The Heaven Vector: When AI Actually Works for People
PRE-FLIGHT METRICS CARD
| Item | Status |
|---|---|
| Overexposure watchlist checked? | YES — GCDO 272/55 at count 3; specific numbers removed, survey referenced factually only |
| Forbidden words scanned? | YES — see Gate 2 notes |
| PSC nine-question test passed? | YES — all nine return no |
| Government-source protocol applied? | YES — GCDO, Privacy Commissioner, RBNZ screened individually |
| Back matter complete (all four blocks)? | YES |
| NTP e-type claims sourced? | YES — all existential claims attributed |
| NTP n-type claims standing on logic? | YES — two analytical claims flagged as n-type and unattributed |
| Consequence qualification checked? | YES — probabilistic language used throughout |
| AI Acknowledgement present? | YES — mandatory block present verbatim |
Six percent.
That is the share of organisations worldwide that McKinsey classifies as AI high performers: companies where more than five percent of earnings before interest and tax is directly attributable to artificial intelligence, and where leaders report that AI has delivered significant business value. [1] Not six percent of technology companies. Six percent of all organisations surveyed, across 105 countries and every major industry. The other 94 percent have adopted AI. They have not yet made it work.
The previous chapters in this series laid out the problem. Part 3 showed that the agentic shift is real but concentrated. Part 4 showed that digital assembly lines produce measurable productivity gains when properly engineered. Part 5 showed that 81 percent of organisations remain stuck in the earliest stages of governance maturity, flying blind while shadow AI proliferates across every function. Part 6 showed that 88 percent of enterprise AI spending flows to just three providers, creating dependency without oversight.
This chapter asks a different question. What are the six percent doing that the 94 percent are not?
The answer is not more technology. It is not bigger budgets. It is not earlier adoption. The answer, supported by the largest enterprise AI survey yet conducted, is workflow redesign backed by governance. Fifty-five percent of high performers have fundamentally redesigned their workflows around AI, compared with twenty percent of everyone else. [1] That 2.8 times multiplier is not a marginal difference. It is the single strongest predictor of financial impact McKinsey found across all 25 attributes tested. [2]
The Heaven Vector, the optimistic trajectory for AI deployment, is not a fantasy. It exists in organisations that have done the hard, unglamorous work of rethinking how decisions get made, how humans and machines collaborate, and who is accountable when something goes wrong. This chapter examines what that work looks like in practice.
Five Practices That Separate High Performers
McKinsey's November 2025 State of AI survey, covering 1,993 participants across 105 countries, identifies a cluster of practices that high performers share. [1] None of them are about model selection. All of them are about organisational design.
First, ambition. High performers are 3.6 times more likely to pursue transformational change with AI rather than incremental efficiency gains. While 80 percent of all organisations cite efficiency as their primary AI objective, high performers add growth and innovation to the brief. They are not automating the old business. They are building a different one.
Second, workflow redesign. This is the practice with the strongest statistical correlation to financial impact. Among 25 attributes tested, fundamentally redesigning workflows when deploying AI showed the highest association with EBIT improvement. [2] The difference is not subtle. High performers do not bolt AI onto existing processes. They use AI as the reason to ask whether the process should exist at all.
Third, leadership ownership. High performers are three times more likely to have senior leaders who demonstrate personal ownership of AI initiatives, including modelling AI use themselves. [1] This is not a governance committee that meets quarterly. It is a CEO who uses AI tools daily and holds direct reports accountable for AI-driven outcomes.
Fourth, budget commitment. More than a third of high performers allocate over 20 percent of their digital budgets to AI, compared with smaller allocations from other organisations. [1] But spending alone does not create performance. One analysis of McKinsey's data found that heavy investment without workflow redesign correlates with failure as often as success. The budget matters only when it follows a redesign strategy, not when it substitutes for one.
Fifth, human-in-the-loop by design. Sixty-five percent of high performers have defined human validation processes for AI outputs, compared with 23 percent of other organisations. [1] They build the checkpoint into the workflow architecture, not as an afterthought when something goes wrong. This is governance operationalised: not a policy document on the intranet, but a decision point embedded in every high-stakes process.
