You’re listening to “Enhancing corporate innovation through generative AI: A case study on Human–AI collaboration,” by T. Pena and colleagues. Published in 2026.

Contents lists available at ScienceDirect

Social Sciences & Humanities Open journal homepage: the linked source

Regular Article

T. Pena a, A.S. Cunha a,, J.V. Cordeiro b, G. Victorino a a NOVA Information Management School (NOVA IMS), Universidade NOVA de Lisboa, Lisbon, Portugal b NOVA National School of Public Health, Public Health Research Centre, Comprehensive Health Research Center, CHRC, NOVA University Lisbon, Lisbon, Portugal

1. Introduction.

Corporate innovation is fundamental for organizations to remain competitive, adapt to changing environments, and meet consumer demands. Innovation assists companies in identifying opportunities, reducing risks, and improving their products and services. However, different companies face challenges in effectively implementing and managing innovation projects. These challenges can relate to limitations in resources, organizational culture, leadership support, and cross-departmental collaboration. Furthermore, companies must balance exploitation, which consists in refining existing services to meet current consumer needs and exploration, which consists in anticipating future demands through risk-taking and experimentation. Achieving this balance requires specific knowledge and skills and remains a critical factor for organizational success in an increasingly competitive market.

Beyond the challenges of managing innovation, organizations must also address the need to cultivate creativity as the driving force behind innovative solutions (T. Amabile, 1988; Phelan & Young, 2003).

Creativity involves generating novel and practical ideas, yet its effective application depends on understanding and implementing a creative process. Encouraging creative confidence, defined as the belief in one's ability to produce meaningful and effective solutions, empowers individuals to contribute to innovation. At an organizational level, approaches such as Design Thinking (DT) help address these challenges by providing a human-centered, iterative methodology for solving complex problems. In this study, DT is conceptualized following Brown (2009) framework, which structures the process into three iterative phases: Inspiration, Ideation, and Implementation. This perspective emphasizes DT as a structured, human-centered methodology for problem-solving, distinguishing it from broader or less formalized approaches to creativity and innovation.

While philosophical perspectives, such as Dewey's, contribute to the broader understanding of experience and creativity, our study focuses specifically on DT as an applied organizational methodology.

1.1. Innovation through trend identification, brainstorming, and feedback.

in DT

This study follows Brown (2009) DT framework, which is structured into three phases: Inspiration, Ideation, and Implementation. These phases provide the conceptual foundation for organizing the analysis of the DT process in this study. DT supports both exploration and practical implementation, as it integrates the needs of users, technological possibilities, and business requirements. By facilitating divergent thinking and enabling flexibility in problem-solving, DT has been shown to aid organizations in achieving innovative outcomes (Brown (2009); Dewey, 1958). Within the three phases of DT - inspiration, ideation, and implementation - key components such as trend identification, brainstorming, and feedback play a pivotal role in driving effective innovation (Brown (2009)).

The selection of these three components—trend identification, brainstorming, and feedback—was guided by their relevance in capturing distinct and complementary stages of the DT process.

Trend identification represents exploratory activities in early innovation stages, enabling organizations to anticipate future consumer needs and reduce uncertainty, despite challenges related to subjectivity and limited predictive capacity.

Brainstorming reflects the generative phase of innovation, supporting divergent thinking and structured idea generation while helping organizations balance creative exploration with practical constraints (T. M. Amabile, 1996; Carlgren et al., 2016; Litchfield, 2008).

Feedback captures evaluative and refinement processes, integrating multiple perspectives to support both divergent and convergent thinking, while also addressing challenges related to bias and organizational dynamics.

Together, these components enable a comprehensive, process-oriented examination of how Generative AI supports different cognitive and collaborative activities within DT (Table 1).

During the inspiration phase, trend identification supports organizations in anticipating future consumer needs and reducing uncertainty in early innovation stages. However, this process remains challenged by subjectivity and limited predictive capacity, particularly in unstructured and uncertain contexts.

In the ideation phase, brainstorming facilitates divergent thinking and idea generation by encouraging the exploration of multiple solutions and challenging existing assumptions. When supported by clear objectives, brainstorming can help organizations balance creative exploration with structured innovation processes (T. M. Amabile, 1996; Robert C Litchfield 2008).

During the implementation phase, feedback supports both divergent and convergent thinking by refining ideas and incorporating multiple perspectives into decision-making. Nevertheless, the effectiveness of feedback in organizational settings may be constrained by hierarchical structures and communication barriers.

Overview of DT phases, components, aims and importance.

1.2. The role of GenAI in DT.

Corporate innovation has advanced through multidisciplinary studies, uncovering key facilitators, barriers, and outcomes for organizational competitiveness. However, challenges remain in bridging theory and practice to adapt quickly to evolving technological and market demands. Emerging technologies such as blockchain, machine learning, and artificial intelligence (AI) present both opportunities and challenges for innovation across industries. Generative AI (GenAI) in particular, which employs techniques such as transformer models like GPT, variational autoencoders, and generative adversarial networks, has the potential to enhance innovation processes, including DT.

