Research
Bibliography
Bibliography.
The sources cited across this site, grouped by the topic they inform. Quality bar: peer-reviewed journals, major research institutions, books from established imprints, and primary reporting from publications of record.
Each citation links to its primary source where available. The framework itself is grounded in 200+ first-person interviews with CHROs and CEOs conducted by Robert Blaga across two decades of leadership development practice; that primary research is the foundation. The published literature summarised here is the supporting record.
01
Enterprise AI adoption rates and stalls
The headline numbers from the major institutional research programmes — McKinsey, BCG, Deloitte, Gartner, RAND, and the U.S. Census Bureau — on what is actually being adopted, at what scale, with what success.
McKinsey & Company. (2025). The state of AI: How organizations are rewiring to capture value. McKinsey Global Survey.
78% of organizations use AI in at least one business function, but only 1% of executives describe their gen-AI rollouts as 'mature.'
Source ↗McKinsey & Company. (2024). The state of AI in early 2024: Gen AI adoption spikes and starts to generate value. McKinsey Global Survey.
65% of organizations report regular use of generative AI, nearly double the prior year, yet only 11% have adopted it at scale across functions.
Source ↗Gartner. (2024, July 29). Gartner predicts 30% of generative AI projects will be abandoned after proof of concept by end of 2025 [Press release].
Gartner forecasts at least 30% of generative AI projects will be abandoned after proof of concept by end of 2025, citing poor data quality, inadequate risk controls, and unclear business value.
Source ↗Deloitte. (2024). The state of generative AI in the enterprise: Q4 2024 report. Deloitte AI Institute.
70% of surveyed executives had moved 30% or fewer of their generative AI experiments into production; scaling cited as the top barrier to value.
Source ↗Ryseff, J., De Bruhl, B. F., & Newberry, S. J. (2024). The root causes of failure for artificial intelligence projects and how they can succeed. RAND Corporation.
More than 80% of AI projects fail — roughly double the failure rate of non-AI IT projects — primarily due to misaligned stakeholder understanding, not technology.
Source ↗Bonney, K., et al. (2024). Tracking firm use of AI in real time: A snapshot from the Business Trends and Outlook Survey (CES Working Paper No. 24-16). U.S. Census Bureau.
Only 5.4% of U.S. firms reported using AI in production as of early 2024; large firms (>250 employees) used it at roughly 3× the rate of small firms.
Source ↗Boston Consulting Group. (2020). Expanding AI's impact with organizational learning. BCG × MIT Sloan Management Review.
Of the value AI creates, roughly 10% comes from the algorithms themselves, 20% from technology and data, and 70% from people, processes, and adoption.
Source ↗
02
Productivity theater: activity vs. outcomes
Where AI usage rises without business metrics moving. Microsoft Work Trend Index, Upwork Research Institute, Dell Innovation Index, and the Brynjolfsson et al. call-centre QJE study on novice-vs-expert value distribution.
Microsoft & LinkedIn. (2024). AI at work is here. Now comes the hard part: 2024 Work Trend Index annual report.
75% of knowledge workers use AI at work; 78% bring their own AI tools (BYOAI); only 39% of users received any company training — exposing a measurement and governance vacuum.
Source ↗Upwork Research Institute. (2024). From burnout to balance: AI-enhanced work models for the future.
77% of employees say AI has added to their workload rather than reduced it; 47% don't know how to achieve the productivity gains their employers expect.
Source ↗Dell Technologies. (2024). Innovation catalysts: How GenAI pioneers are building the future of business. Dell Innovation Index.
Only 31% of leaders can quantify the impact of their GenAI initiatives on business outcomes, despite 76% saying GenAI will be significant or transformative.
Source ↗Brynjolfsson, E., Li, D., & Raymond, L. R. (2025). Generative AI at work. Quarterly Journal of Economics, 140(2), 889-942.
Access to a generative AI assistant in a call center increased productivity 14% on average, with 34% gains for novice workers and minimal effect for experienced workers — value is concentrated, not uniform.