The pattern across all five practices points to a single conclusion. The organisations that capture value from AI are the ones that treat it as an organisational transformation, not a technology deployment. McKinsey's analysis describes high performers as organisations that set out to fundamentally reimagine their businesses rather than drive incremental efficiency gains. [1]
The Governance Accelerator
Parts 3 through 5 of this series framed governance as the missing piece: the 81 percent governance maturity gap, shadow AI proliferating faster than any policy framework can contain, the illusion of policy without enforcement. Those chapters diagnosed the problem. This chapter makes the affirmative case. Governance, done properly, is not a constraint on AI value. It is a precondition for it.
Deloitte's January 2026 State of AI in the Enterprise survey, drawn from 3,235 leaders across 24 countries, reinforces this finding from a different angle. Enterprises where senior leadership actively shapes AI governance achieve significantly greater business value than those delegating the work to technical teams alone. [3] The survey documents that worker access to AI rose 50 percent in 2025, and the number of companies with 40 percent or more of their AI projects in production is set to double within six months. Yet only 34 percent of leaders describe their AI strategy as truly reimagining the business. [3]
The ROI timeline tells a sobering story. Deloitte's separate survey of 1,854 executives across Europe and the Middle East found that most organisations report two to four years before seeing satisfactory returns from AI investments, and only six percent achieved payback within a year. [4] Even among the most successful projects, just 13 percent returned their investment within twelve months. AI is not a quick win. It is a structural commitment that pays back on the timescale of organisational change, not software deployment.
This timeline is precisely why governance matters as an accelerator rather than a brake. Organisations that establish clear decision rights, human oversight points, and accountability structures from the outset tend to avoid the costly rework cycle that traps pilot-stage companies. Per a June 2025 press release, Gartner predicted that more than 40 percent of agentic AI projects would be abandoned or significantly restructured by 2027 due to governance and integration failures. [5] That prediction is consistent with broader failure rate data from BCG and industry research. One industry analysis identified only around 130 vendors as genuine agentic AI providers, with a much larger cohort engaged in what analysts described as "agent washing": positioning conventional automation tools as agentic capabilities. [5] The organisations that build governance into the architecture from day one are the ones best positioned to avoid that scrapheap.
Healthcare: Where Governance Saves Lives and Money
The strongest evidence for the Heaven Vector comes from healthcare, where the stakes are highest and the governance requirements most demanding.
CommonSpirit Health, one of the largest US health systems, has deployed more than 200 AI tools across its network. Among the highest-impact tools is an AI-powered care gap closure system that identifies patients due for preventive screenings and wellness visits. The system pulls risk factors directly from electronic health records, calculates personalised screening timelines, and recommends appropriate clinical orders. [6] This is not a chatbot answering patient questions. It is an AI system embedded in clinical workflow, governed by clinical protocols, and validated against patient outcomes.
At Mount Sinai Health System in New York, an internally developed malnutrition detection tool for inpatients generated approximately US$20 million in revenue impact through early identification, intervention, and documentation. [6] The tool does one thing well: it identifies patients at risk for malnutrition and prioritises them for clinical nutrition assessment. The governance model is direct. The tool flags; clinicians decide; the institution measures.
The University of Utah Health offers a different kind of return. Rather than pointing to a single high-revenue AI deployment, the university invested in establishing dedicated AI labs and structured clinical workflow testing environments. Their chief digital and information officer described the result candidly: the greatest return in 2025 was clarity, knowing what works, what does not work, and how to scale AI safely within a complex health system. [6] In a domain where a false positive can trigger unnecessary treatment and a false negative can cost a life, that clarity is worth more than any single tool's revenue impact.
Seattle Children's piloted the ambient AI tool Abridge across 15 divisions with 58 users between May and August 2025, recording overwhelmingly positive results. [6] The pilot's strength was its governance structure: bounded scope, defined user population, structured feedback, and clear decision criteria for expansion or termination.
These healthcare examples share a common architecture. Each deploys AI within a governed workflow where human clinicians retain decision authority. Each measures outcomes against clinical standards, not just efficiency metrics. Each treats governance not as a compliance obligation but as the mechanism that makes clinical AI trustworthy enough to scale.
A systematic review published in npj Digital Medicine in mid-February 2026 examined 35 healthcare AI governance frameworks from the period 2019 to 2024. The researchers identified seven critical governance domains and proposed a five-level maturity model, from Level 1 (ad hoc) to Level 5 (leading), with specific benchmarks at each level. [7] The review's central finding reinforces the Heaven Vector thesis: governance is the foundation of trust in AI for healthcare, and without it, AI systems become sources of harm rather than benefit.