GenAI use does not necessarily contradict core principles of DT such as empathy, creativity, and experience but might rather strengthen them by helping to overcome specific human-centered design obstacles. Specifically, GenAI can create new data sets based on patterns identified during training, which sopen opportunities for enhancing creative and iterative processes in DT. Despite its potential, the integration of GenAI raises significant ethical and practical concerns. Tools like ChatGPT, which gained widespread adoption in 2023, often produce plausible yet fabricated information, posing challenges around trust, transparency, and accountability. Users may struggle to understand how GenAI systems generate outputs, as reasoning processes are not fully interpretable.

Ethical considerations, including data privacy, social responsibility, and the trade-off between profit and moral judgment, must be addressed to ensure responsible GenAI use. Balancing the benefits and risks of GenAI integration is essential to ensuring that innovation processes, particularly in DT, remain meaningful, effective, and ethically sound.

In conclusion, the integration of GenAI into innovation processes presents both significant opportunities and notable challenges. GenAI can improve the DT methodology by addressing its limitations, particularly in the exploration phase, where it helps identify strengths and patterns that traditional methods might overlook. However, this integration also raises critical concerns, including ethical considerations and issues surrounding data usage and privacy. Ensuring the ethical implementation of AI is essential to maintaining the integrity and trustworthiness of the innovation process.

Research on the incorporation of AI into innovation processes has expanded significantly in recent years, particularly within the domains of computational creativity and human–AI collaboration. Existing studies have explored how AI can support idea generation, enhance creative performance, and augment decision-making processes. However, much of this literature has focused either on isolated creativity tasks or on outcome-based measures of innovation, offering limited insight into how human–AI collaboration unfolds within structured innovation methodologies.

In particular, the role of Generative AI within Design Thinking (DT) processes remains underexplored (B ̈ockle & Kouris, 2023a) (B ̈ockle & Kouris, 2023a).

While DT is widely recognized as a human-centered and iterative approach to problem-solving, there is still limited empirical understanding of how GenAI interacts with specific DT components across different phases of the process, especially in organizational contexts. While the potential of AI to enhance creativity, efficiency, and decision-making in innovation processes is widely recognized, its specific impact on the three phases of DT (inspiration, ideation, and implementation) has not been fully examined.

This study addresses this gap by adopting a process-oriented perspective and examining how GenAI influences key DT components within a real corporate setting. Specifically, we investigate how GenAI supports trend identification, brainstorming, and feedback activities across the DT process. To achieve this, we assess the perceptions of a diverse population within a company regarding the potential and challenges of GenAI use before and after a DT workshop.

Additionally, we compare the outcomes of wisdom-of-the-crowd votes with leadership votes, offering a unique perspective on collective decision-making versus leadership-driven innovation.

By focusing on these dimensions, this research contributes to a more nuanced understanding of human–AI collaboration in structured innovation environments and aims to advance knowledge on the integration of GenAI within DT processes in corporate settings. Ultimately, understanding how GenAI interacts with DT methodologies may help organizations foster more effective and contextually grounded innovation practices.

1.3. Research objectives and research questions.

The purpose of this study is to examine how Generative AI supports human-centered innovation within the DT process in a real corporate context. While prior research has addressed -AI-supported innovation and DT, the role of Gen AI in human-AI collaboration across specific DT components remains unexplored.

Consequently, this research explored the following research questions:

• How does Gen AI influence trend identification, brainstorming and feedback during a corporate DT workshop?

• How do participants’ perceptions of creative confidence and Gen AI usefulness change before and after the workshop?

• How do collective (“wisdom of the crowd”) and leadership evaluations differ across human-Gen AI configurations?

This study adopts an exploratory case-study approach to examine human–GenAI collaboration in DT activities, emphasizing group interaction, task execution, and peer and leadership evaluation.

2. Methods.

This study employed a mixed-methods approach, incorporating quantitative and qualitative analyses to evaluate the influence of GenAI in the DT process within a corporate innovation environment. Although the study was conducted in a real corporate environment, the workshop incorporated structured experimental elements. These included controlled group configurations, standardized tasks across conditions, and predefined use of Generative AI. This hybrid design enabled comparison across different human–AI interaction settings while preserving the ecological validity of a real organizational context. The selection of a mixed-methods approach was guided by the need to capture both measurable changes in participants’ perceptions and richer qualitative insights into interaction dynamics, enabling a more comprehensive understanding of human–GenAI collaboration within DT activities.

The overview of the methodological steps employed in this study, along with their respective descriptions, is presented in Table 2 below.

2.1. Study context and participant selection.

The company selected for this session specializes in outsourcing customer experience services, combining human expertise with AI-driven technologies such as conversational AI and CX analytics. Their services span sectors including banking, healthcare, and retail, with services such as back-office support, technical assistance, customer care, and sales retention. The company's experience with AI technologies provided a relevant context for evaluating the impact of GenAI on the DT process.

Participants were selected by the company through convenience sampling and represented multiple departments, allowing for the inclusion of diverse organizational roles and perspectives. Diversity is a core principle of DT as assit integrates multiple perspectives to promote innovation by improving the flexibility and inclusivity of solutions and ensuring they better address the needs of the target audience. Participants were informed about the context and objectives of the study and provided informed consent to participate. Research ethics principles and legally applicable requirements were fully complied with.

