Bridging the GenAI Execution Gap : Barriers and Solutions for Scaling Generative AI in the IT Services Industry

Sammanfattning

Background. Despite substantial strategic investment and widespread experimentation, a significant proportion of Generative AI (GenAI) initiatives within the IT (Information Technology) services industry remain constrained at the pilot stage, unable to transition into sustained, enterprise-scale, and governed operational deployment. This persistent divergence between experimental success and production execution, referred to throughout this thesis as the GenAI execution gap, represents a critical strategic challenge for IT service organisations whose competitive differentiation, client trust, and contractual accountability depend on responsible AI (Artificial Intelligence) deployment at scale. Purpose. This thesis investigates the barriers that prevent IT service organisations from scaling GenAI beyond pilots and the solutions through which organisations may bridge the resulting execution gap. Using the Technology-Organisation-Environment (TOE) framework as the theoretical lens, the study produces an expert-validated, empirically grounded hierarchy of barriers and solutions specific to the IT services context and identifies cross-dimensional interaction effects that conventional single-method approaches cannot surface. Method. The study employs a sequential four-phase mixed-methods design. Phase 1 comprised a survey questionnaire administered to 20 IT services professionals, measuring the perceived severity of barriers and effectiveness of solutions across seventeen TOE-derived constructs spanning 122 items. Phase 2 applied the Fuzzy Delphi Method (FDM) to these 20 survey responses using Triangular Fuzzy Numbers to validate and prioritise the barrier-solution taxonomy through expert consensus analysis. Phase 3 comprised 6 semi-structured individual expert interviews, selected from the 20 survey/FDM respondents, each of approximately 45 to 60 minutes in duration, conducted to explore the contextual mechanisms and organisational dynamics behind the quantitative findings. Phase 4 applied Braun and Clarke's (2006) thematic analysis framework with a hybrid deductive-inductive coding approach using the TOE framework as the initial theoretical coding template, synthesising the qualitative findings across eleven themes. Results and Analysis. The FDM phase validated eleven consensus-supported barriers and forty-four consensus-supported solutions across the three TOE dimensions. The most critical barriers were concentrated in the Technology dimension: fragmented data governance across client environments (B2.2, crisp = 0.681) ranked first, followed by a cluster of integration complexity and technical debt items (B3.1 to B3.4), which achieved 100 per cent consensus coverage, the strongest consensus signal in the dataset. Organisational barriers, notably change resistance, low AI readiness, and prompt evaluation capability deficits, ranked below Technology in item-level consensus, yet expert holistic weight allocation placed the Organisation dimension as the leading contributor to the execution gap at 38.0 per cent, revealing a systemic rather than item-specific character. The Environment dimension produced no retained barriers due to jurisdictional variability yet yielded the highest mean retained solution effectiveness (0.710), confirming that environmental solutions are universally valuable despite contextual divergence in environmental barrier experience. Thematic analysis enriched and extended these findings, identifying three significant constructs not represented in the structured FDM taxonomy: a risk-tiered AI governance architecture principle; the client-side readiness bottleneck as an external scaling ceiling; and an incentive misalignment dynamic, termed the frozen middle, rooted in the billable-hour commercial model of IT service delivery. Cross-dimensional interaction analysis revealed that security and compliance gatekeeping processes systematically amplify technical integration barriers, and that iterative scaling frameworks produce sustained value only when anchored by explicit governance decision gates. Conclusions. The GenAI execution gap in IT services is a multi-dimensional, socio-technical, and ecosystem-level challenge. It cannot be resolved through technical intervention alone; it requires simultaneous action on data and integration architecture, organisational capability and incentive structures, and external client and vendor relationship management. The findings challenge single-dimensional explanations of the execution gap and support a structured, cross-dimensional model of barrier co-production and solution co-design. The commercial model of IT service delivery itself emerges as a structural barrier that conventional technology adoption frameworks do not capture. Recommendations for Future Research. Future research should pursue longitudinal study designs to track how barrier severity evolves as organisations progress from pilot to production stages. Cross-industry comparative studies would assess the degree to which the identified barrier-solution hierarchy is IT services-specific or broadly generalisable. Governance maturity research should examine how organisations progress through AI governance development stages over time. Post-deployment operational studies should investigate LLMOps (Large Language Model Operations), GenAI operational resilience, and value attribution following production launch. Research into commercial model transformation, from headcount-based to value-based or outcome-based IT service pricing, is identified as a priority gap warranting dedicated empirical investigation.