The fifth of five field notes on the constraints that sit between AI investment and earnings.

· · ·

The problem has a name: the Accountability Vacuum. It is the condition in which AI projects are simultaneously owned by everyone and accountable to no one. The CTO champions the infrastructure. The CDO claims the data strategy. The innovation lab runs the pilots. The business units absorb the costs when pilots fail.

When a model hallucinates in a customer-facing context, produces a biased output, or simply stops delivering the projected return, the organizational response is typically a product tweak, a revised safety document, and a statement about lessons learned. No one is demoted. No one's compensation moves. No one is named.

The accountability vacuum defined

OpenAI's 2025 State of Enterprise AI report describes the structural pattern directly: AI initiatives are owned by multiple stakeholders simultaneously, with no single executive accountable for end-to-end performance and risk. AI centers of excellence advise but do not own profit and loss. They cannot intervene with authority when a model underperforms or causes harm because their authority is advisory — dependent on goodwill rather than enforceable governance.

Menlo Ventures found the same split in its 2025 generative AI enterprise survey: initiatives are championed by innovation teams or individual business units, and no single C-level owner is responsible for AI return on investment across the enterprise. Menlo's data showed that companies with a named AI P-and-L owner produced materially better ROI and abandoned fewer pilots than those operating under distributed experimentation with no central accountability.

Forty-two percent of firms are abandoning most AI projects before full deployment — a sharp increase from the prior year. Because no one owns the outcome, no one improves the process. The same structural errors reproduce themselves in the next initiative. Costs are absorbed as innovation spend rather than attributed to any accountable executive.

Why agentic AI sharpens the problem

Agentic AI sharpens the accountability vacuum because an agent fits no existing ownership category. It is funded like technology but it behaves like a worker — taking actions and making decisions that a human role used to own. Capital has a budget owner. Labor has a manager. An agent that is neither lands in the gap between them, which is exactly where accountability disappears.

PwC's 2025 Responsible AI Survey found that many organizations have policies, frameworks, and references to NIST's AI Risk Management Framework — but treat them as documentation exercises rather than operational disciplines with assigned owners and enforceable controls. PwC's own top recommendation is to clarify accountability through a defined three-lines-of-defense model in which builders, reviewers, and risk owners each have explicit roles they cannot share or delegate away. The recommendation would be unnecessary if enterprises had already solved this.

Deloitte's 2026 research finds that only about one in five organizations has a mature governance model even for the autonomous AI agents they are actively deploying. The other four are running agentic AI in the gap between capital and labor, where accountability disappears.

How leaders escape it

Leaders who escape the Accountability Vacuum do three things differently.

First, they assign ownership with specificity. One executive owns AI outcomes — including risk outcomes — across the enterprise. That person's name is attached to the initiative before launch, not discovered after failure.

Second, they connect governance to decision rights. The Wharton 2025 AI Adoption Report describes a visible shift among higher-performing adopters toward linking accountability to measurable business KPIs — precisely because earlier waves of deployment left outcome ownership undefined. Governance committees in mature programs control deployment gates and budget, not just principles. They can say no and make it stick.

Third, they build consequence into the system. This is the mechanism that GenGov™ and H2AI™ are designed to formalize: assigning structural ownership and transferring capability with accountability attached. Accountability without consequence is a policy document. Consequence without clear ownership is noise. The instruments work because they address both simultaneously — binding a named owner to a measurable outcome and ensuring that the capability to deliver is transferred alongside the responsibility to be measured.

The five constraints that prevent AI investment from converting to earnings are structural, not technical. The Accountability Vacuum is the one that compounds every other constraint: without named ownership, no remediation architecture holds.

The invitation

The Constraint Map℠ surfaces where accountability is absent in your AI portfolio and identifies the ownership architecture needed to close it. The Readiness Assessment delivers that picture in two weeks: named owners, decision rights, enforcement mechanisms, and a consequence structure that makes governance operational rather than advisory.

It begins with a thirty-minute diagnostic scoping conversation that surfaces a specific constraint in your own function, verifiable against data you already hold.

If that is the conversation you are ready to have, we would welcome it.

michael@magruder.co · doug@magruder.co

· · ·

The Observation Deck is a five-part series on the constraints that prevent AI investment from converting to earnings. The series is complete. The Readiness Assessment is the next step.

