Chi-square = 393.246, DF = 183, p < 0.001; RMSEA = 0.064; CFI = 0.930; TLI = 0.920; IFI = 0.931.

(0.34MB).

Third, to confirm that the mediation model is the most suitable model, we tested two alternative moderation models. Theoretically, scholars might suspect that the three elements of business model innovation are complementary or even contradictory in contributing to firm performance. Thus, we conducted an additional group of regressions by treating value creation and value capture innovation as moderators rather than mediators. Model 4 in Table 6 reveals that the interaction between value proposition and value creation innovation is not significantly related to digital start-up performance (β = 0.024, p = 0.561). Similarly, Model 6 shows that the interaction between value proposition and value capture innovation is only marginally and negatively related to digital start-up performance (β = -0.100, p = 0.068). Overall, these results suggest that the moderation model does not fit well with our data.

Finally, we tested eight alternative mediation models to ensure that our mediation model shows the right paths. Table 7 shows that the suggested model (Model 1) of this study has the best model fit when compared to all other models. Models 2 and 3 do not fit well with our data as they suggest that value creation and value capture innovation are more likely to be independent in translating value proposition innovation into digital start-up performance. Models 4, 5, 7, and 9, indicating that the impact mechanism is most likely initiated by value proposition innovation, show consistently worse model fits when compared to Model 1. Although not ideal, Models 6 and 8 also fit our data to an acceptable extent. These results suggest that there may be two alternative models—one triggered by value creation innovation and the other by value capture innovation. These findings have been further discussed in subsequent sections.

Table 7.

Summary of fit indices for alternative mediation models

Model	Model description	Chi-square	RMSEA	CFI	TLI	IFI
Model 1	The suggested model	393.246	0.064	0.930	0.920	0.931
Model 2	Value proposition innovation → Value creation innovation → Value capture innovation → Digital start-up performance	529.806	0.081	0.886	0.872	0.887
Model 3	Value proposition innovation → Value capture innovation → Value creation innovation → Digital start-up performance	510.223	0.078	0.893	0.879	0.894
Model 4		528.329	0.081	0.886	0.871	0.887
Model 5	Value creation innovation → Value proposition innovation → Value capture innovation → Digital start-up performance	441.175	0.070	0.916	0.905	0.916
Model 6	Value creation innovation → Value capture innovation → Value proposition innovation → Digital start-up performance	510.994	0.078	0.893	0.879	0.893
Model 7		507.227	0.078	0.893	0.879	0.894
Model 8	Value capture innovation → Value proposition innovation → Value creation innovation → Digital start-up performance	444.120	0.070	0.915	0.904	0.915
Model 9	Value capture innovation → Value creation innovation → Value proposition innovation → Digital start-up performance	534.648	0.081	0.885	0.870	0.886

Note: RMSEA = root mean square error of approximation; CFI = comparative fit index; TLI = Tucker-Lewis index; IFI = incremental fit index.

This study makes two major contributions. First, as a response to Foss and Saebi (2017), this study opens the “black box” of business model innovation. The study examines the mechanism through which it contributes to business performance, thereby adding to the existing knowledge on business model innovation. The business model innovation architecture has been analyzed by disassembling it into three elements (value proposition, value creation, and value capture innovation) and exploring how the three elements are linked to digital start-up performance.

Our empirical findings demonstrate that value proposition innovation initiates firm performance, while value creation and value capture innovation serve as two key conduits in the mechanism. Further, our robustness checks show that the two conduits are most likely independent—value proposition innovation can be translated into digital start-up performance by conducting either value creation innovation or value capture innovation activities or both. Moreover, our findings suggest that value proposition innovation is more likely to trigger a start-up's value-creation activities in the digital environment, which aligns with the findings of Teece (2010, 2018) and Baden-Fuller and Haefliger (2013). Notably, our robustness tests indicate the possibilities of other impact mechanisms being initiated by value creation or value capture innovation of an organization. These findings demonstrate that value proposition innovation is more likely, but not always, the catalyst for start-up business model innovation.