2.2. Data collection: Pre- and post-workshop questionnaires.

Participants’ prior experience with DT and GenAI was assessed through a dedicated section of the pre-workshop questionnaire. This assessment was conducted to provide contextual information for interpreting the study results and is described in detail in the section below.

To evaluate participants' perceptions of GenAI integration into the DT process, pre- and post-workshop questionnaires were administered for quantitative comparison. The questionnaires employed Likert scale-based questions (1 = strongly disagree; 7 = strongly agree) and were structured according to the DT phases (Inspiration, Ideation, and Implementation) to reflect the workshop activities (B ̈ockle & Kouris, 2023).

Questions addressed participants' self-reported creative confidence, openness to GenAI adoption, ethical concerns and GenAI usefulness, according to components of different DT phases: Inspiration Phase (trendspotting); Ideation Phase (idea generation and brainstorming); Implementation Phase (feedback process). Questionnaires began with a Primary Information section to assess participants' familiarity with DT methodology and their knowledge and frequency of GAI use. Finally, a Demographics section collected data on participants' age, gender, education level, and professional roles to contextualize responses and identify patterns across the group.

Questionnaires were pre-tested in a convenience sample of 3 volunteers. As a result of this test, changes were introduced to the clarity of definitions, expressions and concepts included in the questions.

Overview of the methodology.

The questionnaire was administered before and after the DT workshop. Participants accessed the questionnaire by scanning a QR code projected on a screen using their personal devices. This approach facilitated access, allowing participants to independently complete the pre-and post-workshop questionnaires at the designated times. The workshop included 15 participants; however, only 14 completed the post-test questionnaire, rendering the pre-test response of the non-responding participant invalid.

2.3. DT workshop procedure.

The workshop was conducted in a physical, in person setting and it was structured according to the three DT phases. To improve clarity and reproducibility, each workshop activity was structured as a clearly defined task with specific objectives and expected outputs. In the inspiration phase, participants were asked to identify emerging trends relevant to the company's future challenges, and to formulate a concise opportunity statement based on these trends. In the ideation phase, participants were instructed to do a brainstoming activity.

The brainstorming activity conducted in this study followed a structured approach based on classical brainstorming principles, including encouraging the generation of a high number of ideas, withholding judgment during idea generation, building on others’ ideas, and promoting unconventional thinking. While brainstorming can take multiple forms, this study adopted a group-based, time-constrained ideation format in which participants collaboratively generated ideas in response to a predefined task. This approach was selected to ensure consistency across groups and alignment with the DT ideation phase. In the implementation phase, participants were asked to evaluate the ideas generated by other groups by providing structured feedback, including one positive aspect and one suggestion for improvement.

All tasks were time-constrained and followed the same instructions across groups to ensure consistency and comparability. Data from these tasks informed the results presented in Section 3. Each activity lasted 30 min and concluded with the group submission of the output, followed by individual anonymous voting with their own devices. Participants were randomly and equally divided into three small groups. This assignment aimed to ensure an even distribution of participants across groups while maintaining diversity in terms of departmental background. Each group was assigned a distinct level of interaction with Generative AI, allowing for a structured comparison of different human–AI collaboration configurations.

Each group had a different specification to test the role of GenAI in the DT process: i) Group “Humans”: participants could only rely on their own intellectual capacities to complete the exercises.

ii) Group “GenAI-Enhanced”: participants were required to use GenAI, but only after first completing the exercises using their own intellectual capacities.

iii) Group “GenAI-Free Access”: participants had the flexibility to decide whether and how to use GenAI during the exercises.

A fourth group (“Non-Expert with GenAI”), consisting solely of a research team member using GenAI as the input source, also completed the exercises. The prompts replicated the instructions and descriptions presented to participants during the session. This fourth group was included as an AI only benchmark in order to make it possible to compare between AI only outputs and outputs generated under different human group configurations. The prompts and corresponding outputs for this group are provided in Table S1. ChatGPT (version 3.5) was the GenAI system used for this workshop (B ̈ockle & Kouris, 2023). For the “GenAI-Enhanced” and “GenAI-Free Access” conditions, participants interacted with the tool by submitting text-based prompts aligned with the task instructions provided. For the “Non-Expert with GenAI” condition, prompts replicated the same task descriptions given to participants.

Prompts used with the Generative AI tool were aligned with the written exercise instructions provided to participants for each exercise. The same task descriptions were used across groups to ensure consistency. The comparison between wisdom-of-the-crowd (WOC) evaluations and leadership evaluations was included to capture different perspectives in the assessment of innovation outcomes. While WOC reflects collective judgment and may favor creativity and diversity of ideas, leadership evaluation represents strategic and managerial perspectives, often emphasizing feasibility, alignment with organizational goals, and implementation potential. This comparison enables a more comprehensive understanding of how different evaluation logics influence the perceived value of outputs generated under different human–AI collaboration conditions.

After the completion of each exercise and the submission of each group's output, an individual vote was held. This individual voting process consists of in which participants evaluated and selected the best outcome among all group submissions, excluding their own. This approach leveraged the “wisdom of the crowd” (WOC) a concept suggesting that collective decisions made by a group are often more accurate or effective than those of individual experts. In each exercise, a leadership vote, provided by the company's CEO, was also included. A qualitative analysis aimed to compare decisions made by the leadership with those derive from the WOC. Fig. 1 summarizes the workflow of the DT workshop, including group assignment, sequence of activities, and the role of Generative AI across experimental conditions.