Sources: CIO linkage figure from Dataiku/Harris Poll survey, 2024–2025. Abandoned initiatives from 2025 enterprise adoption synthesis. OpenAI State of Enterprise AI, 2025. Menlo Ventures 2025 generative AI enterprise survey. PwC Responsible AI Survey, 2025. Wharton AI Adoption Report, 2025. Deloitte, "The State of AI in the Enterprise, 2026."

Magruder & Company · magruder.co · customercore.co

© 2026 Magruder & Company. All rights reserved.

Customer Core™, GenGov™, and H2AI™ are trademarks, and Constraint Map℠ is a service mark, of Magruder & Company.

The fourth of five field notes on the constraints that sit between AI investment and earnings.

· · ·

That firm's story is not an edge case. It is a preview of what finance leaders will find when they look carefully at how AI is actually being purchased and used inside their organizations.

The constraint is structural. AI procurement has decentralized faster than finance functions have adapted to track it. Employees subscribe to tools individually and expense them. Department heads approve SaaS platforms with embedded AI features without flagging them to IT or the CFO. Teams build workflows on top of free tiers of foundation models, then upgrade to paid plans when usage scales — often without a formal purchase order. The result is a spending pattern that is fragmented by design, even when no single actor intends to obscure it.

The shape of invisible spend

A separate analysis found that 89 percent of AI activity inside organizations goes unseen by IT and security teams. The average enterprise has more than 1,200 unofficial AI-connected applications in active use. These vendor figures come from firms with a commercial interest in the problem and should be read as directional rather than audit-grade. But the direction they point is consistent with what finance leaders report when they actually attempt to produce a consolidated AI spend number: they cannot.

The spending pattern is fragmented by design. Employees subscribe to tools individually. Department heads approve SaaS platforms with embedded AI features — Salesforce Einstein, Microsoft Copilot, Notion AI — without flagging them to IT or the CFO. Teams build workflows on free tiers of foundation models, then upgrade to paid plans when usage scales, often without a formal purchase order. No one is hiding the spend. The structure of modern SaaS procurement hides it automatically.

What it costs

The cost has two components, and organizations typically underestimate both.

The first is direct financial waste. Roughly 30 to 50 percent of AI-related cloud spend is lost to idle resources, overprovisioned infrastructure, and workloads that were never optimized after deployment. When no one owns the full inventory, no one is positioned to rationalize it. Duplicate tools proliferate. Licenses go unused. Teams in separate business units pay independently for capabilities that already exist elsewhere in the organization.

The second cost is strategic. Average enterprise AI budgets are projected to reach $85,521 per month in 2025 — up 36 percent from the prior year — yet only 51 percent of organizations report that they can confidently evaluate whether their AI investments are delivering returns. Enterprise generative AI investment is projected to grow another 50 percent in the near term, but only 6 percent of companies report achieving 75 percent or more of their expected ROI, and only 15 percent of AI decision-makers report any EBITDA lift from the spend at all.

When spend is invisible, ROI is unmeasurable. When ROI is unmeasurable, the board has no basis for deciding where to invest more and where to stop. The CFO is not managing an AI portfolio. She is managing a collection of unrelated line items that no one has assembled into a coherent picture.

The discipline of spend visibility

Organizations that escape this pattern treat AI spend visibility as a precondition for AI strategy, not a byproduct of it. That means tagging AI resources at the point of purchase, attributing cost by team, project, and workload, and requiring that any AI-related expense — whether a SaaS subscription with embedded AI features or a direct API contract — flow through a defined approval and tracking process.

It means regular audits of sanctioned applications to surface the embedded AI capabilities that teams are already using without governance awareness: the kind of Salesforce Einstein usage that improves sales conversion but never appears in any AI budget discussion. Critically, it means building a single consolidated spend register that the CFO can interrogate at any point in the fiscal year.

The organizations that do this consistently find that their actual AI spend is materially higher than what finance previously tracked — often 30 to 60 percent higher once shadow usage and embedded tool costs are surfaced. That discovery is uncomfortable. It is also the only honest starting point for converting AI investment into measurable earnings.

The firms that will capture durable value from AI over the next three years are not necessarily the ones spending the most. They are the ones that know what they are spending, why, and whether it is working.

The invitation

If your finance team cannot produce a consolidated AI spend number today, the constraint is operating. The Constraint Map℠ surfaces the full inventory of AI activity — sanctioned and unsanctioned — and establishes the spend governance architecture that converts a fragmented cost profile into a manageable portfolio.

The Readiness Assessment delivers that picture in two weeks: a consolidated spend register, an approval and tracking framework, and a ROI measurement architecture built around the data your organization already holds. It begins with a thirty-minute diagnostic scoping conversation that surfaces a specific constraint in your own function, verifiable against data you already hold.

