Old Ways or New Bots: Conceptualizing Status Quo Bias and Corresponding Countermeasures for Robotic Process Automation

SQB describes a biased preference for a current solution or way of doing things (Samuelson and Zeckhauser 1988). The literature distinguishes among three explanations for this preference: 1) Cognitive misperception or loss aversion refers to the "nonrational" preference of individuals to remain with the status quo because potential losses in the context of change are perceived as unrealistically large (Kahneman and Tversky 1979). 2) Rational decision-making captures the perception of "rational" aspects such as net benefits and costs (uncertainty and transition costs) (Kim and Kankanhalli 2009). 3) Psychological commitment comprises other "nonrational" influences, such as sunk costs, control, and social norms. In their initial publication, Samuelson and Zeckhauser (1988) did not explicitly delineate these rational and nonrational explanation approaches. We thus need to consider them in combination.

In the IS domain, SQB has received particular interest in the context of system adoption and acceptance. For example, Kim and Kankanhalli (2009) developed an integrative framework with distinct concepts combining SQB theory, the technology acceptance literature, and the equity implementation model. The constructs found in that study are shown and extended in Table 1. Kim and Kankanhalli (2009) successfully tested parts of this model in the context of an ERP introduction (Kim and Kankanhalli 2009). These authors did not rely on classical theories such as the Technology Acceptance Model or the Unified Theory of Acceptance and Use of Technology, but rather on a selection of related constructs.

To ground our research firmly in both the bias and technology acceptance literature, we rely on the two constructs that are most prominent across the multitude of available models:

We focus on these constructs in our experiment, as researchers have determined that these are the most influential with respect to both the intention to use and the actual use of technologies (Ma and Liu 2004). Furthermore, the relevance of these constructs has been proven in the RPA context (Wewerka et al. 2020).

Table 1 provides an overview of the constructs.

The research has discussed several countermeasures for SQB. Godefroid et al. (2022) reviewed the SQB literature and provided an overview of the described countermeasures. Two countermeasures stand out because of their suitability in the RPA context and their empirical foundations (albeit outside the domain of technology acceptance and use):

In the following, we introduce the research model and the hypotheses underlying the experiment on the basis of the constructs described above. For an overview of the constructs, see Table 1 in Sect. 2.1. The treatments build upon the selected countermeasures (see Sect. 2.2) and are described in more detail in Appendix A (available online via http://link.springer.com). Figure 1 provides an overview over the research model.

The literature review conducted by Lee and Joshi (2017) demonstrates that studies have typically selected only those SQB constructs relevant to their contexts. Thus, not all SQB constructs might be relevant in the RPA case. In the literature, one of the main advantages of RPA is that it is inexpensive and does not require substantial changes to the processes for automation (Syed et al. 2020a, b). Therefore, transition costs should become irrelevant in the context of a technology that requires no large investments and is easy to implement (van der Aalst et al. 2018; Lacity and Willcocks 2016). The RPA introduction effort must be negligible compared with that necessary in the context of an ERP introduction, as described by Kim and Kankanhalli (2009). We thus propose that (H2.3) TC has no influence on PE and (H2.4) TC has no influence on EE.

Continuing this argument, we must also assume that uncertainty costs are irrelevant in the RPA context. When the bot simply replicates the work of people without changing the process or the way they interact with the core system (Syed et al. 2020a, b), uncertainty over the outcome should be minimal. Additionally, in contrast to ERP systems, for which introducing only half a system for test purposes is impractical, RPA can be developed and tested easily for single cases or partial processes first (e.g., Lacity et al. 2015). We thus propose that (H2.5) UC has no influence on PE and (H2.6) UC has no influence on EE.

Zhang et al. (2017) report a positive influence of perceived benefits on behavioral intention. Kim (2011) reports a negative influence of perceived value on user resistance, which we expect to be inverted for the direct determinants of behavioural intention. We therefore propose that (H2.7) NB has a positive influence on PE and (H2.8) NB has a positive influence on EE.

Similar to transition and uncertainty costs, sunk cost also appears not to be relevant in the RPA context. RPA automates processes that are typically conducted digitally but manually (van der Aalst et al. 2018; Plattfaut and Borghoff 2022). Thus, no existing system must be replaced, and no extensive training efforts have become obsolete (Kim and Kankanhalli 2009). Therefore, we propose that (H2.9) SC has no influence on PE and (H2.10) SC has no influence on EE.