Figure 2 presents an overview of participants’ general familiarity with Design Thinking and Generative AI.

2.4. Data processing and statistical analysis.

Results were transferred to IBM® SPSS® software (version 28) for data processing and statistical analysis. Univariate analysis consisting in the descriptive statistics of the frequency analysis of each variable was performed. In this study, most variables demonstrated a normal distribution. Prior to hypothesis testing, data distributions were examined. Most variables approximated normality, and in cases where deviations were observed, they were not substantial. Given the exploratory nature of the study and the robustness of paired t-tests under these conditions, parametric testing was considered appropriate. Paired-sample t-tests were used to compare pre- and post-workshop responses, as the same participants completed both questionnaires and the objective was to assess within-subject changes in perceptions following the intervention.

This test was conducted to determine statistically significant differences between pre- and post-test means. The test assessed whether any observed differences were statistically significant. A positive correlation between variables was considered significant when the p-value was less than 0.05, corresponding to a 95% confidence level. Table 3 summarizes the overall study design, mapping research objectives to DT phases, data sources and analysis methods.

3. Results.

The demographic and professional characteristics of the participants are summarized in Table 4. The participants were evenly distributed by gender (1:1 ratio) and had a median age of 46 years. The majority held higher education degrees (12 out of 14) and originated from different company departments.

3.1. Baseline perceptions of DT and familiarity with generative AI.

As our study focused on the role of GAI in the DT process, we set out to assess participants' self-perception of DT and their familiarity with GAI. Participants were asked to estimate their literacy levels in DT and GAI using a 7-point Likert scale (Fig. 1). The majority of respondents (57.1%) reported having a good level of familiarity with DT (level 5) and perceived DT as well suited for addressing challenges in corporate innovation (92.9%, levels 6 and 7 combined). Regarding GAI, participants demonstrated good familiarity, with 71.5% rating themselves at levels 5, 6, or 7, and 50% reporting frequent use (levels 5, 6, and 7 combined).

3.2. Perceived contribution of generative AI to DT and innovation.

Participants' general perceptions regarding the contribution of GAI to DT are presented in Fig. 3. Overall, 78.5% of participants agreed that GAI can help identify user needs more effectively than traditional DT methods (levels 5, 6, and 7 combined). Furthermore, all respondents agreed that GAI can enhance the innovation process in general (levels 5, 6 and 7 combined). Consistently, 78,6% of participants disagreed (levels 1, 2 and 3 combined) with the statement that GAI provides less useful feedback than humans. However, 50,0% of participants agreed (levels 5, 6 and 7 combined) that GAI could potentially introduce bias into the solutions generated during the DT process, while 28,5% disagreed (levels 1, 2 and 3 combined), and 21,4% expressed a neutral opinion (level 4).

Specific perceptions of DT and GAI according to DT phases are presented in Fig. 4. Regarding the inspiration phase, 71.4% of participants expressed confidence in their ability to identify emerging trends (levels 5, 6, and 7 combined). Notably, all participants were open to incorporating AI-driven trend analysis in the trend-spotting process. However, 49,9% of participants expressed concern about the ethical implications of AI use in trend analysis (levels 5, 6, and 7 combined), while 42,9% were neutral (level 4). In terms of AI's effectiveness, 92,9% agreed that AI could improve trend analysis in the DT process.

Regarding the ideation phase, 57.1% expressed confidence in their ability to generate creative ideas (levels 5, 6, and 7 combined). Openness to integrating AI in the ideation process was 100% (levels 5, 6, and 7 combined). Ethical concerns remained relevant, with 50% expressing concern about the ethical implications of using AI in idea generation (levels 5, 6, and 7 combined), while 42,9% were neutral (level 4). Nevertheless, all participants agreed that generative AI could enhance the efficiency of brainstorming sessions (levels 5, 6, and 7 combined) with 50% indicating extreme agreement (level 7).

Regarding the implementation phase, participants demonstrated confidence in their ability to provide constructive feedback, with 78.5% agreeing (levels 5, 6, and 7 combined). Openness to incorporating AI-driven feedback was also high, with 85.8% expressing agreement.

Summary of study design, data sources and analysis methods.

Table 4

Participants profiles and demographic information.

Ethical concerns about using AI for feedback were less significant than for other DT phases, as 35.7% expressed concern (levels 5, 6, and 7 combined), while 42.9% remained neutral (level 4). Finally, participants widely agreed (85.7%) that AI could enhance the process of giving and receiving feedback.

Overall, the results indicate strong confidence and openness toward the integration of AI across all phases of the DT process, despite some ethical concerns.

3.3. Workshop results: generative AI across DT phases.

To evaluate the contribution of GAI to the DT process, we conducted an experimental DT session. This session included three key exercises corresponding to specific DT phases: identifying trends (inspiration phase), brainstorming (ideation phase), and providing constructive feedback (implementation phase). Participants were divided into three groups (humans only, GAI-enhanced, and GAI-free access) to perform the exercises in parallel. A non-expert member of the research team also utilized GAI, providing results for comparison.