If that is the conversation you are ready to have, we would welcome it.

michael@magruder.co · doug@magruder.co

· · ·

Next in The Observation Deck: The Accountability Vacuum — when nobody owns the AI failure, and you are paying for it.

Sources: Financial services IPO incident from Help Net Security, 2025. Unauthorized tool usage from Larridin, 2025. Shadow AI visibility from Help Net Security, 2025. Personal account usage from Netskope, 2026. Cloud spend waste from MILL5, 2025. Monthly AI budget and evaluation confidence from MILL5, 2025. ROI achievement from Glean, 2025. EBITDA lift from Forrester, 2026.

Magruder & Company · magruder.co · customercore.co

© 2026 Magruder & Company. All rights reserved.

Customer Core™, GenGov™, and H2AI™ are trademarks, and Constraint Map℠ is a service mark, of Magruder & Company.

The third of five field notes on the constraints that sit between AI investment and earnings.

· · ·

This is not an unusual board meeting. It is the norm.

IDC research finds that only 4 of every 33 AI proofs of concept in large enterprises graduate to production — an 88 percent non-conversion rate. S&P Global reports that large enterprises abandoned an average of 2.3 AI initiatives in 2025, each carrying approximately $7.2 million in sunk cost, and the share of companies abandoning the majority of their AI initiatives rose from 17 percent to 42 percent in a single year. These are not figures from companies that ignored AI. They are figures from companies that invested in it seriously and still cannot show the work.

The problem has a name: the governance gap. It is the pattern where AI deployment outpaces the structures needed to make that deployment auditable, accountable, and connected to financial outcomes. AI activity accumulates. Governance does not keep pace. The result is an organization running AI it cannot fully inventory, producing decisions it cannot fully explain, and reporting results it cannot fully defend.

What the numbers confirm

The constraint is structural, not accidental. Most AI programs are launched under board and executive pressure to demonstrate action, with use cases selected for visibility rather than economic rigor. Pilots are designed to succeed in controlled environments, running on curated data, outside the organization's authentication systems, compliance frameworks, and systems of record. When the organization attempts to move a pilot toward production, it collides with integration complexity, regulatory scrutiny, and the absence of any governance infrastructure to guide the transition. The pilot stalls. A new pilot begins. The cycle repeats.

Aona AI's 2026 survey finds that only 23 percent of organizations have a formal AI governance framework, while shadow AI — unsanctioned use of AI tools outside IT and security oversight — is widespread. The deployment-versus-scale gap tells the same story: roughly 23 percent of organizations are running agentic AI somewhere in the business, but only 2 percent have actually scaled it. Deloitte's 2026 State of AI in the Enterprise report finds that only about one in five organizations has a mature governance model even for the autonomous AI agents they are actively deploying. McKinsey's 2025 State of AI survey reports that roughly 39 percent of organizations report any positive EBIT impact from AI — meaning six in ten are deploying AI without seeing material earnings improvement on their own self-reported metrics.

Why governance falls behind deployment

The cost to organizations is concrete and compounding. At the most immediate level, there is the direct financial loss from abandoned initiatives: millions of dollars per program, per year, with no recoverable asset and no institutional learning captured in a usable form. At the operational level, there is the cost of parallel AI ecosystems that no one can map, creating data exposure, compliance risk, and the growing possibility that a regulatorily significant decision has been shaped by an AI system whose logic no one can reconstruct.

At the strategic level, there is competitive erosion. BCG's research identifies that roughly 5 percent of firms characterized as future-built are achieving up to five times the revenue uplift and three times the cost reductions from AI compared to peers operating in the same domains — while approximately 60 percent of companies see negligible financial impact despite substantial spend. The gap between those two groups is not closing. It is widening, and the primary driver is not model quality. It is operating infrastructure.

PwC's global CEO survey found that more than half of CEOs report their companies have not reduced costs or increased revenues from AI over the prior 12 months, and only one in eight report achieving both higher revenues and lower costs from AI investment.

How the pattern breaks

Organizations that escape the governance gap treat AI governance as an operational discipline, not a compliance function. They integrate AI oversight into existing risk, audit, and finance structures so that every deployed AI system has an owner, a documented purpose, defined success metrics tied to business outcomes, and a traceable decision log. They establish gate criteria that a pilot must satisfy before advancing to production — including evidence of integration feasibility, measurable ROI hypothesis, and regulatory defensibility.