Source ↗
03
AI governance failures and data-leak incidents
Named incidents and the empirical record on AI risk: the Samsung leak, Air Canada chatbot liability, Cyberhaven's telemetry on sensitive data sent to AI tools, IBM's Cost of a Data Breach, and Stanford HAI's AI Index incident tracking.
Park, K. (2023, May 2). Samsung bans staff's AI use after spotting ChatGPT data leak. Bloomberg.
Samsung Semiconductor engineers leaked confidential source code and meeting notes to ChatGPT on three occasions within 20 days, leading to an enterprise-wide ban on generative AI tools.
Source ↗Cyberhaven Labs. (2024). The cubicle culprits: AI data risks in the workplace. Cyberhaven Q2 2024 Report.
Sensitive corporate data sent to AI tools grew 485% year-over-year; 27.4% of corporate data fed to AI is now classified as sensitive, up from 10.7% the prior year.
Source ↗IBM Security & Ponemon Institute. (2024). Cost of a data breach report 2024. IBM Corporation.
Global average cost of a data breach reached $4.88M in 2024 — a 10% increase — with shadow data involved in 35% of breaches and adding $670,000 to average costs.
Source ↗Cecco, L. (2024, February 16). Air Canada ordered to pay customer who was misled by airline's chatbot. The Guardian.
A Canadian tribunal ruled Air Canada liable for misinformation provided by its customer-service chatbot — establishing precedent that companies are legally responsible for AI outputs.
Source ↗Maslej, N., Fattorini, L., Perrault, R., et al. (2024). The AI Index 2024 annual report. Stanford HAI.
Reported AI incidents grew 32.3% year-over-year in 2023; standardized responsible-AI benchmarks remain rarely adopted by major model developers.
Source ↗
04
Workflow redesign vs. tool bolt-on
The foundational re-engineering literature (Hammer, Brynjolfsson & Hitt) through the contemporary AI-operating-model work (McKinsey's Rewired, Iansiti & Lakhani's AI Factory, Davenport & Mittal's case studies).
Hammer, M. (1990). Reengineering work: Don't automate, obliterate. Harvard Business Review, 68(4), 104-112.
Firms that 'paved the cow paths' by automating existing processes achieved marginal gains; firms that obliterated and reengineered captured order-of-magnitude improvements.
Source ↗Brynjolfsson, E., & Hitt, L. M. (1998). Beyond the productivity paradox. Communications of the ACM, 41(8), 49-55.
IT investments generated returns only when paired with complementary organizational redesign; firms that invested in IT without restructuring saw flat or negative productivity.
Source ↗Yee, L., Chui, M., Roberts, R., & Xu, S. (2024). Rewired to outcompete. McKinsey Quarterly.
Digital and AI leaders outperform peers by 2-6× on TSR, but the differentiator is rewiring six organizational capabilities — not buying technology. Over 70% of digital transformations still fall short of objectives.
Source ↗Iansiti, M., & Lakhani, K. R. (2020). Competing in the age of AI. Harvard Business Review Press.
AI-centric firms operate at fundamentally different scope, scale, and learning curves than traditional firms, requiring an 'AI factory' operating architecture — bolting AI onto legacy processes preserves the bottleneck.
Source ↗Davenport, T. H., & Mittal, N. (2023). All-in on AI. Harvard Business Review Press.
Across 30+ AI-leading firms, value comes from companies that put AI at the heart of strategy and redesign operating models — not those that run isolated pilots.
Source ↗
05
Skill atrophy, automation complacency, and deskilling
Forty years of human-factors literature on what happens to operators paired with reliable automation — Bainbridge's Ironies, Parasuraman & Manzey's meta-analysis, Endsley's situation-awareness synthesis — alongside the new generation of AI-specific replications in radiology, colonoscopy, and knowledge work.
Bainbridge, L. (1983). Ironies of automation. Automatica, 19(6), 775-779.
The classic paradox: the more automation handles routine work, the less the human operator practices the skills needed to take over when automation fails — and the higher the stakes when they must.