Hsieh (2015) reports a positive influence of social norms on the behavioural intention to use health clouds. Similarly, Polites and Karahanna (2012) report a positive influence of social norms on behavioral intentions. Positive affirmation by people important to the individual should increase the intention to use a system. Venkatesh et al. (2003) conceptualized performance and effort expectancy as direct determinants of behavioral intention; therefore, we also expect a positive influence of social norms. We thus propose that (H2.11) SN has a positive influence on PE and (H2.12) SN has a positive influence on EE.

Hsieh (2015) also reports a positive influence of control on the behavioral intention to use health clouds. He refers to the effect by which, when individuals have more resources and knowledge, they also have a greater intention to use a new system. Hence, we propose that (H2.13) CO has a positive influence on PE and (H2.14) CO has a positive influence on EE.

For our experiment, we considered only those constructs for which we hypothesized a positive or negative influence. We thus excluded those constructs we expected would not have an influence (H2.3–H2.6, H2.9, and H2.10), as we deemed them irrelevant to the exploration of SQB in the RPA context in our experiment and excluding them eased the experimental procedure and data collection. However, we consider all constructs in our subsequent qualitative research phase.

We employ a (previously unplanned) mixed-methods approach, as RQ3 emerged from the results of the first study. To answer RQ1 and RQ2, we performed an online experiment to determine the effects of two promising countermeasures to SQB on RPA acceptance. We learned that SQB functions differently in the RPA context (see Sect. 4.1), which holds interesting implications for theory and practice. The reasons for this difference could not be inferred from the experiment. Therefore, we decided to follow up the experiment by conducting qualitative research to uncover the mechanisms at work via a sequential explanatory approach (Venkatesh et al. 2016). This approach is also inspired by the ideas of Pratt et al. (2022), who argue that researchers need to combine different analytical moves to solve a research problem and highlight the importance of emergence in research rather than preplanning all of the steps, and by Trieu et al. (2022), who use qualitative data to understand the unexpected results from their prior quantitative study. In our qualitative research, we focused on RQ3 and relied on in-depth case data to expand and complete the insights obtained from the online experiment (Venkatesh et al. 2013, 2016; Petter and Gallivan 2004). According to Venkatesh et al. (2013), expansion refers to the use of mixed methods to “explain or expand upon the understanding obtained in a previous strand of a study” (Venkatesh et al. 2013, p. 26), and completeness refers to the use of mixed methods “to make sure a complete picture of a phenomenon is obtained” (Venkatesh et al. 2013, p. 26). In both cases, Venkatesh et al. (2013) suggest following a quantitative study with a qualitative one. In the qualitative phase, we examined both the previously developed research model and additional countermeasures in the RPA context. In the subsequent chapters, we describe the methods used in this work.

We performed an online experiment (Fink 2022) to assess our research model (see Section \* MERGEFORMAT 3.2). Online experiments are an increasingly common method in IS research (Fink 2022; Karahanna et al. 2018). Experiments have advantages regarding the ability to make causal inferences (see, e.g., Karahanna et al. 2018). Moreover, when participants are randomly assigned to treatment and control groups, “omitted variables are equally distributed across these treatments and the explanatory variable […] is thus truly exogenous" (Sajons 2020). For this analysis, we used structural equation modeling (SEM) with partial least squares (PLS) (Hair et al. 2019, 2017), a well-accepted method in IS research (Petter 2018).

Following the guidance of Yin (2018), we conducted a positivist case study to deepen our understanding of SQB in the context of RPA. We base the investigation on the literature on SQB, especially on Kim and Kankanhalli (2009) and Godefroid et al. (2022).

First, we tested the influence of SQB on RPA adoption via an online experiment. The hypotheses tested in the experiment are described in Sect. 2.2. Table 3 presents the path coefficients that we assessed to test our hypotheses. We analyzed significance on the basis of the p-values of the path coefficients.