They build what Deloitte calls a living AI backbone: a governed data and infrastructure architecture that gives every AI system access to reliable inputs and produces outputs that can be attributed, audited, and connected to financial performance. And they establish a single authoritative inventory of what AI the organization is running, updated continuously, accessible to finance and risk leadership, and reviewable by the board without requiring a scavenger hunt across business units.

These organizations also redesign workflows around AI rather than layering AI onto legacy processes. McKinsey's research identifies workflow redesign as the single most impactful organizational factor in determining whether AI delivers EBIT impact — yet only about 21 percent of companies report having fundamentally redesigned even some of their workflows in response to AI.

Governance retrofitted after sprawl is expensive and incomplete. Governance built as a precondition for deployment is what makes AI investment convertible to earnings.

The invitation

If the board meeting in the opening of this piece sounds familiar, you are seeing the constraint clearly. The Constraint Map℠ identifies where in the lifecycle the governance discipline is missing, confirms the pattern against your own operating data, and points to the remediation architecture that closes the gap structurally.

The Readiness Assessment converts that picture into an operating instrument: a governed inventory of AI systems, gate criteria for pilot advancement, and a finance-legible reporting structure that makes the CFO's question answerable. It begins with a thirty-minute diagnostic scoping conversation that surfaces a specific constraint in your own function, verifiable against data you already hold.

If that is the conversation you are ready to have, we would welcome it.

michael@magruder.co · doug@magruder.co

· · ·

Next in The Observation Deck: Invisible Spend — the AI budget your finance team doesn't know exists.

Sources: 95% pilot impact from MIT NANDA, 2025. POC-to-production rate from IDC, 2025. Abandoned initiatives from S&P Global, 2025. Governance framework adoption from Aona AI, 2026. Agentic AI scaling from Capgemini 2025 and Camunda 2026. Mature governance from Deloitte 2026. EBIT impact from McKinsey 2025. Future-built performance from BCG 2025. CEO survey from PwC 2025.

Magruder & Company · magruder.co · customercore.co

© 2026 Magruder & Company. All rights reserved.

Customer Core™, GenGov™, and H2AI™ are trademarks, and Constraint Map℠ is a service mark, of Magruder & Company.

Eighty-seven percent of organizations report having formal AI governance structures. Twenty-two percent say those structures work. (ADR Center, 2025) The gap between those two numbers is where most AI investment quietly disappears.

The second of five field notes on the constraints that sit between AI investment and earnings.

· · ·

Somewhere in your organization there is a completed training report. Two hundred employees certified. Completion rates in the high nineties. A Center of Excellence with a published framework. A governance committee that holds its quarterly meeting and produces its quarterly minutes.

And somewhere else in your organization, a CFO has begun asking a question that no one has been able to answer: where is the productivity gain?

The distance between those two things is the second constraint we map. We call it competence theater, and it is the one most organizations are least willing to name, because the investments that produce it feel responsible. Certifications, Centers of Excellence, governance committees: these are defensible expenditures. The problem is not that they are wrong. The problem is that without outcome discipline they are incomplete, and incomplete looks identical to effective until the question gets asked.

What the numbers confirm

The pattern shows up in the research with unusual consistency.

McKinsey's 2024 State of AI survey found that while sixty-five percent of respondents reported their organizations regularly using generative AI, nearly double the prior year's rate, only forty-six of eight hundred and seventy-six respondents could attribute a meaningful share of their EBIT to a gen AI deployment. The adoption signal was moving sharply. The earnings signal was not moving at all.

The governance data is, if anything, more direct. The ADR Center's 2025 enterprise AI governance survey found that eighty-seven percent of organizations report having formal governance principles and policy structures in place, but only twenty-two percent describe those structures as working effectively in practice. Most characterize their governance as possessing good structures and inconsistent execution. The coverage problem sharpens across the AI lifecycle: seventy-two percent apply governance at development, forty-four percent during post-deployment monitoring, and four percent at system retirement. The infrastructure of readiness is in place. The discipline of outcomes is not.

Training tells the same story from a different angle. Deloitte's 2026 research found that eighty-four percent of companies have not redesigned a single job to fit AI. The training investment lands, the certifications are earned, and employees return to workflows that were never rebuilt to absorb what they learned. The credential validates awareness. It says nothing about application.

One figure captures the contradiction. A 2026 readiness study from Drexel LeBow and Precisely found that eighty-seven percent of organizations reported having the infrastructure they needed, while forty-two percent named that same infrastructure as their single biggest obstacle. Both numbers came from the same population. That contradiction is most consistent with readiness being reported as declared rather than demonstrated.

