Abstract
Generative artificial intelligence (GenAI) is growing in popularity, yet insights on its application to business model innovation (BMI) are scarce. Drawing on a qualitative, multiphase study involving interviews, focus groups, and digital diaries with strategy consultants, we explore how professionals engage with GenAI during BMI. Our findings suggest strategy professionals engage with GenAI through what we term reflexive augmentation, which represents the deliberate, critical engagement with GenAI to decide which tasks should and should not involve GenAI and which tasks to automate or augment through GenAI. We show how this process is shaped by four tensions related to trust, skills, value-add, and client disclosure. We offer actionable insights for managing human-AI collaboration, advancing debate on augmentation and automation at the micro-level, and suggest how organizations can support effective GenAI integration in innovation contexts.
Keywords
Generative artificial intelligence (GenAI), and artificial intelligence (AI) more broadly, 1 are put forward as powerful contributors to the field of innovation. 2 AI represents a “highly capable and complex technology that aims to simulate human intelligence.” 3 GenAI, a type of AI, employs deep learning models to synthesize various types of content, including text, code, audio, visual, and three-dimensional entities, in response to user prompts. 4 Popular, commercially available examples of GenAI models include OpenAI’s ChatGPT tools, Google’s Gemini, Anthropic’s Claude, and Meta’s Llama, all of which are easily accessible and knowledgeable about a great variety of fields. 5 The advent of GenAI represents a significant shift in AI’s technological capabilities. Indeed, according to Krakowski, GenAI represents “a significant leap beyond the capabilities of predictive AI” and, consequently, the theories and insights based on the capabilities of predictive AI do “not adequately address how humans can be expected to interact and work with GenAI.” 6
Earlier in its history AI was mainly considered suitable for automation. AI was typically fully entrusted with well-structured, largely routine analytical tasks. With the recent advent of GenAI, that profile has undergone a change. Today the technology is considered capable of augmenting human performance of highly complex cognitive work, 7 including creative endeavors. 8 AI scholars note that an organization’s decision to use AI to automate or augment can be assessed at the organizational, individual, and task levels. 9 To date, however, management scholars have predominantly focused on AI’s automation and augmentation potential at the organizational level, despite calls to explore the interactions between humans and machines at the micro-level to enhance AI integration management in the workplace. 10 The lack of empirical research on the microfoundations of GenAI use, also in innovation management, 11 is problematic considering the potential of GenAI to enhance outcomes. 12 As suggested in prior research on predictive AI, individuals in the workplace may respond to this technology in strategic ways that may not be anticipated by management (e.g., to protect interests and resources), which ultimately may hinder effective technology integration at the firm level. 13
To further explore human-machine collaboration at the micro-level in the current age of GenAI, we explore GenAI use by professionals for business model innovation (BMI). BMI is a strategic process to change the way in which firms create, distribute and/or capture value. 14 BMI includes different phases with different (strived-after) interim outcomes, 15 thereby providing the opportunity to study the use and impact of GenAI for different types of activities. We specifically examined, at the individual level, choices to either automate or augment tasks and possible challenges or opportunities that might result when using GenAI during BMI.
We collected data from strategy professionals working for a multinational company delivering strategic consulting services to external clients. Data collection took place in two stages, by means of different qualitative techniques, including in-depth interviews with informants (n=49), focus group discussions (n=4), and digital diary keeping (n=8). In addition to primary data, we gathered relevant secondary data, such as guidelines on GenAI use, as developed by the sampled organization. Based on an analysis of these data we distilled how the sampled strategy consultants tend to engage in BMI. We also identified for which specific business model activities consultants used GenAI, automating or augmenting these activities through GenAI use. We identified four types of tensions accompanying GenAI use. Our research has managerial relevance because a better understanding of the interactions between employees and GenAI technology may facilitate technology-implementation strategies at an organizational level. Our research also has theoretical relevance because we provide more fine-grained insight into human-AI collaboration at the micro-level, having inducted a process we label reflexive augmentation. This process is used by individuals to deliberately and critically reflect on GenAI use at the task level and on the tensions related to GenAI use that shape reflexive augmentation.
