Digital transformation and cost stickiness: a perspective based on resource allocation and agency costs Open Access

The phenomenon of cost stickiness was discovered in the late 1990s and challenged the traditional cost accounting assumption of cost symmetry. For instance, Noreen (1991) posits that the relation between costs and cost drivers is far more intricate, raised by non-linear cost behavior issue (Noreen & Soderstrom, 1994, 1997). Some costs rise more with an increase in activity or cost driver, and they decrease with proportionate decreases in activity levels, a phenomenon termed “cost stickiness” (Noreen & Soderstrom, 1997). Anderson, Banker and Janakiraman (2003) find that the Selling, General and Administrative (SG&A) costs of a USA sample increase by 0.55% per 1% demand increase but decrease by only 0.35% per 1% demand decrease. Subramaniam and Weidenmier (2016) report that revenue changes greater than 10% trigger sticky behavior in SG&A expenses and costs of goods sold (COGS).

The 21st century has witnessed unprecedented advancements in digital technologies, fundamentally reshaping global economic landscapes and organizational paradigms in all sectors of the economy. China has become the largest consumer of the digital economy, and its added value reached 53.9 trillion, accounting for 42.8% of gross domestic product (GDP) in 2023. China's digital transformation (DT), evolving from foundational informatization to systemic ecosystem innovation, reflects a strategic integration of policy-driven initiatives and enterprise adaptation. Beginning in the early 2000s, the informatization phase (2002–2010) focused on deploying enterprise resource planning and computer-aided design systems under policies such as the 2002 Guidelines on Promoting Enterprise Management Informatization and the Revitalization Plan for the Software Industry, aiming to digitize workflows in traditional industries. By 2015, the strategic acceleration phase emerged with the Internet + Action Plan, formally embedding “digital transformation” into national policy to bridge Internet technologies with manufacturing and services. In 2016, the 13th Five-Year Plan for National Informatization clarified new orientations for DT and outlined goals and pathways to accelerate industrial digitalization and informatization development. Post 2018, the deep integration phase prioritized artificial intelligence (AI), Internet of Things and data governance, as shown in multiple government reports such as Guiding Opinions on the Development of Digital Economy.

With the rapid advancement of digital technology and AI, scholars have turned their attention to the impacts of digital technology on the cost stickiness. Digital tools enable companies to gather, process and analyze data at a scale and speed never before possible, significantly impacting how businesses operate (Kant, 2023). Digital governance structures bring about more refined and scientific corporate governance (Rosenblat & Stark, 2016), reduce the hierarchical constraints of corporate management structures; prompt more networked and flattened enterprise organizational structures; facilitate precise analysis, scientific decision-making and optimization of enterprise management behavior and reduce internal organizational costs (Erel, Stern, Tan, Weisbach, & Goldstein, 2021), thereby alleviating agency conflicts.

Amid a new wave of economic downturns, enterprises striving for sustained and healthy development should deeply integrate digital technology into production development and effectively reduce costs to enhance core competitiveness. Therefore, investigating the role of DT in reducing cost stickiness not only provides empirical evidence for optimizing management behavior but also facilitates the integration of enterprise information across platforms efficiently. Using data of Chinese listed companies from 2009 to 2021, this paper examines the impact of DT on cost stickiness from the perspective of resource allocation and agency costs. This paper demonstrates that DT significantly alleviates cost stickiness, primarily by improving resource allocation efficiency and reducing agency problems.

This paper contributes to the literature in the following ways. First, previous studies categorized the DT keywords into two groups (Chen & Xu, 2023): underlying digital technology and application of digital technology. This paper further divides the keywords into three groups: underlying digital technology, digital infrastructure construction and application of digital technology, to provide a more comprehensive measurement of enterprise DT. Furthermore, the differences in the impact of the three secondary statistical calibers on cost stickiness were investigated in our robustness checks. Second, the mechanisms through which DT affects cost stickiness can be broadly categorized into environment uncertainty (Chen & Xu, 2023), corporate governance (Gao & Ren, 2024), management efficiency (Zhao & Huang, 2022; Chen & Xu, 2023) and corporate operational efficiency (Wu & Tian, 2022). Our research examines how DT enhances financial information sharing and resource allocation efficiency from the efficiency of financial DT and production DT: enhancing corporate financial information sharing and improving resource allocation efficiency. The pathways of “digital transformation → accounting information → cost stickiness” and “digital transformation → resource allocation efficiency → cost stickiness” are identified as more direct and effective.