Why the credential is not the capability

The reason this constraint is so durable is the same reason the first one is: it is structural, not technical, and it is reinforced by incentives that operate in the wrong direction.

Training programs are measured by completion. Governance committees are measured by meeting cadence. Centers of Excellence are measured by framework publication. None of those metrics are wrong, exactly. They are just insufficient. They measure the inputs to capability without measuring whether capability ever arrived.

UNESCO's 2024 AI Competency Framework for Teachers establishes that genuine AI competence must be demonstrable and progressive, observable in practice rather than declared through credential. The framework's principles apply across organizational contexts wherever capability is declared before it is demonstrated. When an organization stops at awareness, it has built the appearance of capability, not the substance of it. That distinction is invisible in a completion report and obvious in an operating result.

There is also a governance liability that tends to be underappreciated. Committees that meet but do not decide, frameworks that exist but are not enforced, checklists that are completed but not monitored: these structures produce the documentation of diligence without the substance of control. As regulatory scrutiny of AI systems increases across jurisdictions, governance that exists on paper without functioning in practice is increasingly difficult to defend. An organization that has documented its framework thoroughly without enforcing it has not reduced its risk; it has clarified the record of what it chose not to do.

The compounding cost is visible in three places. First, time: organizations that spend a year or more in training and governance cycles before deploying a single production use case are ceding ground to competitors who are already measuring operational gains. Second, organizational trust: employees who complete programs that never change how work is done become skeptical of the next initiative. That skepticism is earned, and it makes every subsequent change harder to move. Third, overhead without return: a Center of Excellence that has operated for two or more years without a portfolio of measurable deployed outcomes is a real cost with no documented result.

How the pattern breaks

What distinguishes organizations that convert AI investment into earnings is not better training or more sophisticated frameworks. It is outcome discipline applied from the start, before the curriculum is designed and before the governance structure is chartered.

The firms that are compounding real gains from AI share a specific orientation. They define measurable success criteria before any capability-building begins. They tie every training activity to a specific deployment target. They treat governance coverage as a lifecycle responsibility, not a development-phase checkbox. And they reduce oversight dependency over time by design, building toward operational autonomy rather than permanent committee review.

The diagnostic question is concrete: set your training completion data next to your list of production AI deployments and ask how many of the certified employees are doing work differently today than they were doing it before the certification. If the answer is unclear, or if the list of deployments is substantially shorter than the list of certifications, the constraint is operating. That gap is measurable against records your organization already holds, and it surfaces in an afternoon.

The Constraint Map identifies where in the lifecycle the discipline is missing, confirms the pattern against your own operating data, and points to the remediation architecture that closes the gap structurally. The Readiness Assessment converts that picture into an operating instrument: a portfolio of deployment targets with success criteria attached, a governance structure that is enforced rather than documented, and a training architecture built around application rather than awareness. Planning cycles that open from that footing are different in character from ones that open from a completion report. Leadership enters the board conversation with an audited picture of what was funded, what changed, and what it produced.

That is the shift that makes the CFO's question answerable. And it is available now.

The invitation

If the organization in the opening of this piece sounds familiar, you are seeing the constraint clearly. Most organizations recognize it once it is named precisely. Few have named it precisely enough to remove it.

Naming it is what the Readiness Assessment is built to do. It is a two-week engagement that produces an operating picture, an evaluation framework, and a conversation guide: the instruments that let a planning cycle open from readiness rather than exposure. It begins with a thirty-minute diagnostic scoping conversation that surfaces a specific constraint in your own function, verifiable against data you already hold.

If that is the conversation you are ready to have, we would welcome it.

michael@magruder.co · doug@magruder.co

· · ·

Next in The Observation Deck: The Governance Gap, what happens when the committee exists and the decisions do not.

Sources: Strategy-embedding and gen AI EBIT attribution figures from McKinsey & Company, "The State of AI in Early 2024." Governance structure and effectiveness figures from ADR Center, 2025 Enterprise AI Governance Survey. Job redesign figure from Deloitte, "The State of AI in the Enterprise, 2026." Infrastructure readiness contradiction from Drexel LeBow and Precisely, 2026. AI competency framework from UNESCO, 2024 AI Competency Framework for Teachers.

Magruder & Company · magruder.co · customercore.co

There is a deck in your organization right now that describes your AI strategy. It was presented to the board. It was polished for the quarterly review. And if you set it next to what your teams are actually doing with AI this week, the two would not recognize each other.