GenAI: Pitfalls and Benefits
With GenAI being an emergent technology, relatively little is yet known about how individuals entangle with it in the workplace and how this affects technology adoption. A micro-foundational perspective studying the use of GenAI on an individual level is relevant because individuals’ actions and cognitions underpin organizational performance, including effective integration of technology. 16
In terms of barriers or pitfalls for GenAI adoption at the individual level, there is research suggesting that factually incorrect GenAI outputs will reduce actors’ trust in the technology, which in turn can create a barrier to its further adoption. 17 In addition, an individual’s decision to adopt GenAI may be hindered by the technology’s limitations in terms of “feeling capabilities” 18 as well as its “inscrutability,” 19 and the opaqueness of the process it uses to source and create a specific output. 20 Emerging evidence that the technology might create or perpetuate racial and social biases may also negatively affect individual adoption. 21
In terms of benefits or value-add, emerging empirical research suggests that GenAI use may result in greater agility and resilience for employees and increase worker productivity. 22 Several scholars have pointed out, however, that the extent of that benefit depends on individuals’ skills and experiences. It has been demonstrated, for example, that knowledge workers’ lack of clarity about GenAI’s (ever-expanding) capabilities can result in suboptimal utilization. 23 Other research suggests that GenAI productivity improvements are heterogenous across workers: GenAI assistance seems particularly beneficial for novice and low-skilled workers with less impact on experienced and highly skilled workers. 24
Innovation is one of the key fields where further consideration of how to leverage GenAI is needed. 25 Existing empirical work establishes the positive effects of GenAI use at the task level, facilitating creative activities such as idea generation 26 and analytical tasks such as summarizing and categorizing information. 27 There is also some conceptual work describing the potential benefits of GenAI for BMI specifically, 28 in addition to conceptual research on how GenAI might evolve from an innovation tool to an autonomous driver of innovation efforts. 29 There is hesitation as to whether truly novel innovation will emerge through GenAI use or whether GenAI will result in less diverse solution spaces on a more aggregate level. 30 Relatively little empirical research is currently available about how humans interact with GenAI for innovation purposes, including for BMI, and, thus, multiple scholars have called for further detailed research. 31
Automation and Augmentation
In terms of theoretical framing of AI use, the perspective of automation or augmentation through AI is considered particularly insightful in the context of this research. 32 Automation is a process whereby machines (e.g., AI) execute a formerly human task with little to no further involvement of human actors with the intent of enhancing efficiency, rationality, and comprehensiveness. 33 Augmentation, on the other hand, represents close human collaboration with machines (e.g., AI) to perform a task 34 to enhance and complement, for example, the human’s ability to use intuition or common-sense reasoning 35 and the machines’ ability to overcome humans’ bounded rationality through high-speed processing of massive amounts of data and complex pattern recognition. 36 Human-AI augmentation might be described as hybrid intelligence, a synergistic relationship between humans and machines where each has complementary strengths that can be combined to augment each other. 37
Empirical research examining the microfoundations of hybrid intelligence and AI-in-the-loop workflows, exploring in-depth how humans might automate or augment activities through AI, is relatively limited in nature. 38 Recent empirical research investigating how professionals—lawyers and accountants—decide to integrate predictive AI through automation or augmentation found that they actively protect their interests and resources, carefully delegating specific work to the AI while creating new types of professional work to replace the work “lost” to AI. 39 Other research, focused specifically on GenAI and using experiments as data collection methods, found that in the context of highly creative tasks—fiction writing and song writing—augmentation and automation coexist and their relative influence is dependent on the type of human capital that users possess. 40 More specifically this research suggested that, while GenAI automates the domain-specific elements of creative tasks, thereby reducing the competitive edge of domain-specific skills, effective GenAI use also requires augmentation through more generic, broadly applicable cognitive skills, such as evaluation and synthesis skills. Another piece of empirical research investigating GenAI in the context of crowdsourcing identified concrete benefits for AI-in-the-loop workflows during idea generation, particularly when there was iterative human input, suggesting the importance of augmentation for creative problem-solving. 41
Despite this emerging empirical research on GenAI use at the task and individual level, there is still much we do not understand about how individuals use and experience GenAI when engaged in innovation.
Method
Our qualitative research examined professionals’ use of GenAI during BMI, including any accompanying challenges and opportunities that might arise. Primary and secondary data collection took place in two sequential stages: a preliminary stage and a main research stage.
During the preliminary stage (Stage 1) we collected data to establish the core activities conducted during BMI and the nature thereof. This stage was necessary due to the lack of fine-grained information in the existing literature on the nature of BMI activities related to novel business models (rather than incrementally improved models). 42 We interviewed thirty-eight informants who worked for a multinational company (anonymously referred to here as “ABC”) that delivers strategic consulting services to external clients, including advice on BMI. The informants were nonrandomly selected based on whether they had been involved in BMI activities in the past twelve months. This was determined by examining an internal database of company ABC, which included recently completed consultancy projects. We continued selecting and interviewing informants until data saturation was achieved (that is, no new information about the type of BMI activities and their sequence emerged). Each consultant was interviewed for an average of sixty minutes; interviews were conducted online and transcribed in full. The first and second authors performed the data collection and the first round of analysis. These preliminary insights were discussed and further refined together with the third author, who has over twenty years of experience in BMI. We also conducted three focus groups where we gathered four or five of the informants we had interviewed to deepen and refine emerging findings. Based on the data gathered we examined how the sampled strategy consultants engaged in BMI in terms of activities, outcomes, and temporal structure. We clustered the identified business model activities into five key phases, each with two different subphases including constituent activities and outcomes (see Table 1, which provides a summary of our results). While the intensity of BMI activities tended to be somewhat client dependent, in broad strokes, the sampled strategy consultants typically engaged in the BMI activities shown in Table 1 when hired by clients. The sampled strategy consultants tended to support client firms with the conceptualization of new business models; business model implementation is normally carried out without their active involvement. We classified the identified ten subphases as onstage or backstage based on whether the informants’ activities tended to be performed together with representatives of the client company and hence being directly visible or not. 43 The onstage, visible activities performed together with the client representatives are generally considered more valuable to the client and core to why consultants are hired. (Find further detail about the data analyses performed in this preliminary research stage in Appendix A.)