The remainder of this paper is organized as follows: Section 2 reviews the relevant literature and develops the hypotheses; Section 3 details the study's data and research methodology; Section 4 presents the empirical results and analysis and Section 5 discusses the results and concludes.

Digitization means the adoption of digital technologies in daily activities from personal to organizational contexts (Gregori & Holzmann, 2020). Digitization allows organizations to create greater customer value through novel resource combinations (Amit & Han, 2017; Tortora, Chierici, & Farina Briamonte, 2021), fortify traditional manufacturing industries (Borangiu, Trentesaux, Thomas, Leitão, & Barata, 2019), enhance product and process innovation and create value in new ways to effect fundamental organizational change (Hanelt, Bohnsack, & Marz, 2020; Vial, 2019). Through DT, enterprises are able to use the internet, big data and other digital technologies to gain access to a larger range of counter parties and vast amounts of information about traded products, which facilitates firms' understanding of products, reduces their search costs (Dana & Orlov, 2014; Kuhn & Mansour, 2014), transforming relationships between enterprises and other market participant, fostering greater openness, connectivity and interaction among organizations (Adner, Puranam, & Zhu, 2019). DT corresponds to new values, practices and structures that influence existing arrangements and rules (Holzmann & Gregori, 2023).

Prior research indicates that DT enhances enterprise value creation and stimulates the entrepreneurial spirit (Galindo-Martín, Castaño-Martínez, & Méndez-Picazo, 2019). The application of digital technologies can assist enterprises in breaking away from traditional management models (Song, Escobar, Arzubiaga, & De Massis, 2021), optimizing organizational structures and production management processes and reducing costs. The application of data mining, machine learning and AI technologies enables intelligent decision-making within companies (Loebbecke & Picot, 2015), reduces managerial discretion, optimizes and monitors the selection of board members and decreases the costs of management decision failures (Erel et al., 2021). The use of digital technology as an organizational management tool, replacing “managers” in the enterprise (Kellogg, Valentine, & Christin, 2020), has reduced internal management costs.

The occurrence of various costs reflects the managers' allocation of different resources. Increasing or decreasing such resources within a company will result in corresponding adjustment costs. Due to higher adjustment costs associated with reducing resource inputs than with increasing them, changes in enterprise costs reflect not only fluctuations in current sales but also are influenced by current production capacity and expected sales (Cooper & Kaplan, 1991). Intelligent production lines in manufacturing enable enterprises to switch between different varieties, types and batches, thereby overcoming the limitations of previous single production lines that could produce only specific varieties and batches. The empirical results demonstrate that DT can significantly reduce cost stickiness by reducing adjustment costs and management's optimistic expectation (Chen & Xu, 2023). Thus, the application of intelligent production lines can reduce asset specificity and production adjustment costs, thereby reducing cost stickiness. Based on the cost adjustment perspective, our first hypothesis is as follows:

According to the adjustment-costs view, the higher cost of adjusting enterprise resources is an important cause of cost stickiness. If a company's resources are highly versatile and liquid (such as general-purpose equipment or short-term leased assets), the adjustment costs are low. When demand declines, the company can quickly and inexpensively dispose of or convert these resources, thereby exhibiting lower cost stickiness. Conversely, suppose the company has invested heavily in specialized assets (such as dedicated production lines, custom molds or highly specialized employees). In that case, these resources are difficult to reallocate to other uses or to sell on the market. Thus, the adjustment costs required to reduce expenses (such as contract termination penalties or significant asset write-downs) may exceed the cost of maintaining idle resources. As a result, management would rather bear the cost of idleness than undertake aggressive adjustments, leading to high-cost stickiness.