That distance has a name. It is the first of the constraints we map, and it is the most common one we find: strategy without validation. The strategy exists as a document, not as a working instrument. It was built to communicate intent, not to govern execution. And the gap between the two widens every quarter, while no one is measuring the distance.

This is not a story about a shortage of strategy. It is a story about strategy that cannot be checked against reality, which is a different and more expensive problem.

What the numbers say

The pattern is visible across the research, and the figures are not subtle.

Seventy-five percent of executives admit their company’s AI strategy is “more for show” than actual internal guidance. Only twenty-seven percent have fully embedded an AI strategy across their business units. And the human cost shows up in the same surveys: seventy-three percent of CEOs report stress or anxiety about their company’s AI strategy, and sixty-four percent say they fear losing their job if they fail to lead the transition.

Read those together and a picture forms. Most leaders know the strategy is performative. Most have not embedded it where the work happens. And the people who own it are carrying real anxiety about a document they privately suspect does not govern anything.

That is not a confidence problem. It is a validation problem.

Why it persists

The reason this constraint is so durable is that it is structural, not technical. No tool fixes it, because no tool created it.

A strategy becomes performative the moment it is built to be presented rather than to be run. It gets written for an audience, the board, the review, the all-hands, and an audience needs a clear story more than it needs a working instrument. So the strategy optimizes for clarity of intent. It says where the organization is going. What it almost never says is how anyone would know, this quarter, whether the organization is actually going there.

Without that second half, there is nothing to validate against. The strategy cannot be wrong, because it was never made measurable enough to be wrong. It simply drifts, quarter over quarter, further from the operations it was meant to direct, and because no one is measuring the distance, no one can say how far the drift has gone until something forces the question.

For the CEO, this is an operational hazard: decisions are being made under a strategy that does not describe the actual operation. For the board, it is a fiduciary blind spot: the quarterly update is built from metrics chosen to look like progress rather than to prove it.

How it becomes removable

This constraint is removable, and the work to remove it is concrete.

There is a quick way to test whether this constraint is operating in your own organization. Open your AI strategy document and check its last meaningful edit date against the dates of your last few AI purchases. If the spending moved and the strategy did not, the strategy is describing a company that no longer exists. That gap, between when the document was last governed and when the money was last committed, is the recognition signal. It is one of several the diagnostic looks for, and it is the kind of thing you can verify in an afternoon against records you already hold.

The move from there is to convert the strategy from a statement of intent into an instrument you can check against your own operation. That begins not with our framework but with your business: how the function actually works, where AI investment is already flowing, and what result each piece of that investment is supposed to produce. From that footing, the Constraint Map℠ names the specific impediment, confirms the recognition signal against your own experience, and points to the remediation pattern that closes the gap structurally rather than episodically.

A validated strategy changes the whole posture of the conversation. Instead of defending a deck in front of an increasingly skeptical board, leadership presents an audited picture: here is what we are funding, here is the result it is producing, here is how we know. Board reporting stops being damage control and becomes a demonstration of capital efficiency. The same investment that was a source of anxiety becomes something you can defend on an earnings call.

That shift is available now. It is the difference between an organization that enters its next planning cycle allocating from clarity and one that allocates from vendor pressure and instinct.

The invitation

If you recognized your own organization in any of this, you are seeing it clearly, and that clarity is the part that can be acted on. Most leaders arrive at the same recognition; few have named the constraint precisely enough to remove it.

Naming the constraint is what the Readiness Assessment is built to do. It is a two-week engagement that produces an operating picture, an evaluation framework, and a conversation guide, the things that let a planning cycle open from readiness rather than exposure. It begins with a thirty-minute diagnostic scoping conversation that surfaces a specific constraint in your own function, verifiable against data you already hold.

If that is the conversation you are ready to have, we would welcome it.

michael@magruder.co · doug@magruder.co

· · ·

Next in The Observation Deck: Competence Theater, the gap between training that reports high completion and capability that never reaches production.

Sources: “more for show,” CEO stress, and job-loss figures from Writer Inc. and Harris Poll, 2026 survey of approximately 900 CEOs. Strategy-embedding figure from Gartner CEO Survey, April 2026.

Magruder & Company · magruder.co · customercore.co

Executive Summary | Five Points of Orientation

The Core Premise. The market is flooded with AI advice, yet organizations remain paralyzed by invisible, internal constraints. True AI preparedness is not a matter of drafting better strategy decks. It is an execution problem.