Figure 2 presents the results graphically. In full support of H1, we observe a very strong, significant, positive effect of performance expectancy on the behavioral intention to use RPA (H1.1 +). We also observe a significant effect of effort expectancy on the behavioral intention (H1.2 +). We expected this result, as several studies have tested the UTAUT across domains (Venkatesh et al. 2012).

Full size image

We observe only partial support for H2. We note a highly significant, positive influence of the net benefits on the performance expectancy (H2.3 +) and the effort expectancy (H2.4 +). We intuit that a positive perception of the benefits of RPA is reflected in a positive expected performance and a high perception of usage ease. We also observe a significant effect of control on the performance expectancy and the effort expectancy (H2.7 + , H2.8 +). This finding is also logical, as individuals who feel more in control of the technology will be confident about their performance and effort expectancy. However, we find no significant effect of loss aversion on either performance or effort expectancy (H2.1; H2.2). Thus, in our case, the participants did not fear the loss of the current way of working. This effect could arise because the first processes to be automated are often rather repetitive administrative tasks (Syed et al. 2020a, b) that may not be missed. Similarly, we find no significant effect of social norms on performance or effort expectancy (H2.5-, H2.6-). This finding implies that, at least in the RPA context, the opinion of colleagues and friends does not affect the perceptions of the technology.

In contrast to our expectations, which were developed on the basis of the literature (see Sect. 2.3), we observe only weak support for H3. We observe a weak effect of the success story on one SQB construct and no significant effects of the additional information treatment. The success story treatment slightly positively influenced the perception of net benefits (H3.6). This finding aligns with the literature that has recommended success stories as a tool to influence rational decision-making (Linnerud et al. 2019; Kim 2011; Shealy et al. 2019; Bekir and Doss 2020), but we expected a stronger effect. Interestingly, we cannot confirm that simply giving individuals additional information about the technology and the associated change is a valid countermeasure. Thus, at least in our specific context of RPA adoption, this often discussed recommendation from the literature (Hsieh et al. 2014; Zhang et al. 2016; Merriman et al. 2016; Weiler et al. 2019; Lorenc et al. 2013; Henkel et al. 2019; Kim 2011; Li et al. 2016) does not seem to apply.

The limited effect of our countermeasures could be explained in four main ways: (1) The first possibility is that our participants did not suffer from SQB concerning RPA, or that the model, originally developed in an ERP context, might not fully apply to the RPA context. This explanation aligns with the data obtained thus far, which indicated that only the net benefits and control significantly influenced the technology adoption constructs. These two concepts also appear in contexts beyond SQB (Lee and Joshi 2017). It seems unlikely that SQB should not apply in this situation, given that prior studies have found SQB effects in multiple circumstances (see above). In our hypothesis development, we identified three constructs (transition, uncertainty, and sunk costs) that did not fit the characteristics derived from the literature. The SQB constructs might need to be reconceptualized in the RPA context. (2) The second potential explanation follows media richness theory (Ishii et al. 2019). The provided texts might not be rich enough to sway people’s opinions, given the complexity of the technology. Therefore, a video format might be more effective for the level of complexity required for delivery. (3) The third possibility is that not all of the often-cited countermeasures work against SQB. If so, researchers should reconsider their recommendations, and practitioners should allocate resources elsewhere. Our use of real-world examples aimed at encouraging NGO-affiliated individuals to adopt RPA implies that even the authors of these texts might need to reevaluate their strategies. (4) If the SQB constructs themselves or their influence varies by context, the countermeasures would also differ in their effectiveness. Table 4 summarizes our hypotheses and findings.

Given the unexpected and intriguing results of our experiment, we decided to conduct a comprehensive qualitative analysis. Therefore, in the next step, we present a case study examining both the SQB model and potential countermeasures in the RPA context.

In our data analysis, we identified several aspects of SQB at work. We identified countermeasures that have been established in the literature (Godefroid et al. 2022) and additional countermeasures from practice. In the following, we highlight the differences in the SQB constructs in the RPA context and then detail the observed countermeasures.