For instance, Anderson et al. (2003) find that cost stickiness is more common when physical assets and human capital comprise a larger share of a company's resources. Banker and Chen (2006) find that countries with higher labor market adjustment costs tend to exhibit greater cost stickiness. Furthermore, when companies bear more social costs (such as employee benefits, social donations, etc.), they experience higher levels of cost stickiness. The effectiveness of DT in reducing cost stickiness depends on the efficiency of resource allocation. Enterprises with high resource allocation efficiency possess mature data analysis systems that enable them to rapidly integrate massive data into specific resource deployment directives. They can precisely channel digital investments into core bottleneck areas and fields yielding the highest returns, avoiding ineffective “scattergun” investments. When a certain business lines contract, the freed-up resources (personnel, equipment and capital) can be quickly identified and redeployed to promising new businesses or innovation projects. High resource allocation efficiency also reduces management's reluctance to maintain redundant resources due to high “adjustment costs” (such as layoffs and rehiring), thereby lowering cost stickiness.

In contrast, companies with low resource allocation efficiency often suffer from rigid departmental barriers, lengthy approval processes and centralized decision-making authority. Even when digital systems issue early warnings, data insights remain trapped in reports, unable to guide timely action. Resources released through digitalization remain confined to their original departments and are unable to flow effectively across the organization. Employee skills cannot be transformed, idle equipment cannot be shared and cost stickiness persists. In this scenario, both the high adjustment costs (e.g. layoffs) and holding costs (maintaining idle resources) contribute to sustained cost stickiness. Inefficient scenarios include blindly following digital trends, dispersing resources across multiple uncoordinated projects and even creating numerous “data silos” and “zombie systems.” Thus, the company not only fails to reduce cost stickiness in existing operations but also creates new, higher-level “digital cost stickiness” due to substantial additional depreciation, amortization and maintenance expenses.

Digital governance structures can incorporate formal overlay controls (e.g. role management and access permissions), thereby increasing certainty and reducing tolerance for erroneous transactions (Hanisch, Goldsby, Fabian, & Oehmichen, 2023). Firstly, DT accelerates the organization's response to market changes, improves decision-making efficiency and allocates resources accordingly. When demand declines, DT helps managers better understand resource utilization and make timely reductions in resource inputs. Similarly, when demand increases, it enables managers to accurately predict the resources needed to expand capacity and promptly increase resource inputs to meet the growing market demand.

Secondly, the use of digital technology transcends enterprise boundaries, enhancing information exchange between upstream and downstream enterprises (Wang & Li, 2018). This information sharing improves communication efficiency between the various links of the industrial chain, enabling enterprises to better coordinate and adjust critical business activities such as supply chains and production processes. Therefore, when sales decline, DT can help respond quickly to market changes, reduce resource reallocation costs along the industrial chain and consequently diminish cost stickiness.

Thirdly, DT enhances managers' willingness to dispose of unused resources. In traditional business models, firms retain unused resources in order to avoid adjustment costs when sales decline. However, DT helps firms transfer their unused resources through sharing or selling technologies and equipment and acquire them at a lower cost. Therefore, firms are more inclined to dispose of idle resources, reduce resource adjustment costs and improve asset utilization efficiency. Thus, our second hypothesis is as follows:

The agency problem causes cost stickiness, as managers want to retain obsolete resources to “avoid personal consequences of retrenchment” (Anderson et al., 2003) or to build an empire and enjoy a “quiet life” (Chen, Lu, & Sougiannis, 2012). Dierynck, Landsman, and Renders (2012) find that managing earnings through discretionary accruals leads to higher cost stickiness. Kama and Weiss (2012) argue that when management decisions are driven by incentives to avoid losses or achieve profit targets, managers adjust resources more rapidly when activity levels decline than when they rise.