Summary of Staged Activities Involved in BMI and GenAI Use.
The informants we sampled did not mention any use case with GenAI in this specific subphase. That does not mean that there are no use cases for GenAI in this specific subphase. It only means that we did not find any examples in our dataset.
Our informants described “Big Ideas” as ideas with truly transformational potential in relation to a particular industry, for example by reshaping the boundaries of the industry, or bringing together participants from other ecosystems for broad-ranging, visionary, innovative future solutions.
In the main study (Stage 2) we explored more deeply the use of GenAI for BMI-related activities. The informants were not the same as those from Stage 1 but did work for the same multinational (ABC) as the informants from the preliminary stage. The informants were nonrandomly selected based on an internal call for research participants, in which we specified that ideal candidates would be strategy consulting practitioners with recent experience in BMI and who had used GenAI in their work in some way. We recognize that this may have biased our sample to relatively early adopters of GenAI with a level of interest in GenAI as a topic. However, as it became apparent within our findings, the sample was relatively diverse in terms of level of experience and proficiency with GenAI. Beyond BMI the informants also performed other activities related to strategy consulting more generally, but our research focus was on GenAI used for BMI. When we conducted the research, there were no company directives on GenAI use at ABC; the consultants had the agency to make decisions about GenAI use at the task level as they saw fit.
To collect granular data, we asked eight of the Stage 2 informants to complete a digital diary of their interactions and experiences with GenAI tools in their BMI-related work activities during an eight-week period. 44 We supplemented the diary data by means of conducting semi-structured interviews of sixty to ninety minutes with all the individuals who completed the diaries. During the interviews we probed further about the use of GenAI recorded in the diaries. We asked, for example, questions to determine whether GenAI was used to automate a task (e.g., “Did you use the GenAI output in an unedited way?”) or augment a specific task (e.g., “Did you spend time to improve upon or edit the GenAI output?” and “Did you use GenAI output as a starting point for, for example, brainstorming?”) and informants’ decision criteria to do so. These interviews were conducted via Zoom and transcribed in full. After the first round of data analysis we conducted a focus group with five informants who completed the digital diaries to further discuss emerging themes. In addition we conducted interviews with three senior company leaders to understand its GenAI-related strategy, guidelines, training opportunities, resources, and capabilities available for internal use. We reviewed existing internal documentation and GenAI models in use at the company.
To analyze the data gathered during Stage 2, we adopted an inductive approach that allowed us to inductively explore and analyze the complex, dynamic phenomena of human-machine interaction from the perspective of those experiencing it. We first analyzed the type of tasks (“use cases”) associated with GenAI and mapped them against the BMI activities we had identified during Stage 1. We then analyzed these human-machine interactions deeply, immersing ourselves in the data and coding statements reflecting GenAI use along with any associated challenges and opportunities. We agreed, upfront, to classify GenAI use as automation or augmentation in line with previous research. 45 We considered a task to be automated where an informant deployed GenAI to conduct said task by GenAI, without substantive further involvement from the individual. On the other hand, we coded for task augmentation when human and GenAI contributions jointly shaped task completion.
Emerging findings were discussed by all three authors and with the Stage 2 informants during the focus group discussions. We consulted extant literature for existing theoretical insights. Stage 2 data analysis (see Appendix A) resulted in the identification of different forms of tension arising from GenAI use and the delineation of a process that the informants used to make the deliberate decision to augment, automate, or abstain from GenAI use at the task level. Building on the theoretical insights of Schön on reflective practice, 46 we have labeled this process reflexive augmentation, and elaborate on it in more detail at the end of the following section.
Findings
Our findings are presented in three main subsections. The first contains a description of the BMI activities distilled from our findings and our analysis of GenAI use for specific tasks during BMI. The second subsection presents the findings related to the tensions experienced by the consultants while using GenAI during BMI. In the third subsection we interpret our findings at a more abstract level and set forth our notion of reflexive augmentation. We use informant numbers, rather than names, in our finding as agreed upon prior to data collection.
BMI Activities
Based on the findings from the first research stage, we identified BMI activities and how these activities were normally structured from a temporal perspective. We clustered these activities into five distinct phases, as shown in Table 1. Table 1 also includes the outcomes sought during each phase and the performance nature of each phase (i.e., onstage or backstage). The informants were generally engaged in all five subphases on BMI, although the degree of engagement could fluctuate per client firm.
As indicated earlier, our research informants had the agency to decide how and when they wanted to use GenAI. We found twenty GenAI use cases for BMI, 47 which we mapped to the BMI subphases shown in Table 1.