Second, by addressing the relationship between business model innovation and performance from the demand side, this study injects fresh insights into cross-fertilization between business model and demand-side research. The demand-side perspective (Adner & Zemsky, 2006; Priem, 2007; Priem, Li, et al., 2012; Priem, Butler, et al., 2013) explains how a firm better creates value by understanding and meeting consumer demands (Ye et al., 2012; Schmidt et al., 2016; Manral & Harrigan, 2018; Priem et al., 2018; Rietveld, 2018). Similarly, business model scholars highlight the importance of value creation while emphasizing the role of consumers (Teece, 2010; Baden-Fuller & Haefliger, 2013; Priem et al., 2018). Scholars are now increasingly recognizing the importance of focusing on the demand side in business model research. While recent works have highlighted the role of freemium business models (Rietveld, 2018), heterogeneous consumer and demand complementarity in business model design (Guo et al., 2020), business model diversification (Sohl et al., 2020; Aversa et al., 2021), and adaption (Denoo et al., 2021), the role of consumers in business model innovation has been ignored.

This study provides a specific explanation of how consumers and demand factors contribute to the relationship between business model innovation and firm performance by using a demand-side perspective in the creation of theoretical models and hypotheses. Value proposition innovation refers to new offerings for consumers and emphasizes consumers’ role in creating products or services (Morris et al., 2005; Chesbrough, 2007; Johnson et al., 2008; Zott & Amit, 2010; Baden-Fuller & Haefliger, 2013; Chandler et al., 2014). Our results clearly suggest that the most important factor in the success of business model innovation in digital start-ups is value proposition innovation aimed at meeting consumer demand. It initiates the two impact paths and contributes to firm performance through value creation and value capture innovation. Additionally, this study shows that there are two paths to transform consumer value into firm performance. These two intermediary mechanisms indicate that demand and consumer factors may need to be combined with other factors (such as business processes) to contribute to firm performance.

This study has important practical implications. First, the study emphasizes the need for digital start-ups to be business model innovators. In the process of business model innovation, start-ups could disassemble the overall business model into three components and innovate by redesigning value proposition, value creation, and value capture activities. Further, start-ups must pay attention to the sequence of the three types of innovation activities in the digital economy. Our findings suggest that a digital start-up would be better off conducting value proposition innovation activities considering its existing value offerings. The new value proposition can then be realized through value creation or value capture innovation. On the one hand, digital start-ups can create and capture more value by developing new capabilities, utilizing new technologies, attracting new partners, or building new processes. On the other hand, they can also do so by designing novel revenue models or new cost structures.

Second, heterogeneous consumer demand is the source of a firm's competitive advantage. We suggest that digital start-up entrepreneurs prioritize understanding consumer needs and address them by designing novel value propositions. Consequently, a series of value creation and value capture activities should be designed and conducted to meet the needs of consumers and improve consumers’ willingness-to-pay, thereby ensuring successful translation of business model innovation into business success.

This study has certain limitations that should be addressed in future research. First, generalization of our findings to other economies must be done cautiously as China, being one of the world's largest and most active digital economies, provided an ideal context for this study. Moreover, the features of the digital economy vary across countries. China's digital economy is largely market-driven, owing to its demographic dividend. Conversely, the digital economy of the United States, led by Silicon Valley, is more technology-driven. Furthermore, considering “Industry 4.0” as a national strategy, Germany's digital economy is famous for the digitization of manufacturing industries. Thus, it would be interesting for future studies to investigate how start-ups create and capture value in various digital economies. Second, our research design is cross-sectional, limiting our ability to identify causal relationships. A longitudinal design would help uncover the dynamic process of how digital start-ups reinvent their business models over time. Third, our measures are based on entrepreneurs’ views, which may or may not correspond to reality. Future studies must employ objective measures to validate our propositions. Furthermore, future research must also focus on developing objective measures for both business model and business model innovation.