DT reduces cost stickiness by enhancing information transparency and supervision mechanisms (such as traceability in data sharing), thereby restricting management's opportunistic behavior and enabling faster resource adjustment. When principal-agent conflicts are not severe, DT is used to optimize processes and improve efficiency. During business downturns, management actively utilizes digital tools (such as data analysis) to identify and eliminate redundant resources, thereby reducing cost stickiness. However, when severe principal-agent conflicts exist, management may perceive DT as an ideal excuse to expand investments and increase controllable resources. They may extensively recruit digital talent and purchase expensive information technology (IT) equipment, even when such investments are economically unjustified. During business downturns, management becomes highly reluctant to cut resources from the “digital empire” they have built, as doing so would undermine their personal authority and prestige. Consequently, they strive to retain redundant digital resources and personnel, resulting in an increase, rather than a decrease, in cost stickiness after DT.

As cost stickiness results from deliberate interventions by managers in resource adjustments, effective corporate governance systems can reduce information asymmetry, bringing the level of cost stickiness closer to the optimal cost response level, thereby effectively mitigating cost stickiness (Chen et al., 2012; Ibrahim & Ezat, 2017). An opposing viewpoint suggests that the presence of agency problems may reduce cost stickiness (Dierynck et al., 2012). To alleviate the pressure of performance evaluation, management may minimize the projects that are long-term beneficial to the company but may impair performance assessment, thereby exerting a restraining effect on cost stickiness (Kama & Weiss, 2012).

Firstly, with the deep involvement of digital technology in decision-making, the traditional manager-centered decision-making model is shifting towards a data-analysis-based model in which digital technology acts as a co-participant. The emergence of this new decision-making paradigm enables managers to make decisions more efficiently (Rahwan, Cebrian, & Obradovich, 2019) and promotes more efficient information communication. DT enhances internal information communication efficiency within enterprises by improving information transparency, reducing supervision costs and broadening participation channels, thereby significantly increasing the effectiveness of major shareholders' oversight of management and helping curb short-termism in managerial decision-making. Zhong (2018) argues that DT, with the open and borderless nature of digital contexts, holds significant potential value in enhancing information transparency, thereby promoting the circulation, transmission and sharing of internal information within enterprises (Shen & Yuan, 2020). After enterprises undergo DT, robust computing capabilities and enhanced information transparency enable accurate assessment of corporate profitability and risks. Shareholders gain access to more effective information on management decision-making, thereby strengthening the internal oversight of management. Such enhanced internal oversight serves as an ex ante constraint and deterrent against agency problems such as cost stickiness, ultimately enhancing the level of internal corporate governance (Qi, Cao, & Liu, 2020).

Secondly, in the backdrop of DT, enterprise information becomes increasingly diverse, including various non-structured data (Warren, Moffitt, & Byrnes, 2015). Through the integrated application of big data and other digital technologies, managers can leverage digital mining, analysis systems and other decision-support systems to enhance decision-making efficiency and quality, thereby reducing irrational decision-making behaviors.

Finally, DT can increase the oversight of external stakeholders and reduce agency costs. Enterprises can utilize networks and digital platforms to proactively and rapidly disseminate information to the market. At this point, external investors can access more comprehensive information, which significantly reduces information asymmetry. Moreover, the application of digital technology facilitates easier monitoring of corporate operations by external stakeholders, expanding oversight channels (Qi et al., 2020).

In summary, DT can effectively supervise the behavior of agents, exerting governance effects and thereby reducing cost stickiness. Thus, our second hypothesis is as follows:

Using Chinese A-share listed companies from 2009 to 2021, this paper investigates the impact of DT on the' stickiness of enterprises. Table 1 presents the sample screening process. We begin with a firm-year sample from the China Stock Market and Accounting Research database. After excluding firm-year observations in the financial and insurance sector, specially treated and ^*ST firms, firms without continuous sales and cost of sales for two successive years, firms with sales and operating cost changes around the 0.5% deciles and firms with missing financial statement and DT data, our final sample consists of 25,331 firm-year observations. All continuous variables are winsorized at the 1% and 99% levels.