3.3.1. Inspiration phase: Trend analysis.

In this phase, participants were asked to identify trends that would impact the company's future challenges related to customers' needs, ultimately leading to spaces of opportunity for value creation. Following a group discussion, each team submitted a proposed opportunity space in the form of a brief and concise sentence (Table 3). Participants then anonymously voted for the best outcome, excluding their own group's submission, to reflect the “wisdom of the crowd” (WOC). Leadership also voted for the best outcome. The GAI-free access group received the highest number of WOC votes (33.3%), while the leadership selected the outcome generated by the GAI-enhanced group (Table 3).

3.3.2. Ideation phase: Brainstorming.

In this phase, participants remained in their assigned groups and were tasked with generating and brainstorming ideas involving technology that addressed the opportunity space identified in the previous exercise. Each group selected the idea they believed had the greatest potential for implementation and value creation, submitting it as a brief sentence. The analysis focuses on the final selected ideas generated by each group, as these represent the outcome of the collaborative ideation and decision-making process within each condition. While multiple ideas were generated during the brainstorming phase, the study emphasizes the selected outputs to enable consistent comparison across groups and evaluation conditions. As in the inspiration phase, both a WOC vote, and a leadership vote were conducted.

In this phase, the idea generated by the non-expert using GAI received the highest WOC vote (40%), while the leadership selected the idea generated by the GAI-free access group (Table 5). These results suggest that GAI-supported approaches demonstrated strong potential in identifying impactful trends, as evidenced by the WOC and leadership votes. These results indicate that GAI can play a valuable role in ideation.

3.3.3. Implementation phase: Providing constructive feedback.

In this phase, participants remained in their assigned groups, and each group presented their final idea in a 3-min pitch. The other groups were requested to anonymously provide constructive feedback in the form of a brief paragraph, consisting of one positive aspect and one suggestion for improvement. Each group then selected the feedback they found most useful for both their own idea and the ideas of other groups, excluding their own submission. Leadership also voted anonymous on the feedback provided by all groups. Additionally, the non-expert using GAI contributed anonymous feedback, which was included for

Results of the Inspiration and Ideation Phases of the GAI-DT workshop.

Inspiration (trend analysis) and Ideation (brainstorming) outcomes according to group composition and respective wisdom of the crowds (WOC) (n = 15) and leadership (n = 1) voting results.

comparison.

Regarding the feedback provided to the humans group's idea, the feedback submitted by the AI-free access group won the WOC vote, while the feedback provided by the GAI-enhanced group was selected by the leadership (Table 6). For the idea generated by the GAI-enhanced

Results of the Implementation Phase of the GAI-DT workshop.

group, the feedback provided by humans won the WOC vote, whereas the leadership selected the feedback submitted by the non-expert using GAI. Lastly, for the idea generated by the AI-free access group, the WOC vote resulted in a tie among the feedback provided by different groups, while the leadership selected the feedback provided by humans

Implementation outcomes (feedback given) according to group composition and respective wisdom of the crowds (WOC) (black dotted box) and leadership (blue shade) voting results. N/A - not applicable.

(Table 4). These results highlight the importance of human judgment in providing actionable and constructive feedback, while GAI shows significant collaborative promise for this DT phase as well.

Taken together, these results demonstrate that GAI can effectively complement human creativity and judgment throughout the DT process, while human contributions remain critical, particularly in collaborative and evaluative stages where collective and expert perspectives play a key role.

3.3.4. Pre and post workshop perceptions of generative AI in DT.

We next aimed to evaluate the impact of the DT-GenAI workshop on participants’ perceptions of both their self-potential and the potential of GenAI within the DT process.

Regarding the inspiration phase, participants reported a significant increase in confidence in identifying emerging trends after the workshop (71.4% pre-workshop vs. 100% post-workshop at levels 6 and 7 combined) (Fig. 4). In the ideation phase, the workshop resulted in a statistically significant improvement in participants' confidence in generating creative ideas (35.7% pre-workshop vs. 64.1% post-workshop at levels 6 and 7 combined). Regarding the implementation phase, participants demonstrated more openness to incorporating AI- driven constructive feedback following the workshop, indicating a positive shift in their attitudes toward AI-supported feedback mechanisms (26.6% pre-workshop vs. 57.1% post-workshop at level 7).

Additionally, perceptions regarding GenAI's ability to identify user needs more effectively than traditional methods also improved significantly (21.4% pre-workshop vs. 85.7% post-workshop at levels 6 and 7 combined). To highlight the most relevant changes in participants' perceptions, Fig. 5 presents only the questionnaire items for which statistically significant differences were observed between pre- and post-workshop responses (p < 0.05).

Taken together, our results suggest that the DT-GenAI workshop positively influenced participants' perceptions of their self-potential and GenAI's potential to improve critical elements of the DT process, including trend analysis, idea generation, and constructive feedback incorporation.