The use of GenAI was uneven across the different phases. The majority of the use cases we identified were related to the initial “Setting the Stage” phase of BMI (Phase 1), and involved analytical tasks related to industry and client research, the synthesis of large volumes of information, and an analysis of specific topics and hypotheses. We also observed GenAI use for more creative activities, such as developing user personas and bringing them to life through user journeys. We did not observe GenAI usage at all during certain BMI subphases, however, including the “Inspiration” phase (Phase 2), and the generation of “Big Ideas” phase (Phase 3b). These activities tend to be seen as more strategic and core in nature. They are performed onstage alongside the client, whereas the use cases where GenAI was applied tended to be more supportive of BMI-related activities that, while necessary, are typically performed backstage.
Tensions When Using GenAI
The GenAI models that the sampled informants chose to use included those that were proprietary to the company, as well as publicly available tools such as ChatGPT 4 and Llama 3. Even if assessing the GenAI models’ functionality was not germane to our research, functionality did influence the presence of tension (e.g., there was tension when the model did not provide a source for its output). ABC did not mandate GenAI use but there was evidence that they sought to motivate GenAI adoption, furnishing employees with prompt libraries, providing training, and ensuring access to public and private GenAI tools.
Notwithstanding these organizational initiatives the strategy professionals from our sample encountered several tensions when using GenAI models for BMI activities. We clustered these tensions into four types, which we will now discuss. The below subsections on the four tensions discuss the first-order categories listed in Appendix B. These first-order categories represent the perspectives, experiences, and language of our informants related to the specific tension. We illustrated these first-order categories using quotations from our interviews and used bold text to indicate when we refer to a specific first-order category. Appendix B provides a table summarizing all tensions and their constituting first-order categories. Appendix B also provides additional supporting quotations from our dataset.
Tension 1: Investing In, Yet Doubting Skills of/with GenAI
The first type of tension arose when the informants did not clearly understand whether their skills and abilities were causing the GenAI models to produce undesired outcomes or whether GenAI is actually incapable of accomplishing a particular task or activity. Tension 1 led the informants to wonder, “Is it me or is it GenAI?” There was either confidence or doubt (or both) regarding the individual’s and the technology’s skills and abilities in terms of achieving the outcomes sought from the GenAI interactions.
The informants had different perceptions of the tasks GenAI could perform and the related value they could obtain by investing in development of AI skills to “make it work.” They sought to obtain value from GenAI use by minimizing their personal effort in the development of AI skills—time is a key resource for consultants—but to produce that value, they needed to invest effort into learning and formulating effective engagement strategies for GenAI use. As informant 4 pointed out, to effectively use GenAI tools a consultant must invest considerable time in mastering these tools “to get the most out of it.”
Obtaining output from GenAI engagement typically occurs via prompts, which are natural language instructions used to provide the GenAI with context and guide its response generation. Some informants were more confident about their prompting skills than others. Informant 6, for example, shared the tactics he had developed over time to get the answers he sought from the GenAI tool:
“I treat it like a child, where I will give the GenAI tool almost more information than the tool needs. I just give it as much as I can think of. And then, I am very specific about structure. . . so the GenAI tool doesn’t have much room for interpretation.”—Informant 6
Other informants, however, were less confident about their proficiency at prompting:
“Maybe I’m not good enough at prompting. I tried a couple of different examples. Perhaps if I had tried some different things, it might have worked a little better.”—Informant 3
To stimulate individuals’ efforts to invest in the development of AI skills, management may stimulate education about GenAI’s capabilities. Some informants felt that the extensive training and available prompt libraries offered by ABC to help upskill with GenAI were insufficient for their needs. As informant 7 said, “We do have prompt assistants, but I think it’s still somewhat limited,” while informant 5 found that searching prompt libraries was “very time-consuming.” Some informants expressed uncertainty about whether the lack of a desired outcome was due to their level of skill, the GenAI’s overall capabilities (i.e., what it can or cannot do), or due to providing too little or too much context:
“I’m just also trying to figure out. . . is it because what I asked is too specific? Or is what I am asking for not clear enough, or does the GenAI require more context?”—Informant 2
When informants were doubtful about their ability to obtain the results they sought from GenAI or became frustrated while attempting to do so (e.g., when prompting was unsuccessful), this might lead them to abandon GenAI use for a specific task altogether, as noted by informant 7, who said, “If it didn’t give me what I wanted, I just moved on.” Informant 6 expressed a similar sentiment:
“The first attempt wasn’t what I wanted, but I just didn’t feel like taking the time to try and train it. I was late and stressed, so I just started writing.”—Informant 6
The decision to abandon GenAI use or not using GenAI for a specific task is influenced by contextual factors related to time or the accuracy level needed for a particular task, as informant 1 said, “If I only have half an hour before this is due to a client, I’m not going to take the risk of using a tool that didn’t work earlier.” This sentiment is mirrored in the words of informant 7, who said, “It takes time to ask follow up questions, to try to formulate my prompts, but then there are circumstances when I just look for speed.”
Tension 2: Acknowledging GenAI’s Value-Add, Yet Feeling Superior to GenAI
The second type of tension was related to the informants recognizing the value of GenAI for a variety of use cases but seeing themselves as superior in certain aspects, reflecting the sentiment that “GenAI is good, but not like me.” Some informants also questioned whether GenAI would ever be able to replace them entirely given its perceived inherent limitations when compared to humans.