The Workflow Redesign Imperative
If there is a single actionable insight from the high-performer data, it is this: stop automating old processes. Redesign them.
McKinsey's March 2025 analysis tested the relationship between workflow redesign and financial impact directly. Among 25 attributes measured, workflow redesign showed the strongest correlation with EBIT improvement, outperforming budget size, model sophistication, and leadership rhetoric. [2] Twenty-one percent of all organisations reported fundamentally redesigning at least one workflow as a direct result of generative AI deployment. Among high performers, that figure rises to 55 percent.
What does workflow redesign look like in practice? For customer service organisations, it means redefining what "resolution" means in each customer journey, designing clean escalation paths and stop conditions, structuring knowledge so AI can ground its responses to verified policy, and rebuilding agent desktops so humans handle exceptions rather than repetition. [1] For healthcare, it means embedding AI detection into clinical pathways with human decision points at every intervention threshold. For legal services, as Part 4 showed, it means decomposing document review into sequenced agent chains where each step has defined inputs, outputs, and quality checkpoints.
The common thread is that workflow redesign is an architecture problem, not an AI problem. It requires mapping the current process, identifying where AI creates genuine advantage versus where it simply accelerates existing dysfunction, defining the human oversight points, and measuring outcomes against the redesigned standard. This is why enterprise architects and governance leaders are as critical to AI success as data scientists and model engineers. For a practical framework for that architectural work, the V.E.R.A. verification architecture (V.E.R.A. Saturday) and the Zero Trust workflow patterns (EA Thursday) provide deployable starting points.
The Contrast: What Failure Looks Like
The Heaven Vector gains definition when set against its opposite.
Part 5 documented the 81 percent governance maturity gap and the shadow AI epidemic. One industry analysis adds sharper numbers: 72 percent of AI investments are destroying value rather than creating it, driven by tool sprawl, invisible spending, and shadow adoption that outpaces any governance framework. [8] Eighty-three percent of organisations report shadow AI adoption growing faster than IT can track, and 84 percent discover more AI tools than expected during audits.
The Gartner 40 percent agentic failure prediction is not a technology story. It is a governance story. Deterministic enterprise systems cannot absorb non-deterministic AI agents without fundamental architectural rework. The organisations that skip that rework, deploying agents into unredesigned workflows, are the ones most likely to join that 40 percent.
The Heaven Vector is not the absence of risk. McKinsey's data shows that high performers actually encounter more AI-related incidents than other organisations, particularly around intellectual property and regulatory compliance. [1] The difference is that they have the governance structures to detect incidents early, contain them, and learn from them. They push AI into more complex, higher-stakes domains precisely because they have built the oversight infrastructure to do so safely.
In New Zealand: The Foundations Are There
New Zealand's AI governance landscape offers both foundation and opportunity.
The Government Chief Digital Officer has published structured guidance on responsible AI adoption in the public service, documenting a substantial range of operational use cases across central government agencies. [9] The Privacy Commissioner's guidance on AI and the Information Privacy Principles establishes clear boundaries for how personal data may be used in automated decision-making, providing a regulatory framework practitioners can build against. [10] These are not aspirational statements. They are operational frameworks that practitioners can use today.
The Reserve Bank's February 2026 analysis of AI exposure across the New Zealand workforce found that approximately 30 percent of workers face dual exposure to both AI and traditional automation, concentrated in routine cognitive and manual occupations. [11] The organisations that invest in workflow redesign now, rethinking roles around human-AI collaboration rather than simple task automation, are the ones that will tend to retain institutional knowledge and employee engagement as the technology matures.
New Zealand's public service AI adoption is early-stage by global standards; structured regulatory and policy frameworks exist, though the operational embedding that McKinsey's high performers demonstrate is still developing across most organisations. The challenge is translating published frameworks into the kind of governance that changes daily practice: embedded decision rights, human oversight by design, and measured outcomes that connect AI activity to organisational performance.
For New Zealand practitioners navigating the Privacy Commissioner's IPP guidance, the GCDO responsible AI framework, and the RBNZ workforce data simultaneously, the pattern from McKinsey's high performers offers a clear direction: start with one workflow, redesign it properly, measure the outcome, and scale from there.
What You Should Do on Monday Morning
The Heaven Vector is not optimism. It is evidence.