Cost of sales change (lnCOGS). Following Anderson et al. (2003) and Dierynck et al. (2012), the lnCOGS is measured as the natural logarithm of the ratio of the cost of sales in year t to the cost of sales in year t−1.

The independent variables include three aspects: sales change (lnREV), sales decrease (Dec) and DT.

Sales change (lnREV) is measured as the natural logarithm of the ratio of sales revenue in year t to sales revenue in year t−1.

Sales decrease (Dec) is an indicator variable that equals 1 if sales revenue in year t is less than sales revenue in year t−1 and 0 otherwise.

DT is measured as the natural logarithm of the firm's total DT words in the annual report of year t plus 1. Following Chen and Srinivasan (2023), Goldfarb, Taska, and Teodoridis (2023) and Wu, Hu, Lin, and Ren (2021) and government policy documents related to the digital economy such as the “China Digital Economy Development Report” (2022) and the “2021 China Enterprise Digital Transformation Index Study”, we manually recognize and conduct the DT dictionary through the following three aspects: digital foundational technologies, digital infrastructure and digital technology applications (see Figure 1). Then, using Python, we match the words in Figure 1 with annual reports.

Driven by a combination of technological advancements, market demands, competitive pressures and regulatory requirements, the emergence of DT is more likely to occur in specific industries. Figure 2 reports the industry distribution of DT. The IT industry leads in DT, followed by the leasing and business services, communication and cultural industries, health and social work, professional, scientific and research services, papermaking and printing and education.

Following prior research, we control for several economic determinants of cost stickiness and related decisions. Economic determinants (Econ_var) include employee intensity (EInt), asset intensity (AInt), successive decrease (Sdec) and economic growth (GDPgrow).

EInt is measured as the number of employees divided by total sales revenue (Chen et al., 2012).

AInt is measured as the total assets divided by sales revenue (Anderson et al., 2003).

Successive decrease (Sdec) is an indicator variable that equals 1 if sales revenue in year t−1 is less than sales revenue in year t−2 and 0 otherwise (Chen et al., 2012). Economic growth (GDPgrow) is measured as the province-level GDP growth rate in year t (Anderson et al., 2003; Hartlieb, Loy, & Eierle, 2020).

Independ is measured as the proportion of the number of independent directors to that of the total directors. Top1 is measured as the proportion of shares held by the largest shareholder of the total shares. Mshare is measured as the proportion of management shareholdings to the total shares. Chief executive officer (CEO) Dual (Dual) is an indicator variable that equals 1 if the CEO is also the chairman and 0 otherwise. Age is measured as the natural logarithm of the number of years that the company has been listed. Soe is an indicator variable equal to 1 if the firm is a state-owned enterprise and 0 otherwise. Lev is measured as the total liabilities divided by the total assets. Size is measured as the natural logarithm of the total assets. All variables are defined in Appendix Table 1.

Following Anderson et al. (2003), we introduce the following logistic regression model to test the existence of cost stickiness:

Where i and t denote firm i and year t. $lnCOGSi,t$ is measured as the natural logarithm of the ratio of the cost of sales in year t to the cost of sales in year t−1; $lnREVi,t$ is measured as he natural logarithm of the ratio of sales revenue in year t to sales revenue in year t−1; $Deci,t$ is An indicator variable that equals 1 if sales revenue in year t is less than sales revenue in year t−1 and 0 otherwise. $β1$ represents the percentage increase in cost of sales when sales increase by 1%; $β1$ +$β2$ represents the percentage decrease in cost of sales when sales decrease by 1%. Therefore, when $β2$ <0, it indicates the presence of cost stickiness in the firm.