SQB has often been used in IS to better understand user acceptance of new systems in a variety of contexts (see e.g., Kim and Kankanhalli 2009; Lee and Joshi 2017; Godefroid et al. 2022). However, our findings support the notion that existing models of SQB (Kim and Kankanhalli 2009) do not fully explain SQB in the RPA context. Different literature reviews have highlighted the need for research on RPA adoption (Plattfaut et al. 2022). While several works have researched rational (non-biased) decision-making regarding RPA acceptance (see e.g., Tschandl et al. 2022; van Looy 2022; Vollenberg et al. 2024; Wewerka et al. 2020), an examination of biased influences has thus far been missing. Our experiment and subsequent case study show that SQB affects RPA adoption and that it functions differently than does the adoption of other technology classes. Apparently, the antecedents of intention to use a technology from traditional user acceptance research are stable across different contexts. However, the single components of SQB across the three dimensions (i.e., cognitive misperception, rational decision making, psychological commitment) are context-sensitive, both regarding nuances in their definition and regarding their importance. In the case study, we uncovered possible explanations for the changed behaviour as well as mechanisms leading to this difference and new countermeasures specific to RPA. Those findings enabled us to re-conceptualise SQB for RPA.

In the first study, we revealed a significant influence of SQB on RPA adoption, essentially in response to RQ1. However, unexpectedly, the tested countermeasures were not determined to be effective overall (RQ2). Among the literature-based constructs used to measure SQB (Kim and Kankanhalli 2009), net benefits and control affect RPA technology acceptance. Moreover, unexpectedly, loss aversion and social norms did not have an effect. These findings indicate that the SQB studies must be examined closely with respect to their transferability to user-driven technologies such as RPA. The literature and the known effects relate to more traditional IS, such as ERP systems (Godefroid et al. 2022). The findings revealed by the first study highlight the need to test existing SQB countermeasures in this context and to find new countermeasures, as simply providing users with more information on the change is insufficient; instead, our case study revealed the necessity of obtaining an in-depth understanding of how the bot operates. Similarly, simply presenting success stories has a limited impact, at least in the RPA context. Again, our case study might offer an explanation, as we noticed that social norms present initially during the RPA introduction are not relevant throughout the RPA implementation. This finding explains why the treatment did not have a significant effect on the experiment and adds to the body of knowledge concerning SQB (Kim and Kankanhalli 2009).

In the case study, we gained further context into how the antecedents and countermeasures work to answer RQ3. The findings allowed us to reconceptualize SQB in the RPA context. We found evidence indicating that SQB functions differently in the RPA context. For example, we noted that uncertainty costs in the RPA context include the perceived interaction with the technical environment and that control can be aided by the availability of skilled key users. In addition, we found the construct of sunk costs to not be relevant in the RPA context, which stands in contrast to the SQB findings pertaining to other technologies (Rimbeck et al. 2024). We identified eleven countermeasures that were employed in the case company. While we observed some initial signs of low acceptance, by the time we conducted the interviews, acceptance had increased significantly. Therefore, the countermeasures seem to have worked in the two observed strands of development. While seven measures appeared in the SQB literature (for an overview, see Godefroid et al. 2022), we identified four new countermeasures in the case study.

Interestingly for the RPA community, six of the eleven identified countermeasures were previously known in the literature on RPA CSF (Plattfaut et al. 2022). This result highlights the relevance of SQB regarding RPA: If actions that serve as SQB countermeasures have been identified as critical for the success of RPA implementations, then overcoming SQB could be critical to successful RPA implementations, and practitioners are already doing so, possibly without being aware of SQB as the reason. Therefore, our findings on how SQB operates in the RPA context and the identified countermeasures can help researchers and practitioners design RPA initiatives that are more successful.

In the following sections, we discuss our findings in light of the extant SQB literature; reconceptualize SQB in the RPA context; and critically reflect on the bias perspective and its value in the RPA context and, in turn, in the BPM context.

Our findings contribute to the emerging stream of literature on RPA, the theory of BPM, and the cognitive bias research in the IS domain. Moreover, we contribute to the knowledge base on integrating cognitive biases into the literature on technology acceptance.

We advance our understanding of RPA, especially in light of the scarce literature on its acceptance and implementation (see, e.g., Syed and Wynn 2020; Waizenegger and Techatassanasoontorn 2020; Santos et al. 2020). This paper provides insights into the mechanisms of RPA acceptance, how it can hinder implementation projects, and possible countermeasures. Research can adapt existing RPA implementation frameworks to account for these findings. SQB can hinder or prohibit successful system introduction, which should be particularly relevant when the implementation relies on specialty departments being willing and motivated to implement and operate information systems.