The Reality Gap. Current enterprise AI deployment has devolved into performance art, backed by staggering metrics. 95% of companies fail to achieve measurable financial impact from generative AI. 88% of agent pilots never reach production, and the share of companies abandoning most of their AI initiatives surged from 17% to 42% in a single year. Spending is not the problem: only 15% of decision-makers report any EBITDA lift from the AI they have already bought.

The Five Constraints. The gap between AI investment and earnings is not one failure but five, each nameable and remediable. Strategy Without Validation: board decks built to communicate intent rather than govern execution. Competence Theater: training that yields completion rates and enthusiasm but not production capability, with 84% of companies never redesigning a single job to fit AI. The Governance Gap: 23% of organizations are running agentic AI somewhere, but only 2% have actually scaled it. Invisible Spend: 47% of shadow AI runs through personal accounts, entirely outside the enterprise’s view. The Accountability Vacuum: an agent is funded like technology but behaves like a worker, landing in the gap where ownership disappears.

The Bottom Line. Unvalidated AI is a fiduciary blind spot for the board and an operational hazard for the CEO. To turn structural AI investment into a reportable market advantage on earnings calls, leadership must abandon performative metrics and systematically map and remediate their actual execution constraints, the work the Constraint Map℠ is built to do.

· · ·

Every conversation about AI surfaces a different dimension of the opportunity: platforms that unlock new functions, governance frameworks that build trust, change programs that accelerate adoption. The organizations moving with the most confidence are the ones who recognize that coordinating those dimensions is not the challenge. Preparing for them is. Preparedness means holding the whole shape of the change at once, with a methodology designed for that purpose from the start.

What Magruder & Company brings to this moment is not a platform, not a framework adapted from adjacent work, and not a theory assembled in response to a trend. It is decades worth of alignment research, strategy design, and execution practice distilled into an adaptive system, one built specifically to prepare enterprise organizations for what is already arriving, while the conditions for thoughtful preparation still exist.

01 · The market has not failed to produce AI advice.

The market has not failed to produce AI advice. It has failed to produce AI preparedness. The distinction matters because organizations are not suffering from a shortage of vendors, frameworks, consultants, or platforms. They are suffering from a shortage of integrated thinking about what AI actually requires of an enterprise before it can compound rather than consume.

McKinsey’s 2025 State of AI report found that 88 percent of organizations now use AI in at least one business function. Yet only 6 percent of those organizations qualify as high performers, defined as generating 5 percent or more of EBIT from AI.¹ PwC’s 29th Global CEO Survey, published in 2026, found that only 12 percent of CEOs say AI has delivered both cost and revenue benefits. Fifty-six percent report no significant financial benefit at all.² These organizations are not failing at technology. They are failing at governance, at organizational design, at the preparedness that has to exist before the technology can produce what was promised.

The technology firms sell adoption. The management consultancies sell strategy. The IT advisors sell integration. The governance specialists sell policy. Each of these is a real discipline with real value inside its lane. But none of them integrates the full scope of the change. None of them was designed to. Their engagement models are built around the assumption that the organization will coordinate the pieces once the engagement ends. That assumption has always been the source of transformation failure, and AI has not changed it. It has accelerated the cost of it.

What is missing from the market is not another framework or another platform. What is missing is a living practice that begins where the organization actually is, understands how it actually operates across every functional vertical and leadership layer, and builds preparedness in context with the business rather than asking the business to adapt to the engagement model. A living practice that tracks research that shifts week to week, maintains continuity across every domain AI is reshaping, and develops best practices that do not yet exist in the marketplace.

Our practice does not exist as a product category. It exists as a discipline. It was built in the field over years of applied engagement, not assembled from traditional frameworks.

02 · The nature of the change.

The change is already in motion. It does not wait for readiness and it does not arrive as a single event requiring a single response. Every function will change. Every role will change. Every process, reporting structure, financial model, and people system will change. Not simultaneously, not as a mandate from above, but continuously, in waves, beginning now and accelerating through the next several years.

Two failure modes define the field. The first: the organization that reacts episodically, buying its way through each wave under vendor pressure and time constraints, compounding cost and confusion with every response. The second: the organization that prepared systematically, so that when change arrives it responds from readiness rather than exposure. This is Q2. The organizations that engage now will enter Q3 with a methodology already running. The ones that wait will spend Q3 and into 2027 paying the premium that unpreparedness always extracts: overspending on platforms before the governance exists to use them, on consultants before the internal knowledge exists to sustain what they build, on adoption programs before the organizational design exists to absorb them.