Source ↗Carr, N. (2014). The glass cage: Automation and us. W. W. Norton & Company.
Carr documents how automation erodes tacit expertise, situational awareness, and craft across domains from aviation to medicine — leaving operators dependent and degraded.
Source ↗Parasuraman, R., & Manzey, D. H. (2010). Complacency and bias in human use of automation: An attentional integration. Human Factors, 52(3), 381-410.
Meta-analysis confirms automation complacency and automation bias are robust, replicated effects: high-reliability automation reliably causes humans to under-monitor and accept incorrect outputs.
Source ↗Endsley, M. R. (2017). From here to autonomy: Lessons learned from human-automation research. Human Factors, 59(1), 5-27.
Across 30 years of aviation, military, and driving research, higher automation reliably degrades operator situation awareness and lengthens reaction time when manual takeover is needed.
Source ↗National Transportation Safety Board. (2014). Asiana Airlines flight 214 accident report (NTSB/AAR-14/01).
NTSB identified flight crew over-reliance on automated systems and degradation of manual flying skills as contributing causes — the canonical aviation case of automation-induced skill decay.
Source ↗Kosmyna, N., et al. (2025). Your brain on ChatGPT: Accumulation of cognitive debt when using an AI assistant for essay writing task. arXiv:2506.08872.
EEG analysis showed LLM-assisted essay writers exhibited the weakest neural connectivity and lowest recall of their own writing; brain-to-brain coupling reduced by ~55% versus the brain-only group.
Source ↗Lee, H.-P., et al. (2025). The impact of generative AI on critical thinking. CHI '25.
Survey of 319 knowledge workers: higher confidence in GenAI correlated with less critical thinking effort; workers were more likely to defer judgment when they trusted the tool.
Source ↗Budzyń, K., et al. (2025). Endoscopist deskilling risk after exposure to AI in colonoscopy. The Lancet Gastroenterology & Hepatology.
Endoscopists' adenoma detection rate dropped roughly 20% (from 28.4% to 22.4%) when performing standard colonoscopies after regular exposure to AI assistance — direct measured deskilling in clinical practice.
Source ↗Dell'Acqua, F., et al. (2023). Navigating the jagged technological frontier (HBS WP 24-013).
BCG consultants using GPT-4 outperformed peers on tasks inside the AI 'frontier' by 12-43%, but on tasks outside the frontier they were 19 percentage points more likely to be wrong — they followed the AI off the cliff.
Source ↗Dell'Acqua, F. (2022). Falling asleep at the wheel. Harvard Business School Working Paper.
Recruiters paired with high-quality AI exerted less effort, were less attentive, and performed worse than recruiters paired with lower-quality AI — high reliability triggered complacency.
Source ↗Agarwal, N., et al. (2023). Combining human expertise with artificial intelligence (NBER WP 31422).
When radiologists received AI predictions, they systematically under-weighted their own private information; combined human+AI performance was often worse than AI alone or human alone.
Source ↗
06
Resistance to AI and the psychology of adoption
Algorithm aversion (Dietvorst et al.), algorithm appreciation (Logg et al.), the identity-threat mechanism behind AI resistance (Cao et al.), and the labour-relations framing of algorithmic management (Kellogg et al.).
Dietvorst, B. J., Simmons, J. P., & Massey, C. (2015). Algorithm aversion. Journal of Experimental Psychology: General, 144(1), 114-126.
After seeing an algorithm and a human each make the same forecasting error, people abandon the algorithm and stick with the human — even when the algorithm is demonstrably more accurate overall.
Source ↗Logg, J. M., Minson, J. A., & Moore, D. A. (2019). Algorithm appreciation. Organizational Behavior and Human Decision Processes, 151, 90-103.
Lay people show 'algorithm appreciation' — they weigh algorithmic advice more than identical human advice — but experts in the domain do the opposite, weighting algorithms less. Resistance concentrates where it matters most.
Source ↗Cao, L., et al. (2023). The dark side of AI identity. Computers in Human Behavior, 147, 107816.
