Table 1 Data collection and data... | Emerald Publishing

Table 1

Data collection and data analysis step

Phase	Activity	Outcome
Questionnaire development	Developed a semi-structured questionnaire grounded in a comprehensive review of the literature on decision-making, human autonomy and AI use in managerial contexts. The questionnaire was designed to capture subjective, experience-based insights from senior professionals	A final instrument composed of two sections: (1) socio-demographic information and (2) ten open-ended questions addressing human–AI autonomy dynamics
Pilot test and prevalidation	Conducted a pretest with a convenience sample to refine the questionnaire. Two academic experts in decision-making and two senior professionals with practical AI experience participated. Their feedback informed iterative revisions to ensure methodological clarity and validity	Finalized and validated version of the questionnaire, refined for conceptual clarity, relevance and ease of response
Sample identification and data collection	Identified large enterprises via the AIDA database based on European Commission and Italian criteria. Sent invitations to 1,256 top managers across sectors. Data were collected through self-administered interviews distributed via e-mail across three mailing waves	122 valid interviews collected and screened, yielding a rich data set for qualitative analysis. Nonresponse bias was tested and ruled out through statistical comparison and follow-up calls
Data analysis (interviews)	We applied the Gioia Methodology (Gioia et al., 2013; Magnani and Gioia, 2023) to inductively code and analyze 1,262 meaning-rich responses. First-order concepts were derived from participants’ terms, then aggregated into second-order themes and theoretical dimensions. NVivo was used to support coding, organization and transparency throughout the process	67 codes and 23 latent themes identified, organized across four recursive phases of decision-making (Frame, Evaluate, Commit, Enact) and aligned with the DIKW model
Focus group triangulation and theory building	Held two follow-ups focus groups to validate and enrich findings: one with five academic experts in AI and decision-making, and one with five senior managers from diverse industries (e.g. telco, IT, FMCG, manufacturing, services). findings were compared with interview-based themes to assess coherence, divergence and conceptual robustness	Data triangulation confirmed the four-phase model of the human–AI autonomy loop (HAIL) and supported theory refinement by integrating practitioner and academic validation

Phase	Activity	Outcome
Questionnaire development	Developed a semi-structured questionnaire grounded in a comprehensive review of the literature on decision-making, human autonomy and AI use in managerial contexts. The questionnaire was designed to capture subjective, experience-based insights from senior professionals	A final instrument composed of two sections: (1) socio-demographic information and (2) ten open-ended questions addressing human–AI autonomy dynamics
Pilot test and prevalidation	Conducted a pretest with a convenience sample to refine the questionnaire. Two academic experts in decision-making and two senior professionals with practical AI experience participated. Their feedback informed iterative revisions to ensure methodological clarity and validity	Finalized and validated version of the questionnaire, refined for conceptual clarity, relevance and ease of response
Sample identification and data collection	Identified large enterprises via the AIDA database based on European Commission and Italian criteria. Sent invitations to 1,256 top managers across sectors. Data were collected through self-administered interviews distributed via e-mail across three mailing waves	122 valid interviews collected and screened, yielding a rich data set for qualitative analysis. Nonresponse bias was tested and ruled out through statistical comparison and follow-up calls
Data analysis (interviews)	We applied the Gioia Methodology (Gioia et al., 2013; Magnani and Gioia, 2023) to inductively code and analyze 1,262 meaning-rich responses. First-order concepts were derived from participants’ terms, then aggregated into second-order themes and theoretical dimensions. NVivo was used to support coding, organization and transparency throughout the process	67 codes and 23 latent themes identified, organized across four recursive phases of decision-making (Frame, Evaluate, Commit, Enact) and aligned with the DIKW model
Focus group triangulation and theory building	Held two follow-ups focus groups to validate and enrich findings: one with five academic experts in AI and decision-making, and one with five senior managers from diverse industries (e.g. telco, IT, FMCG, manufacturing, services). findings were compared with interview-based themes to assess coherence, divergence and conceptual robustness	Data triangulation confirmed the four-phase model of the human–AI autonomy loop (HAIL) and supported theory refinement by integrating practitioner and academic validation

Source(s): Authors’ own work