Our informants shared use cases and activities for GenAI use that were seen as helpful. One set of tasks that the informants thought GenAI could perform autonomously, with no human moderation or editing needed, was summarizing large volumes of information. As one informant said:
“We use the GenAI tool for my project quite a lot because we do tons of interviews, and AI has been instrumental to summarizing all the interviews and producing an executive summary from the meeting notes.”—Informant 7
Another commonly described use for GenAI that involved creative tasks was asking it to provide an initial set of brainstorming ideas or hypotheses as a place to start:
“A lot of my use cases for the GenAI tool are for ideation. I want to get as many ideas as possible when using the GenAI tool, so some can be high quality, some probably not so much. That’s definitely of value, just providing many ideas.”—Informant 4
GenAI was also described by many of the informants as providing a helpful starting point rather than an end point for their creative endeavors, as noted by informant 2, who said it “can be used as a draft for thinking.” Informant 7 explained it like this:
“It provides a great starting point as I start to develop the detailed recommendations for our approach to tackling the program and envision what the future-state customer journey looks like for the client.”—Informant 7
Informants considered GenAI to be helpful with other creatively generative tasks, such as creating customer personas or journeys, especially when the informant had little industry knowledge:
“Probably the best experience I had with it was in building a ‘day in the life’ journey. This was for a life science or MedTech company that was looking to grow its diabetes segment. It made a whole story of different things I didn’t know about, like checking medication and a sample eating schedule. So that was really helpful.”—Informant 8
Informants’ perception that GenAI outcomes were often useful points of departure was also due to its perceived inherent limitations. They noted that the outcomes can be seen as generic and “fluffy,” with GenAI-generated wording sounding unlike what the informants would produce. Informant 3, for example, made the observation after receiving GenAI output that “no human would have ever written this response.”
Some informants noted that complex, high-value, time-consuming tasks—such as creating client presentations—were not yet something a GenAI tool could do effectively:
“What are the PowerPoint capabilities of this? The problem with that is, it only reads the text and only exports the text to a PowerPoint slide. But the whole point of PowerPoint is that it’s not a Word document.”—Informant 1
The current limitations of the technology mean that the informants were not worried that their positions would become fully replaced. They saw it mainly as a technology that augments their performance and related outcomes—now and in the near future:
“It has already been a pretty invaluable tool for me, but I definitely don’t think it’s something that would fully replace what we do, just based on the fact that it doesn’t really have that human aspect as much; but it is really good for parts of the job that are not human oriented or needing that emotional aspect.”—Informant 5
Some informants expressed sadness that the technology is taking over certain parts of their jobs that they enjoy, such as specific creative activities. Informant 8 said, “I am still sad that GenAI is replacing us a little. I like the creative parts. . .. I have a sad feeling about the creative stuff that’s dying.”
While the informants in our sample had the agency to decide how to use GenAI technology in their work, there was an understanding that adopting this technology was not entirely optional, given that the company was actively stimulating its use. There was also the acknowledgement that using GenAI might become crucial to a consultant’s success:
“You used to get a big head start with the right education and the right networking skills. I think, increasingly, the right familiarity and flexibility in using these GenAI tools will be another key differentiator for people in their day-to-day work.”—Informant 4
Tension 3: Trusting, Yet Distrusting GenAI Outcomes
The third tension was related to wanting to trust GenAI, but feeling unable to, which led some informants to wonder, “Can I really trust GenAI?” While informants did want to trust GenAI, given its potential to make their work more effective and efficient, the limitations of the technology gave them pause. Sometimes that hesitation was associated with unreliable outcomes, resulting in pressure to check GenAI output:
“It’s tough because you want to trust it, it makes your life a lot easier to trust it, but obviously you want to make sure it’s right. . .”—Informant 6
All informants touched upon the theme of mistrust with reference to GenAI, citing inaccurate outputs and the tools’ collective lack of ability to source and validate information, as well as skepticism about the tools’ black box nature. This mistrust did not necessarily lead to nonadoption. Instead, informants tended to take remedial action, such as fact-checking, to mitigate their mistrust of the tools’ output. Some informants did indicate that fact-checking was not always possible due to various reasons (e.g., time pressure or lack of expertise):
“It’s a little dicey because the GenAI outcomes don’t have a source, so you’re taking this thing as gospel; but at the same time, you have to, ’cause you operate under time pressure. So I definitely took some time to double check a lot of things that didn’t seem right, but that was based on what I thought didn’t seem right, and I am sure there were things I’ve missed that were wrong, but I haven’t double checked.”—Informant 1 “There’s a balance there—I’d say it’s about ‘How much do you trust?’ versus ‘How much do you need to double check, not being a subject matter expert who knows what is obviously wrong?’”—Informant 6
One tactic some informants used to deal with their mistrust of GenAI technology was to consciously adjust the way they used the technology, only using it for use cases or activities where factual information is less relevant:
“When it’s qualitative, when there is no right or wrong answer, it’s a great ideation assistant. I would say I take like 65 percent of what it comes up with, as far as the ideation, but that’s great because that significantly reduces my time.”—Informant 5
Overall the informants did not tend to be emotionless when discussing GenAI limitations. One informant, for example, approached the GenAI as though it were human, reproaching it for its confidence despite its unreliability, generating further mistrust:
“The issue that irritates me the most is how confident it [GenAI] will be at providing you a response, and then you tell it ‘You’re wrong,’ and it’s like, ‘Oh, yes, I was wrong, but here’s the right answer.’ Okay, well, I don’t trust that any more than your last answer, and you seem just as confident.”—Informant 3
Activities where a higher level of accuracy is needed saw a much higher level of augmentation by consultants, whether by fact-checking or otherwise cross-referencing the accuracy of the AI output. In contrast, in cases where there was not necessarily a right or wrong answer, the informants tended to accept the outcomes without further verification, though in some cases these were used as brainstorming fodder or as the jumping-off point for ideation.