Following Anderson et al. (2003), Chen et al. (2012), Venieris, Naoum, and Vlismas (2015) and Golden, Mashruwala, and Pevzner (2020), we introduce the following logistic regression model to examine the impact of DT on cost stickiness and test Hypothesis 1:

Where i and t denote firm i and year t. $DTi,t$ represents the DT variable for firm i in year t; $Econ_var$ denotes the economic control variables$.β3$ measures the impact of DT on cost stickiness, where $β3$ >0 indicates that the DT reduces cost stickiness, while $β3$ <0 suggests that the DT exacerbates cost stickiness. The model includes industry and year fixed effects (FE) to control for unobserved heterogeneity across industries and over time.

We firstly introduce the following ordinary least squares model to examine the impact of corporate DT on resource allocation efficiency and agency costs:

Where i and t denote firm i and year t, Y represents resource allocation efficiency (measured as the total factor productivity (TFP)) and agency costs (measured as total asset turnover (ATO) and information transparency (AQ)). ATO is calculated as sales revenue divided by total assets. AQ is measured as the sum of the absolute values of manipulative accrual items over the past three years multiplied by −1.

Secondly, we introduce Y into Model (2) to examine the effects of resource allocation efficiency and agency costs on cost stickiness, as shown in Model 4.

Table 2 presents the descriptive statistics of the main variables. The mean of the lnCOGS is 0.113 (median = 0.102), while the mean of the sales change (lnREV) is 0.106 (median = 0.098), which is consistent with the principle of matching costs and revenues. The mean of the sales decrease (Dec) is 0.289 (median = 0), suggesting that 28.9% of the total samples exhibit a decrease in sales. The mean of the DT is 1.534 (median = 1.386), with a standard deviation of 1.415, indicating significant variation in the degree of DT across companies. The mean of successive decrease (Sdec) is 0.114 (median = 0), indicating that 11.4% of the companies experienced a successive two-year sales decrease. The mean and median of EInt and AInt are 1.318 (median = 1.039) and 2.558 (median = 1.919), respectively.

Table 3 reports the results of the influence of DT on cost stickiness. Columns (1) present the results of Model 1 to test the existence of cost stickiness. Columns (2)–(4) present the results of Model 2 to test the influence of DT on cost stickiness. The results of Column (1) show that the coefficients of lnREV and Dec×lnREV are 0.979 (t = 163.875) and −0.055 (t = −4.417), respectively, both significant at 1% level, indicating that when sales increase by 1%, the cost of sales increases by 0.979%; however, when sales decrease by 1%, the cost of sales only decreases by 0.924%. The results prove the existence of cost stickiness in Chinese listed companies. The results of Column (2) show that the coefficients of Dec×lnREV and Dec×lnREV×DT are −0.079 (t = −4.984) and 0.016 (t = 2.621), both significant at the 1% level. Columns (3)–(4) introduce economic and firm-level control variables, and the results are consistent with those of Column (2). The results demonstrate that corporate DT can effectively reduce cost stickiness, supporting Hypothesis 1.

From the perspective of the resource allocation mechanism, corporate DT can reduce cost stickiness by improving resource allocation efficiency. To verify this mechanism, we use TFP to measure the efficiency of resource allocation. Following (Schoar, 2002; Levinsohn & Petrin, 2003; Giannetti, Liao, & Yu, 2015), we construct a Cobb–Douglas production function to obtain a proxy for TFP.

Where i and t denote firm i and year t; R denotes the natural logarithm of the sales revenue; $lnK$ denotes the natural logarithm of the net value of fixed assets; $lnL$ denotes the natural logarithm of the number of employees and $lnM$ denotes the intermediate input of the company, measured as the natural logarithm of cash paid for commodities or labor. TFP is the residual of this regression.

We divide the sample into two subsamples based on the median of corporate asset allocation efficiency (TFP) and repeat Model 2. As reported in Table 4, the coefficient on DEC×lnREV×DT is 0.016 (t = 2.92) for the high-TFP subsample and 0.026 (t = 4.54) for the low-TFP subsample, both significant at the 1% level. The intergroup test results indicate a statistically significant difference. The results indicate that enterprise DT significantly improves resource allocation efficiency and alleviates cost stickiness, thereby supporting hypothesis H2.