To consolidate the empirical findings across the three DT phases examined in this study, Table 7 provides an integrative overview of the roles assumed by GenAI and human participants during trend identification, brainstorming and constructive feedback activities. The table aims to synthesize the core DT activities, the role of GenAI, the role of human participants and the corresponding empirical evidence derived

Summary of human and Generative AI contributions across DT phases.

from both quantitative and qualitative results. This comparative representation can help clarify how GenAI was mainly associated with exploratory and generative support functions, while human participants remained primarily involved in contextual interpretation, feasibility considerations and evaluate judgment. By bringing together participant perceptions, wisdom of the crowd voting patterns and leadership evaluations, this table emphasizes a cross-phase view of human-GenAI collaboration within the DT process and summarizes the key results that inform the subsequent discussion. The insights presented in Table 7 are derived from a synthesis of both quantitative results (including survey responses and voting outcomes) and qualitative observations collected during the workshop, enabling the identification of patterns across the different DT phases.

4. Discussion.

Our study aimed to explore how GenAI influences key components of the innovation process, including trend identification, brainstorming, and feedback, by assessing the perceptions of a diverse population within a company before and after a DT workshop. A key strength of this study lies on incorporating Generative AI into DT exercises carried out in a real corporate innovation workshop, rather than in a purely laboratory-based setting, although structured experimental elements were incorporated within the real organizational context. In addition, the use of multiple experimental conditions represents and added advantage, raging from human-only to AI-assisted and AI-dominant configurations, by enabling detailed insights into diverse forms of human-in-the-loop collaboration.

This hybrid setup reflects a balance between ecological validity and experimental control, allowing the study to capture realistic organizational dynamics while maintaining comparability across conditions.

Overall, our findings suggest that GenAI use can significantly improve the perceived effectiveness of the inspiration, ideation and implementation stages of the DT process. Both qualitative and quantitative data indicate that GenAI assistance empowers participants, boosting their self-confidence and supporting higher rates of innovation and idea generation. This observation is coherent with prior research showing that AI can improve creative processes by reducing cognitive barriers and inspiring divergent thinking. However, these findings should be interpreted with caution. The observed increases in participants’ confidence and perceived usefulness of Generative AI may be influenced by their limited prior exposure to GenAI in DT contexts.

As such, part of the observed effect may reflect a learning or novelty effect rather than a sustained improvement in creative capability or performance.

Similarly, improvements in self-reported confidence in activities such as identifying emerging trends may be partially attributed to participants’ limited prior experience with structured Design Thinking tasks. Therefore, these results should be understood as indicative of short-term perceptual changes rather than definitive evidence of enhanced creative performance.

The mixed-methods methodology of this study constitutes an additional strength, as it facilitates the integration of perception-based metrics with qualitative analysis of workshop outputs and evaluation patterns, making it possible for a more comprehensive understanding of human-GenAI collaboration.

Participants exhibited strong confidence and openness toward integrating GenAI across all phases of the DT process, despite expressing some ethical concerns. These results emphasize GenAI's valuable role in specific components such as trend identification, brainstorming, and feedback incorporation, indicating its potential to complement human creativity throughout the DT process.

Despite improving perceptions of the potential of GenAI across all phases of the DT process, participants' views on ethical considerations related to GenAI remained largely unchanged after the workshop. This stability may be related to their initial positive perception of DT and only average familiarity with GenAI, suggesting that familiarity with innovation processes and AI technologies may contribute to stabilizing attitudes toward their integration.

The implementation phase findings of our study emphasize the role of GenAI in supporting the DT process by supporting human creativity and judgment. Nevertheless, human contributions are indispensable, particularly in evaluative stages where expert insights and group perspectives, such as providing actionable and constructive feedback are vital. This aligns with previous research, which highlights that human feedback provides critical contextual insights that AI alone cannot offer. While GenAI shows strong promise as a collaborative tool, human contributions remain critical for ensuring balanced, effective innovation.

The integration of GenAI with human expertise has been reported to result in improved outcomes across various fields. For example, a study at Harvard University's Graduate School of Design found that combining human designers' contextual and aesthetic insights with AI's computational efficiency produced the most innovative and practical architectural designs. This is in line with our results, which emphasize the importance of combining human creativity and judgment with AI's analytical capabilities to improve perceived outcomes in the DT process. Similarly, studies have explored the role of GenAI in journalism and creative writing, revealing that co-authored articles combining human creativity with AI's efficiency produced content that was both engaging and structurally robust.

This is coherent with the potential we observed in the ideation phase of our DT workshop, where GenAI improved idea generation, while human contributions added context, ensuring an effective collaborative process. In parallel, a study reporting the collaboration between professional musicians and GenAI demonstrated that AI's ability to generate intricate melodies was improved by human input, which infused compositions with emotional and expressive qualities. This finding is also coherent with our results, particularly in the feedback stage, where the combination of human insights and AI capabilities led to refined outputs, but the human factor was assessed as fundamental. Also in healthcare, AI diagnostic tools were combined with human expertise to achieve higher accuracy and reliability in medical imaging analysis.

In line with this, our results emphasize that while GenAI improves creative confidence and operational efficiency, human judgment remains critical in evaluative phases to ensure contextually appropriate and ethically sound decisions. In the law field, complex language interpretation and understanding have also been shown to be activities where human involvement is essential, despite AI excellence in speed and accuracy for detecting technical flaws. This is coherent with our observation that GenAI can support efficiency and precision, but human expertise is indispensable for nuanced tasks requiring deep contextual understanding.