Tension 4: Being Open, Yet Closed About GenAI Use
The fourth and final type of tension was related to informants hesitating to disclose whether they had used GenAI, wondering, “To reveal or to conceal?”
In our setting this tension was specifically related to whether the informants had disclosed their use of GenAI for specific tasks or activities to their clients. A client typically hires a consultant because they have specialist expertise or knowledge that the client does not possess. Clients expect that consultants will deliver value that justifies their (often substantial) fees. The informants recognized this and a major issue they raised was managing their clients’ perceptions of value when using GenAI. Our findings suggest that the informants might be more willing to disclose their use of GenAI for activities that they saw as of lower value or as value-add to clients. They were less likely to disclose the use of GenAI for activities that they perceived as delivering higher value to clients.
Cases where the consultants tended to be more revealing about their use of GenAI were when the tasks performed by GenAI were perceived to be lower in value, but which create client efficiency gains, such as synthesizing information:
“What worked really well is when we use the tools to automate our tasks, even just saving time for our own team. It cut eight hours of work from our side, and it shows value, shows them the clients that we are able to automate this repetitive manual task. And they’re super happy when they see the executive summary of each individual interview. They’re like, ‘Wow, you guys save the time and you’re also saving our money.’”—Informant 7
Another instance where GenAI use was more likely to be publicly shared by informants with clients was when it had created additional output that was not expected as part of the formal agreement between the consultants and the clients, such as efficiency gains in the form of value-add:
“The only thing in the last year that was obviously AI generated, that I put in front of a client, was a mock-up of future car interfaces: coming up with a few different future-looking cars that fit the specifications that they were talking about. But they hadn’t paid us to render a new car. It was just a visual that helped out with grounding some of these ideas in a real thing.”—Informant 1
Another scenario was openly using GenAI, jointly with the client, where GenAI tended to be used as a real-time assistant to converge information generated in a workshop:
“We brought it into client workshops where we got a bunch of different people to input what they wanted their marketing vision to look like. And then we had the GenAI tool create a whole new vision statement for the company based on all of these different inputs. . .. And it created this really nice statement.”—Informant 6
Because our informants are engaged in and financially rewarded for the consulting work they do, which involves critical thinking, when GenAI indeed facilitates that work, they concealed that information to avoid diminishing the perceptions of value or unique expertise they bring to clients:
“I would not tell my client that I use GenAI. I don’t want them to think that I’m using GenAI for this. I want them to think that I’m doing a lot of research and I’m working my *** off, and that, like, I’m really smart.”—Informant 5
Some informants also suggested that their clients did not necessarily need to know the mechanics used to generate results, might not even care about them, or would consider them irrelevant; hence, they remained concealed.
“If the clients are expecting something in twenty-four hours, and I give it to them in two, it doesn’t matter if I used AI, if I did it myself—really fast—if I asked a friend, or if I got a genie to do it. They’re paying the same amount; they’re just getting more. . ..”—Informant 1 “The clients just don’t care how you get there. And plus, we still need to do a bunch of work, because GenAI is mostly just used for the brainstorming of foundational starting points.”—Informant 7
Finally, some informants reflected on the professional boundaries and ethical challenges of being caught using GenAI without disclosing it:
“I would be personally and professionally embarrassed if, in a slide deck somewhere, there was a stray input not deleted. I think that’s like cheating on your homework, even though we’re all supposed to be doing it, it’s just professionally. . . it would feel weird.”—Informant 1
Some informants suggested that the buck stops with them in terms of GenAI output accountability. Put differently, the informants did not fully automate certain tasks and activities but took on a developmental editor role and, thanks to this, the clients receive a level of assurance they would not be able to obtain by directly engaging with the technology on their own:
“I want GenAI to get really, really smart. But I actually do like that it messes up. I want it to mess up. I don’t want it to be completely perfect because if it’s completely perfect, I stop questioning it. That’s where it becomes unethical for me to present it to a client, because that is not my work anymore. Why am I charging a client X dollar amount an hour when they could have just searched this on ChatGPT, and they didn’t use any of my brain? You can’t hold a computer accountable.”—Informant 5
Reflexive Augmentation
Based on our findings, we identified a process that we have named reflexive augmentation, which is the deliberate, critical engagement of a human actor with GenAI in innovation contexts. With this concept, we built, in part, on the work by Schön that established how professionals, through “conversations with the situation” engage in reflective practice to adapt, learn, and improve their practice, either in real time or afterwards. 48
Reflexive augmentation is focused on realizing hybrid intelligence, with individuals deliberately and critically reflecting on GenAI use and adapting and acting accordingly. The inclusion of the term reflexive indicates that the cognitive process goes beyond deliberately reflecting upon one’s ways of working; it is a process that shapes actors’ actions and practices in terms of GenAI use in the present and future. For example, strategists in our sample adjusted their GenAI use after assessing GenAI outcomes lacked reliable source information, which resulted in GenAI automation being limited to tasks that did not require a high level of accuracy. Reflexive augmentation is a process that is not solely reactive (i.e., thinking back) but also proactive and forward-looking in nature. This is demonstrated, for example, by sampled strategists who reflected on the implications extensive use of GenAI might have on their future job prospects. Reflexive augmentation is a process that includes contextual understanding. This is demonstrated, for example, by sampled strategists who made a distinction between activities being performed in front of or together with clients, which we labelled onstage activities, versus activities without clients, which we labelled backstage activities, when deciding on GenAI use.