Similarly, we contribute to the literature on user-driven IT, regarding which an SQB perspective is also missing. Offering business departments the ability to create and alter their own software is one of the reasons why organizations utilize RPA (François et al. 2022), and RPA projects rely heavily on the involvement of process participants and potential RPA users in the development and operation of automations (see, e.g., Oshri and Plugge 2022; Syed et al. 2020a, b). When users are responsible for creating and maintaining their own IT systems, user acceptance naturally plays an important role, but acceptance also seems to operate slightly differently for user-driven IT than it does for traditional IS (McGill et al. 2003). Regarding RPA, we showed that the increased involvement of end-users in the process of creating system specifications, compared with traditional requirements engineering, leads to a greater perception of transition costs and additional maintenance efforts, resulting in a decreased perception of net benefits. Since, in the case of user-driven IT, the burden of these efforts (in addition to the actual development effort) relies entirely on the users’ shoulders, SQB could play an important role in this context.

More broadly, we contribute to the wider BPM and process automation literature (see e.g., Grisold et al. 2024). Our research also suggests that organizations must address the cognitive biases of employees, who are asked to either share process information with external consultants who are responsible for developing process automations or to develop their own automations via a citizen developer approach. Our research showed that the decision to cooperate in automation with RPA is not entirely rational. Human actors are biased in their perspectives, especially toward the status quo. However, SQB functions differently in the RPA context than it does in traditional information systems (Kim and Kankanhalli 2009). As such, the emergence of new technologies (like RPA) emphasizes the importance of incorporating behavioral aspects such as technology acceptance into the field of BPM. Aided by such insights, managers can actively shape the dynamics when new process automation techniques such as RPA are introduced.

Our findings extend the insights that have been presented concerning SQB measurement, countermeasures and the influence of this factor acceptance (RQ1). In line with the literature review by Lee and Joshi (2017), we find that not all of the SQB constructs are relevant in the RPA context. In line with our initial hypothesis, we observed that the sunk cost did not have a significant influence. We also contribute to a reconceptualization of SQB in the specific context of RPA. We build on the work of Kim and Kankanhalli (2009), who reconceptualized the initial definition of bias provided by Samuelson and Zeckhauser (1988) for information systems. However, our findings highlight the necessity of developing and adapting such models to suit specific IS contexts. Only by formalizing these influences for the IS domain can we begin to measure them and counter them actively. Similarly, we contribute our four additional SQB countermeasures, whose general applicability remains to be tested.

In addition, we show the value of integrating cognitive biases into the literature on technology acceptance. In line with prior research (Kim and Kankanhalli 2009), SQB was one factor inhibiting the acceptance of the technology (here: RPA). As such, we argue that studies on the acceptance of technologies need to take bounded rationality into account. Based on this, we derive implications for future research.

Our work has several implications, which we situate by reference to the core topics of BPM and RPA research, thus strengthening our understanding of the systematic differences of new emerging technologies and advancing the cognitive bias research in IS.

Our findings contribute a behavioral view on the topic of BPM, stressing the importance of including nonrational behavior in the BPM toolbox. While several studies have researched RPA acceptance from a rational perspective (see, e.g., Tschandl et al. 2022; van Looy 2022; Vollenberg et al. 2024), more research is required to follow the guidance offered by Simon (1955), to stop basing our models and theories on the false assumption of rational behavior. The cognitive bias research in IS has taken significant steps in this direction, with many biases identified and studied in the IS context (Godefroid et al. 2021). While we provided a conceptualization of SQB for the RPA context, similar conceptualizations are still missing for other contexts. Such conceptualizations are essential for understanding context-dependent acceptance behavior, as our example highlights. Further research is necessary to increase our ability to explain acceptance behavior in this context. Future empirical research should also determine whether and how our findings can be transferred to other BPM or IS technologies.