Employees resist AI most strongly when it threatens occupational identity, not when it threatens productivity — change-management failures are identity failures.
Source ↗Kellogg, K. C., Valentine, M. A., & Christin, A. (2020). Algorithms at work. Academy of Management Annals, 14(1), 366-410.
Algorithmic management creates six new forms of control (the '6 Rs') and provokes systematic worker resistance; treating AI rollout as a technology project rather than a labor-relations project is the dominant failure mode.
Source ↗
07
Leadership requirements for AI transformation
The empirical literature on what differentiates organisations that capture AI value: senior engagement, operating-model rewiring, and the principal-agent framing of AI governance.
Westerman, G., Bonnet, D., & McAfee, A. (2014). Leading digital. Harvard Business Review Press.
Across 400+ large firms, 'Digital Masters' (strong digital + strong leadership capability) outperformed peers by 26% on profitability; technology capability alone correlated with underperformance.
Source ↗Fountaine, T., McCarthy, B., & Saleh, T. (2019). Building the AI-powered organization. Harvard Business Review, 97(4), 62-73.
The biggest barriers to AI value capture are organizational and cultural, not technical: shifting from siloed to interdisciplinary work, experimental to evidence-based decisions, and rigid to agile.
Source ↗Candelon, F., et al. (2024). AI adoption in 2024: 74% of companies struggle to achieve and scale value. BCG.
Only 26% of companies have developed the capabilities to move beyond proofs of concept and generate tangible value from AI; the gap is leadership focus and operating model — not tech budget.
Source ↗Ransbotham, S., et al. (2023). Achieving individual—and organizational—value with AI. MIT Sloan Management Review & BCG.
Across ~3,000 managers in 28 industries, organizational value from AI is roughly 5× more likely when the CEO is personally involved in AI strategy and when AI is integrated into employees' daily workflows.
Source ↗Jarrahi, M. H., & Ritala, P. (2025). Principal-agent theory and the future of AI governance in organizations. California Management Review.
Frames AI agents as principal-agent problems within organizations; argues governance must shift from tool oversight to delegation-of-authority design.
Source ↗
08
Failed AI implementations: case histories
Named cases that have entered the literature as canonical examples of AI-system failure — Watson Health, Amazon recruiting, the Optum/Obermeyer healthcare-bias case, Epic's sepsis predictor, Zillow Offers.
Obermeyer, Z., Powers, B., Vogeli, C., & Mullainathan, S. (2019). Dissecting racial bias in an algorithm used to manage the health of populations. Science, 366(6464), 447-453.
A widely-used U.S. healthcare risk algorithm covering ~200 million people systematically under-identified Black patients for high-risk care management.
Source ↗Dastin, J. (2018, October 10). Amazon scraps secret AI recruiting tool that showed bias against women. Reuters.
Amazon scrapped a multi-year internal AI recruiting tool after discovering it systematically penalized resumes containing 'women's' — trained on a decade of male-dominated hiring data.
Source ↗Wong, A., Otles, E., Donnelly, J. P., et al. (2021). External validation of a widely implemented proprietary sepsis prediction model. JAMA Internal Medicine, 181(8), 1065-1070.
Epic's sepsis prediction model — deployed at hundreds of U.S. hospitals — missed 67% of sepsis cases in external validation and generated alerts on 18% of all hospitalized patients.
Source ↗Parker, W., & Putzier, K. (2022, February 10). Zillow's shuttered home-flipping business lost $881 million in 2021. The Wall Street Journal.
Zillow shut down its AI-driven home-flipping unit after losses exceeding $880M and a 25% workforce cut, citing inability of its pricing algorithm to forecast home prices accurately at scale.
Source ↗Strickland, E. (2019, April 2). How IBM Watson overpromised and underdelivered on AI health care. IEEE Spectrum.
IBM's $5B+ investment in Watson Health collapsed after the system gave unsafe cancer treatment recommendations; IBM sold Watson Health assets in 2022 for a fraction of cost.
Source ↗