Reflexive augmentation is shaped by tensions that individuals may experience when making sense of GenAI-related events, feelings, or actions. As visualized in Figure 1, these tensions induce actors to reflect on the nature of tasks (e.g., front stage versus backstage), their own professional value both for clients and their own organization, ethics-related issues related to GenAI use, and (task-specific) capabilities of GenAI tools and their own GenAI abilities. Assuming there is human agency to decide on GenAI use, the reflexive augmentation process resulted in decisions to either automate a specific task, reject the use of GenAI for a specific task, and/or augment a specific task by working with GenAI when engaged in innovation.
Framework for Reflexive Augmentation.
Reflexive augmentation at the individual level is also influenced by contextual factors, including the organization’s GenAI use policy, the existence and extent of organizational resources for effective GenAI use, and available GenAI model functionality. While we did not explicitly study outcomes, based on tentative evidence (e.g., informants suggesting enhanced efficiency through GenAI use) and prior research, 49 we suggest that effective GenAI use will generate value at the individual and organizational level, for example enhancing efficiency and effectiveness of innovation processes. GenAI use or nonuse by individuals at the workplace will also, ultimately, influence effective technology integration at the firm level. Our conceptual framework is visualized in Figure 1.
Conclusions
This study explored interactions between professionals and GenAI technology to gain insights into the opportunities and challenges of integrating GenAI in the workplace from a micro-level, human-centered perspective. Data were collected with a focus on consultants engaged in innovation, specifically BMI. The findings revealed that GenAI can be used differently during specific business model phases, and specific tensions can arise during GenAI use. From these findings we identified a reflective process that we call reflexive augmentation, which professionals adopt to decide how to use GenAI at the task level and realize hybrid intelligence.
Theoretical Implications
Our research on human and GenAI-related dynamics presents different theoretical contributions. First, we answer the call from prior research to empirically examine human-AI collaboration at a micro, individual level. 50 Much of the research on automation or augmentation was published before the widespread diffusion of GenAI and so the decision to automate or augment a task through AI, whether predictive or otherwise, was predominantly considered and taken at the organizational level. 51 GenAI, alternatively, puts the power to make decisions about whether and how to use it more in the hands of the individual. GenAI has been described as a fundamental technological shift compared to predictive AI, with transformative potential, requiring additional insights into the realization of hybrid intelligence in this new technological era. 52 We contribute to extant literature by means of showing how individuals, with agency to decide how and when to use GenAI, use a process we have called reflexive augmentation, where they reflect, in an attentive and critical way, on which of their tasks should or should not involve GenAI and which tasks to automate or augment through GenAI. Our research demonstrates the continuing relevance of Schön’s ideas on reflective practice in the new technological era. 53 Indeed, considering the “jagged technological frontier,” 54 in which the capabilities of AI are constantly evolving, the relevance of reflexive augmentation seems ongoing and not only beneficial for those professionals who are relative novices in GenAI use.