Leftwich (1983), Asquith, Beatty, and Weber (2005), Watts (2003) and Armstrong, Guay, and Weber (2010) suggest that information transparency reduces information asymmetry and decreases agency problems. Furthermore, Chinese firms are characterized by agency conflicts between controlling shareholders and outside minority shareholders (Fan & Wong, 2002; Peng, Wei, & Yang, 2011). We divide the sample into two subsamples based on the median of the separation of voting rights and cash flow rights (SEP) and then repeat Model 2. Thus, we examine whether DT reduces corporate cost stickiness by alleviating agency problems. We use the separation of voting rights and cash flow rights (SEP), total asset turnover (ATO) and information transparency (AQ) as proxies for agency costs. ATO is calculated as sales revenue divided by total assets. AQ is measured as the sum of the absolute values of manipulative accrual items over the past three years multiplied by −1.

We divide the sample into two subsamples based on the median of the SEP, ATO and AQ in the three years prior to DT implementation. Table 5 reports the results. Columns (1) and (2) show that the coefficients of DEC×lnREV×DT in subsamples of SEP are 0.014 (t = 2.30) and 0.023 (t = 4.66), respectively, which are significant at 1% level and 5% level, and the intergroup test results indicate a statistically significant difference. Columns (3) and (4) show that the coefficients of DEC×lnREV×DT in subsamples of ATO are 0.017 (t = 3.76) and 0.019 (t = 3.27), respectively, both significant at 1% level, and the intergroup test results indicate a statistically significant difference. Columns (5) and (6) show that the coefficients of DEC×lnREV×DT in subsamples of AQ are 0.020 (t = 2.49) and 0.025 (t = 3.53), respectively, which are significant at the 1% and 5% levels, respectively. The intergroup test results indicate a statistically significant difference.

The results indicate that DT significantly improves asset utilization efficiency and firms' information transparency, thereby reducing cost stickiness. The results support hypothesis H3.

The degree of DT may affect cost stickiness differently at different stages of a company's life cycle. This study employs the cash flow method (Dickinson, 2011) to divide the sample into three life stages: growth, maturity and decline, to examine the impact of DT on cost stickiness. The regression results in Table 6 reveal that in the maturity subsample, the coefficient for Dec×lnREV×DT is 0.041 (t = 3.520), significant at the 1% level. In contrast, in the growth and decline subsamples, the coefficients are not significant. The results indicate that DT reduces cost stickiness more significantly for companies in the maturity stage compared with those in the growth and decline stages. The impact of DT on reducing cost stickiness is more pronounced for mature-stage companies because they typically have well-established organizational structures, strong profitability and stable cash flows that help them withstand the risks associated with transformation failures.

Managers are generally believed to possess their own styles when making investment, financing and other strategic decisions, thereby imprinting their personal mark on the corporate practices (Bertrand & Schoar, 2003). Imprinting theory suggests that individuals form imprints that match environmental characteristics during their developmental stages, which continuously affect their behavioral decisions (Marquis & Tilcsik, 2013). A CEO's early career experiences, values, ideologies and personality can significantly influence corporate decision-making (Brockman, Campbell, Lee, & Salas, 2019; Custódio & Metzger, 2014).

Therefore, the IT background imprint formed through long-term learning and work experience enables CEOs to leverage their extensive IT research and development (R&D) expertise and keen perception of innovation opportunities, allowing them to more effectively evaluate and manage the implementation of DT projects, reduce uncertainties during the corporate transformation process and advance DT practices. Following (Lim, Stratopoulos, & Wirjanto, 2014; Haislip & Richardson, 2018), we define ITCEO as a CEO with a background in computer science, communications, electronic information or software or with relevant work experience in IT research and development, design or production. We divide the full sample into two subsamples based on CEOs' IT backgrounds: ITCEO and non-ITCEO subsamples and repeat the regression in Model 2. Table 7 reports the results. Columns (1) and (2) show that the coefficients of Dec×lnREV×DT in the subsamples of CEOtec and non-CEOtec are 0.035 (t = 6.11) and 0.010 (t = 1.91), respectively, which are significant at the 1% and the 10% level, and the intergroup test results indicate a statistically significant difference. This empirical evidence suggests that an IT background helps CEOs promote the inhibitory effect of corporate DT on cost stickiness.