The literature also proposes a variety of potential SQB countermeasures (Godefroid et al. 2022), which remain to be tested for their effectiveness either individually or in combination. While we showed that a combination of countermeasures was effective for our case company, different combinations of countermeasures (including those identified in this work) should be evaluated in different contexts to examine which characteristics of those contexts contribute to the success of the countermeasures. As SQB appears to operate differently in different contexts, further research is needed to identify the other contexts in which RPA SQB (and, by extension, user acceptance) functions differently as well as to determine which countermeasures are effective in which contexts.

On the basis of our findings, this research also contributes to managerial practice by providing managers with a better understanding of why they should expect certain behaviors and why they should apply specific countermeasures as well as by expanding their repertoire of applicable SQB countermeasures. Understanding RPA in comparison with traditional systems can help relevant actors conceptualize corresponding training concepts.

Our case study highlights the importance of giving those tasked with handling RPA bots a general understanding of how bot interaction functions. Whereas an ERP system user may find knowledge of how the ERP’s interface with Microsoft Outlook functions to be irrelevant, the “end user” of an RPA bot must understand the nature of its interaction with its (technical) environment to anticipate potential sources of errors and to assess the technologies’ capabilities correctly. Thus, our findings should help managers and consultants to create organizational policies that support their employees in the process of implementing the necessary changes more effectively. This, in turn, can help increase the success rate of RPA projects.

In addition, we provide managers with hands-on measures that they can implement in RPA development projects. The countermeasure to reduce the transition cost, for example, is especially important in BPM contexts. Our findings indicate that the effort to create the necessary process documentation when automating with RPA is not a trifling matter, and end-users have to shoulder a higher percentage than in regular projects.

Even though we applied the utmost care in designing our research, some limitations remain: First of all, we admit that the multi-method approach presented here was not planned as such (see Sect. 3.1). Naturally, a different methodological approach might have led to slightly different conclusions. However, in line with the arguments of Venkatesh et al. (2013) and Pratt et al. (2022), we believe that following up on the discrepancies between the expected results and our experimental results was the right course of action, that our findings are worth sharing, and that they can hopefully convince other researchers to engage with this interesting topic.

The first study involved only 150 participants, which was sufficient only to provide a strong indication, not guarantee a final result. Nonetheless, we believe that our results highlight the importance of researching SQB countermeasures further. Second, although we synthesized our measurement model from the literature, as we selected only some of the concepts due to our context, our choice might not have exhausted the full range of relevant concepts. Third, other studies have combined SQB with user resistance. Therefore, in our research effort, we expected an inverted effect for the intention to use. However, intention to use and user resistance might be more dissimilar than we thought, thus explaining the lack of significant effects observed with respect to most of the SQB constructs. Fourth, the use of Prolific-provided (vs. directly recruited) participants might have skewed the results despite the positive results that Prolific has yielded in previous cognitive bias research (Bahreini et al. 2020) because we could not verify the exact nature of the participants’ nonprofit affiliation or the type of nonprofit work with which they were involved. Moreover, the participants on Prolific exhibit a slight bias toward being from Western, educated, industrialized, rich, and democratic countries (Prolific 2025). As such, the sample might be biased. Furthermore, especially in online experiments and surveys, non-response bias might be a problem. When we compared early and late responders to test for nonresponse biases (see, e.g., Koch and Blohm 2016), we did not observe any statistically significant differences. This result suggests that nonresponse bias might not be at play. In addition, following media richness theory, the text format we chose might not have been sufficient to relay the complex level of information required to understand RPA (Ishii et al. 2019). Finally, different texts, e.g., a success story written by a nonprofit and not a software provider, might have had different effects.

Similarly, the expressive power of a single case study with a moderate number of interviews is always limited. Nonetheless, the data were sufficient to challenge some of our hypotheses according to the positivist case study approach, thus allowing us to contribute important theoretical insights in this context.

Finally, our theoretical model imposes a practical limitation: In the qualitative interaction with practitioners, we encountered reservations regarding the bias perspective. In this context, the model developed by Kim and Kankanhalli (2009) has certain drawbacks, as it includes rational aspects to capture bias. This procedure implies that people are biased because they (perhaps rationally) determine that the net benefits of a new IS for their work do not always garner favor. As the IS research community is actively engaged with the practitioners’ domain, we need to be careful to develop theoretical models that researchers and practitioners can apply without offending their counterparts.