Our research extends prior research by showing how human-technology collaboration is shaped by tensions resulting from the interactions between humans and the technology. One tension relates to individuals’ assessments on the real or perceived value of GenAI and confirms prior research on how humans draw ontological boundaries between themselves and machines. 55 Our informants drew a clear distinction between what they felt GenAI could do and what they felt they were uniquely placed to do in the context of BMI. The tensions we identified related to GenAI skills and trust in GenAI output—and associated behaviors—reinforce emerging research on the importance of human agency to optimize GenAI outcomes. 56 The fourth tension we identified concerns whether to be open about GenAI use. This tension has received limited attention in prior research, yet it is highly relevant given growing interest in AI ethics 57 and possible negative effects on a firm’s trustworthiness and reputation. 58
With the present research we also contribute further insight into GenAI use during BMI. There is conceptual research that points to the benefits that GenAI might bring for BMI. 59 However, empirical research on how humans entangle with GenAI when innovating business models is scarce. Our research helps to bridge this gap by identifying specific use cases for GenAI in the context of BMI. We observed individuals applying GenAI mainly to backstage, supporting BMI activities, but not necessarily to the most visible, strategic, and high-value BMI activities. We also reveal certain behind-the-scenes considerations that strategy consultants consider when deciding whether to disclose their use of GenAI in BMI and how such disclosure might influence their clients’ perception of their value. By walking this tightrope, strategy consultants sought to create value for themselves and for clients from their GenAI use without also diminishing their professional value in the longer term. Our findings suggest that GenAI is being used as a capable assistant who operates backstage rather than a fully fledged member of the team performing strategic, onstage BMI activities. While the informants in our sample engaged in automating and augmenting specific BMI tasks through GenAI, the technology is being used in a rather limited way when engaged in BMI, at least for now. GenAI was used primarily as a support for backstage activities, such as structuring and analyzing data or creating user journey maps. GenAI use did not feature prominently in what are considered core onstage BMI phases, such as identifying possible new strategic directions together with a client. The consultants’ GenAI use may have been driven by their desire to justify and protect their professional value to clients and, more implicitly, to their own organization.
Management Implications
Our research has implications for innovation practitioners and managers. For practitioners, we show what tensions can be expected when engaging with GenAI for innovation purposes. These tensions are summarized in Table 2. By acquiring advanced GenAI “driving” skills, and using reflexive augmentation to decide how to use GenAI at the task level, innovation practitioners can create a competitive advantage for themselves as employees in the future and may avoid being fully replaced by AI technology. 60
Managerial Implications.
For organizational leaders with responsibility for initiating and embedding GenAI capability in their organizations, our research offers valuable insights. Our research suggests that employees may limit GenAI use due to the tensions it may create or elicit. A limited or ineffective use of GenAI at the employee level may result in ineffective integration of technology at the firm level. It is important that organizational leaders address the tensions as experienced at the employee level and ultimately stimulate effective GenAI use at the firm level. Organizational leaders can do so by engaging in specific practices, which we have summarized in Table 2. In summary, managerial implications relate to continuous information sharing and learning on ever-evolving GenAI capabilities, further integration of reflexive augmentation at the individual and firm level, and protocols to maintain accuracy and AI ethics.
Overall, the insights garnered by our research help leaders to create appropriate strategies to encourage more encompassing GenAI use, supporting GenAI adoption via deep human understanding and helping to develop employees of the future.
Limitations and Future Research
Based on our qualitative research on professionals engaged in BMI, we identified reflexive augmentation as a micro-level process guiding GenAI integration. Considering its theoretical roots in Schön’s reflective practice, we assess reflexive augmentation might also be identified in other professional contexts. As illustrated by a recent example in which reflexive augmentation in public policy consulting appeared to be lacking, 61 future research is needed to examine further its presence or absence within different contexts. Future research is also called for considering that our research was based on a relatively limited number of professionals working for a specific multinational consulting company.
We identified four tensions related to GenAI use that shape reflexive augmentation. Based on our qualitative research approach we are unable to assess whether these four tensions were felt in an equally intense way by each informant. Future research may also be needed to further explore how the identified tensions shape reflexive augmentation over time and whether new types of tensions might emerge. Much remains to be explored in the fast-moving world of GenAI. For example, as the technology improves, new use cases might emerge, potentially shifting the role of GenAI to a more central one in the context of innovation, something which could be studied with a longitudinal research approach rather than a cross-sectional one as we adopted. More in-depth research on the links between individual and organizational dynamics will also be important to better untangle, for example, how reflexive augmentation at the individual level influences GenAI integration and, ultimately, outcomes at the organizational level.
Supplemental Material
sj-docx-1-cmr-10.1177_00081256261445458 – Supplemental material for The Human Side of Generative Artificial Intelligence in Business Model Innovation
Supplemental material, sj-docx-1-cmr-10.1177_00081256261445458 for The Human Side of Generative Artificial Intelligence in Business Model Innovation by Norman Azabagic, Gerda Gemser and Edward Giesen in California Management Review
Supplemental Material
sj-docx-2-cmr-10.1177_00081256261445458 – Supplemental material for The Human Side of Generative Artificial Intelligence in Business Model Innovation
Supplemental material, sj-docx-2-cmr-10.1177_00081256261445458 for The Human Side of Generative Artificial Intelligence in Business Model Innovation by Norman Azabagic, Gerda Gemser and Edward Giesen in California Management Review
Footnotes
Supplemental Material
Supplemental material for this article is available online.
Notes
Author Biographies
Norman Azabagic is a Senior Lecturer and Academic Lead for the Creative Destruction Lab program at Monash University in Melbourne, Australia (
Gerda Gemser is a Full Professor of Entrepreneurship & Innovation and Chair of Entrepreneurship at the University of Melbourne in Australia (
Edward Giesen is an Executive Partner in the IBM Enterprise Strategy Consulting Practice and is based in Amsterdam (
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