In a fiercely competitive market, companies face risks such as constrained operating space and declining profit margins. Managers will seek transformation opportunities to address current operational deficiencies. Therefore, market competition can regulate resource allocation through the “invisible hand” and alleviate the agency problems in corporate decision-making regarding DT (Zhang, Liao, Du, & Wang, 2025).

We introduce the Herfindahl–Hirschman index (HHI) to measure the market competition level. We divide the sample into two subsamples based on the HHI median: High-HHI and Low-HHI and repeat the Model 2 regression. Table 7 reports the results. Columns (3) and (4) show that the coefficients of DEC×lnREV×DT in the subsamples of High-HHI and Low-HHI are 0.024 (t = 4.57) and 0.017 (t = 2.95), respectively, which are both significant at 1% level, and the intergroup test results indicate a statistically significant difference. The results suggest that the inhibitory effect of enterprise digitization on cost stickiness is more significant in High-HHI enterprises.

The imbalance in development across China leads to regional disparities in enterprise development. Economic development, resources and market advantages in developed regions contribute to the smooth implementation of DT. We divide the whole sample into three subsamples according to the companies' locations: Eastern, Central and Western regions. The results in Table 8 indicate that only the coefficient of Dec×lnREV×DT in the Eastern region is significant at the 5% level. The results suggest that the inhibitory effect of DT on cost stickiness is more pronounced in the developed regions.

Considering the possible endogenous relationship between DT and the disturbance term due to the omitted variables, we select the natural logarithm of local general public fiscal expenditure on science and technology (CSE) from the “China Urban Statistical Yearbook” as an instrumental variable for two-stage least squares (2SLS) regression. Local science and technology expenditure guarantees local enterprise innovation and drives enterprise transformation and upgrading. Additionally, there is currently no evidence that local public expenditure on science and technology affects enterprise cost stickiness. Table 9 reports the results of two-stage least squares (2SLS) regression. Column (1) shows that the coefficient of CSE is significantly positive at the 1% level. Column (2) shows that the coefficient of Dec×lnREV×DT is significant at the 5% level. The 2SLS regression and analysis show that our earlier results for H1 are robust when controlling for the potential endogeneity between DT and cost stickiness.

Digitization transformation is not a mandatory disclosure item for listed companies, which may lead to sample selection bias. We employ the Heckman two-stage model to solve this problem: In the first stage, a Probit regression is utilized to obtain the Inverse Mills Ratio (IMR), where the dependent variable is DT_dum, is an indicator variable that equals 1 if the enterprise undergoes DT and 0 otherwise; the independent variable is the proportion of enterprises undergoing DT within the industry (DT_Ratio). In the second stage, we introduce the IMR and repeat the regression. The results of Table 10 show that the coefficient of Dec×lnREV×DT remains significantly positive at the 1% level, indicating that H1 is robust when controlling for the potential endogeneity between DT and cost stickiness.

We take the digital underlying technology (Tec), digital infrastructure construction (Infra) and digital technology application (App) as alternative proxies for DT, which are the natural logarithm of one plus the frequencies of Tec, Infra and App, respectively. The results of Columns (1)–(3) in Table 11 show that Dec×lnREV×Tech, Dec×lnREV×Infra and Dec×lnREV×App are 0.011 (t = 1.645), 0.017 (t = 1.971) and 0.021 (t = 2.615), significant at the 10%, 5% and 1% level, respectively. The results are consistent with the previous findings.