At the same time, it is important to situate these findings within the broader innovation literature. A growing body of research examines AI, digital transformation, and technological innovation primarily through macro-level or firm-level indicators, such as patent outputs, innovation efficiency measures, policy shocks or econometric estimations based on large-scale panel data and machine learning techniques. These studies provide valuable insights into how digital technologies, environmental information disclosure, media attention and policy frameworks may influence innovation performance at an aggregate or systemic level.

However, they typically conceptualize innovation as an outcome variable and offer limited visibility into how innovation processes unfold within situated organizational practices, or how human actors experience and make sense of collaboration with AI systems during creative and evaluative activities. In contrast, the present study adopts a micro, process-oriented perspective, focusing on participants’ perceptions, sensemaking, and evaluative judgments within a DT workshop supported by Generative AI. Rather than seeking to replicate outcome-oriented measures, this study aims to complement the existing literature by providing process-level and experiential insights that remain underexplored in macro-quantitative studies of digital or green innovation.

In this sense, the study also complements existing research on Generative AI in ideation and creativity within the Human–Computer Interaction field, by focusing on process-level dynamics and human–AI collaboration within structured Design Thinking activities in an organizational context.

In summary, research studies collectively reinforce the idea that human-AI collaboration highlights their respective strengths in different fields including architecture, journalism, music, healthcare, and law. Our findings similarly suggest the importance of integrating GenAI into the DT process, demonstrating its ability to improve creativity, efficiency, and decision-making when balanced with human judgment and insight. The findings related to participants' increased confidence and perceived usefulness of GenAI reflect self-reported perceptions rather than objectively measured performance improvements. These perception-based results are therefore interpreted as indicative of participants' experiential evaluation of GenAI support during the workshop. These interpretations should also be considered in light of the potential for response bias.

In particular, participants' increased confidence and positive perceptions may partly reflect their initial exposure to Generative AI within a structured workshop setting, rather than long-term changes in behavior or capability. The observed increase in participants’ confidence during ideation and feedback phases is supported by the statistically significant pre–post differences reported in Fig. 5.

Our study also aimed to investigate the outcomes of WOC votes and leadership decisions, offering insights into collective versus leadership-driven innovation preferences. Across all DT phases, despite not being coincidental, leadership and WOC votes often reflected the valorization of human contributions or hybrid approaches. These results underline the importance of integrating collective opinions with leadership expertise.. Differences between leadership voting and wisdom-of-the-crowd preferences suggest distinct evaluation logics. Leadership evaluations appeared more closely aligned with strategic feasibility and organizational relevance, whereas collective voting tended to favor creative novelty and exploratory ideas.

This distinction helps explain observed divergences in preferred outputs and highlights the complementary roles of managerial judgment and collective assessment in DT workshops supported by Generative AI.

Our study suggests that GenAI can potentially complement human creativity and judgment throughout the DT process. While human contributions remain critical, particularly in collaborative and evaluative stages, the integration of GenAI appeared to facilitate a more inclusive and balanced approach to corporate innovation. From a practical perspective, the findings offer insights that may be relevant for managers and innovation practitioners seeking to experiment with GenAI within structured DT processes. Future research should explore the long-term implications of GenAI integration in diverse organizational settings to further understand its impact on innovation processes.

4.1. Limitations of the study.

This study presents some limitations that should be considered when interpreting the findings. One limitation of this study is the small sample size inherent to this single-case exploratory study limits statistical power and constrains the generalizability of the findings. As such, the statistical results should be interpreted as indicative rather than confirmatory. In particular, comparisons between groups should be interpreted with caution, as the limited sample size reduces the robustness of between-group statistical inference. These comparisons are therefore intended to provide exploratory insights rather than definitive conclusions. Given the exploratory nature of this single-case study and the small sample size, the statistical analysis focused on significance testing.

Effect size measures were not included and should be considered in future studies with larger samples to better assess the practical magnitude of observed effects.

In addition, the study relies partly on self-reported perceptions measures collected through pre and post workshop questionnaires. Such measures may be subject to response bias, as participants’ improved confidence and openness toward GenAI may represent their early exposure to technology rather than long-term improvements in creative ability or performance. Furthermore, the study relies partly on self-reported perceptions measures collected through pre and post workshop questionnaires. Such measures may be subject to response bias, as participants’ improved confidence and openness toward GenAI may represent their early exposure to technology rather than long-term improvements in creative ability or performance.

Additionally, these perception-based measures may partly reflect short-term exposure effects, particularly given participants’ limited prior experience with Generative AI and structured DT activities.

Furthermore, the DT workshop's brief duration makes it difficult to evaluate longer term innovation results or the long-term benefits of human-GenAI collaboration. The findings therefore capture participants’ immediate perceptions rather than longitudinal impacts.

Finally, the study used a specific Generative AI tool and model version within a defined usage context and interface. Differences across AI systems or interaction modalities may lead to different outcomes, which limits the transferability of the findings to other Generative AI technologies.

4.2. Future research directions.

Building on the findings and limitations of this exploratory case study, several directions for future research may be considered. First, future studies could seek to explore the phenomena examined in this research across multiple organizational contexts, industries, and innovation settings. Extending empirical investigations beyond a single corporate case may help to assess whether similar patterns of human–GenAI collaboration emerge in different organizational environments. Second, future research might benefit from adopting longitudinal research designs to examine the potential longer-term effects of integrating Generative AI into DT activities.