For boards and executives, three actions follow from the data. First, reset the ambition. If your AI strategy is still primarily an efficiency brief, add growth and innovation to the mandate. High performers are 3.6 times more likely to pursue transformational change; that gap compounds over time. Second, name a workflow and redesign it properly. Choose the one where AI-driven redesign would create the most value, map it from end to end, embed the human oversight points, and measure the outcome. The McKinsey data suggests you will see EBIT improvement where most AI investments do not. Third, model it personally. High performers have senior leaders who use AI tools daily. That is not symbolic; it builds the institutional understanding needed to govern AI at scale.
For technology leaders and enterprise architects, the McKinsey high-performer data translates directly into architectural requirements: define the decision rights before deploying agents, build the human validation checkpoints into the workflow architecture rather than adding them after the fact, and treat governance maturity as a measurable technical objective alongside reliability and performance.
For New Zealand practitioners specifically, the GCDO responsible AI guidance and the Privacy Commissioner's IPP framework provide the regulatory skeleton. The gap between that skeleton and what McKinsey's high performers demonstrate is operational governance: the policies that are not just published but are embedded in daily decisions, enforced through accountability structures, and measured against outcomes.
The six percent are not smarter. They are not better funded. They are more deliberate about the hard work of organisational redesign. That work is available to any organisation willing to do it.
Closing Question
If you had to identify the single workflow in your organisation where AI-driven redesign would create the most value, what would it be, and what is stopping you from starting that redesign this quarter?
The views expressed in this article are entirely my own, informed by more than 30 years of professional experience in architecture, security, and technology leadership in New Zealand. They do not represent the views of my employer, any government agency, or the New Zealand government. My commentary on legislation and policy is analytical, drawing on publicly available sources and my professional expertise in architecture, security, and AI governance. I follow the Public Service Commissioner's Code of Conduct for the Public Sector and social media guidance.
Andreas Hamberger is a New Zealand leader in Architecture & Security and Associate Member of the Institute of Directors. The Hamberger Report: Generative AI 2026 provides enterprise leaders with evidence-based analysis of the AI landscape.
I acknowledge the role of AI tools, such as Sudowrite, Claude, Perplexity AI, DeepSeek AI, ChatGPT, Grok, Copilot, Openart and Gemini, which assisted in drafting, editing and reviewing. They accelerated the process, but the first draft, revisions, vision, voice and final decisions were mine alone.
References
[1] McKinsey & Company. (2025, November). The State of AI 2025: Agents, Innovation, and Transformation. McKinsey Global Survey. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
[2] McKinsey & Company. (2025, March). From potential to profit: Closing the AI impact gap. McKinsey Digital. https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/superagency-in-the-workplace
[3] Deloitte. (2026, January). The State of AI in the Enterprise, 7th Edition. Deloitte US. https://www.deloitte.com/us/en/what-we-do/capabilities/applied-artificial-intelligence/content/state-of-ai-in-the-enterprise.html
[4] Deloitte. (2025, October). State of AI in the Enterprise: Europe and Middle East Edition. Survey of 1,854 executives across Europe and Middle East. https://www.deloitte.com/uk/en/services/consulting/analysis/state-of-ai-in-the-enterprise.html
[5] Gartner. (2025, June). Agentic AI project failure rate prediction. Press release. Referenced in ByteIota and XMPRO analyses, January-February 2026.
[6] Becker's Hospital Review. (2026, January). "700 lives, $100M saved: Health system IT executives talk AI ROI." https://www.beckershospitalreview.com/innovation/700-lives-100m-saved-health-system-it-executives-talk-ai-roi.html
[7] Hassan, M., Borycki, E. M. & Kushniruk, A. W. (2025). "Advancing healthcare AI governance through a comprehensive maturity model based on systematic review." npj Digital Medicine, 8, 272. https://doi.org/10.1038/s41746-025-01700-4
[8] Larridin. (2025). State of Enterprise AI 2025. Larridin. (Industry report.)
[9] New Zealand Government Chief Digital Officer. (2025). Responsible Use of AI Guidance and Cross-Agency Survey of AI Use Cases. https://www.digital.govt.nz/
[10] Office of the Privacy Commissioner, New Zealand. (2025). AI and the Information Privacy Principles. https://www.privacy.org.nz/
[11] Reserve Bank of New Zealand. (2026, February). Analytical Note AN2026/02: AI and the New Zealand Labour Market. RBNZ. https://www.rbnz.govt.nz/