While the present study captures participants’ immediate perceptions following a short-term workshop, longitudinal approaches could allow researchers to explore whether observed changes in creative confidence, openness to GenAI, and perceived usefulness persist over time and how they may relate to evolving innovation practices and organizational outcomes.

Third, subsequent studies may further explore the distinction between perception-based outcomes and more systematically assessed performance measures. Combining self-reported perceptions with additional evaluative approaches may provide a more nuanced understanding of GenAI's role across the inspiration, ideation, and implementation phases. In this context, future studies with larger samples could complement significance testing with the reporting of effect size measures, which may help to better estimate the practical magnitude of observed changes while enhancing statistical robustness.

In addition, future studies should explicitly assess participants’ prior experience with both Generative AI and DT activities. These factors may meaningfully influence user perceptions, learning dynamics, and the development of creative confidence, and should therefore be more systematically accounted for in future research designs.

Furthermore, future research could examine variations in human–GenAI interaction modalities, including different levels of AI autonomy, alternative prompt strategies, and the use of diverse Generative AI models. Given that the present study relied on a specific tool, model version, and interaction configuration, exploring alternative technological setups may help to clarify how different forms of human–AI collaboration influence creative processes and evaluative dynamics.

Finally, future studies may continue to deepen the examination of ethical considerations associated with the use of Generative AI in DT. While ethical concerns related to bias, transparency, and accountability appeared relatively stable in the present study, further research could investigate how ethical awareness evolves with sustained use and how organizational guidelines and safeguards might be more explicitly embedded within DT practices supported by Generative AI.

In addition, future research could further benefit from incorporating qualitative methods, such as semi-structured interviews or focus group discussions, to capture more in-depth insights into participants’ experiences, perceptions and sensemaking processes during human–AI collaboration in DT activities. While the present study relied primarily on structured workshop tasks and perception-based measures, qualitative approaches could provide richer contextual understanding of how participants interact with Generative AI, how ideas evolve during collaborative processes and how individuals interpret and evaluate AI-supported outputs.

Overall, these future research directions may contribute to a more cautious and context-sensitive understanding of how Generative AI can be integrated into human-centered innovation processes.

5. Conclusions.

This study explores the opportunities and challenges associated with the use of GenAI within the DT process, drawing on an exploratory case study conducted in a corporate innovation workshop. Conducting the study within an operational corporate environment, rather than a controlled one, strengthens the validity of the findings and enhances their relevance for real-world innovation practices. By examining the role of GenAI in human-centered design activities, the study provides exploratory insights into how organizations may experiment with GenAI in DT settings. Understanding how GenAI interacts with DT may help organizations reflect on the potential and limitations of integrating generative technologies into innovation practices.

Our findings contribute to the scientific understanding of the role of GenAI in corporate innovation by identifying general optimism about AI's potential, leading to expected predisposition toward GenAI adoption. The persistence of ethical concerns also highlights the need for further exploration of challenges and allies to its effective use. Future policy must prioritize maximizing GenAI's potential for innovation while addressing ethical challenges and incorporating the perspectives of professionals who will be tasked with driving its implementation.

This study has important implications for research, policy, and practice, offering relevant insights for academia, industry, and government. Our exploratory case study aims to inform a diverse audience, including scholars, researchers, professional designers, educators, students, and leaders in business and government. Rather than providing definitive conclusions, these insights offer a basis for future research to further examine human–GenAI collaboration in organizational innovation contexts.

Declarations.

CRediT authorship contribution statement

T. Pena: Writing – review & editing, Writing – original draft, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. A.S. Cunha: Writing – review & editing, Writing – original draft. J.V. Cordeiro: Writing – review & editing, Writing – original draft, Validation, Supervision, Methodology, Formal analysis, Conceptualization. G. Victorino: Writing – review & editing, Writing – original draft, Validation, Supervision, Methodology, Investigation, Funding acquisition, Formal analysis, Conceptualization.

Data availability statement

The data used in this study are available upon request. Researchers interested in accessing the data can contact the first author (TP) for further information (the email address).

Ethical statement

The study was conducted in accordance with the ethical principles of the Declaration of Helsinki. Ethical approval was obtained from the competent Institutional Review Board (NOVA IMS Ethics Committee – OTHER 2024-6-122809). All participants provided informed and explicit consent prior to their involvement. To ensure privacy and confidentiality, all data were fully anonymized.

Declaration of generative AI and AI-assisted technologies in the writing process

During the preparation of this work the authors used a Large Language Model (ChatGPT-4o) to assist with grammar and language editing of the manuscript. The tool was used under author supervision and did not contribute to the generation of content. The authors declare that the manuscript content is entirely original and take full responsibility for the integrity and accuracy of the final text.

Funding statement

Declaration of competing interest

None to declare.

Acknowledgments.

This work was supported by national funds through FCT (Fundaç ̃ao para a Ciˆencia e a Tecnologia), under the project - UID/04152/2025 -Centro de Investigaç ̃ao em Gest ̃ao de Informaç ̃ao (MagIC)/NOVA IMS -the linked source (2025-01-01/2028-12-31) and UID/PRR/04152/2025 the linked source

Appendix A. Supplementary data

Supplementary data to this article can be found online at the linked source.