AI did not create the gap between what organizations know and what they can act on. AI is amplifying it, making whatever was already broken faster, more expensive, and harder to see. The governance that should have been in place before the first platform purchase was made is now being retrofitted under pressure. The organizational design that should have preceded the adoption program is now being discovered in the middle of it. The measurement frameworks that should connect AI investment to earnings quality are still being debated in rooms where the spending has already started.

The goal is not transformation as a parallel crisis layered on top of daily operations. It is the elimination of serial crisis creation. A prepared, adaptive posture that lives inside how the organization already works, not beside it as a second emergency.

03 · What is blocking you, by name.

Every organization carries constraints it cannot see in itself. Not because the people are incapable. Because the observer and the observed are the same.

The Constraint Map℠ diagnostic names those impediments. Not a vague readiness score. Not a maturity model with levels that affirm without directing. Named constraints, with recognition signals and remediation patterns. The five most common across the organizations we work with:

Strategy Without Validation. The organization has an AI strategy. It was presented to the board. It lives in a deck that was last updated for the quarterly review. And it has no measurable connection to what the organization is actually doing with AI day to day. The strategy exists as a document, not as a working instrument. It was built to communicate intent, not to govern execution. The gap between the strategy and the operations it was supposed to direct widens every quarter, and no one is measuring the distance.

• 75% of executives admit their company’s AI strategy is “more for show” than actual internal guidance³

• 95% of companies fail to achieve measurable financial impact from generative AI⁴

• Only 27% have fully embedded an AI strategy across business units⁵

• 73% of CEOs report stress or anxiety about their company’s AI strategy; 64% fear they could lose their job if they fail to lead the transition⁶

For the CEO, this gap is an operational hazard. For the board, it is a fiduciary blind spot. When an AI strategy cannot be validated, quarterly board updates inevitably descend into performative metrics. Conversely, an organization running a validated system converts preparedness into a reportable market advantage, giving leadership the empirical foundation required to defend structural investments on earnings calls and during strategic reviews.

Competence Theater. The organization invested in AI training. Completion rates are high. Leadership reports confidence that the workforce is ready. But the people doing the implementation tell a different story. The training produced awareness, not capability. The workshops produced enthusiasm, not competence. The pilots produced demos, not production systems. The organization is performing readiness rather than building it, and the distance between what leadership believes and what the front line experiences grows with every confident update.

• 82% of enterprise leaders provide AI training, yet 59% still report an AI skills gap⁷

• 90% of executives say their AI efforts are effective; only 39% of technical teams report meaningful impact⁸

• 88% of AI agent pilots never reach production, and the share of companies abandoning the majority of their AI initiatives surged from 17% to 42% year over year⁹, ²³

• Only 35% have a mature, organization-wide AI upskilling program despite 82% offering “some form” of training¹⁰

• 84% of companies have not redesigned jobs to fit AI, and the confidence is its own tell: 87% report having the necessary infrastructure while 42% name that same infrastructure as their biggest obstacle²⁴, ²⁵

The Governance Gap. AI is in production. The governance that should surround it is not. Agents are running, models are generating outputs, decisions are being shaped by systems that have no formal review cadence, no escalation logic, and no defined authority structure. The organization deployed faster than it governed, and the gap between deployment velocity and governance maturity is where risk accumulates. Not theoretical risk. Operational risk that is already producing incidents the organization may not yet recognize as AI-related.

• 78% of executives lack confidence their organization could pass an independent AI governance audit within 90 days¹¹

• 72% of firms are in production with agentic AI, yet 60% lack formal governance for those systems¹²

• 82% of enterprises have unknown AI agents running in their IT infrastructure¹³

• Only 19% have fully implemented AI governance frameworks; 67% believe their company has already suffered a data leak or breach from unapproved AI use¹⁴

• 23% of organizations are running agentic AI somewhere in the business; only 2% have actually scaled it²⁶

Invisible Spend. The CFO approved the AI budget. What the CFO did not approve is the shadow spending happening in every department that found its own tools, built its own workflows, and started its own experiments outside the governance perimeter. The visible AI investment is the number on the budget line. The actual AI investment includes every unauthorized tool, every untracked API call, every department-level subscription that never reached procurement. The difference between those two numbers is growing, and no measurement architecture exists to close it.

• Only 12% of CEOs say AI has delivered both cost and revenue benefits; 56% have seen no significant financial benefit at all¹⁵

• Only 14% of 200 U.S. finance chiefs report clear, measurable impact from AI investments¹⁶

• 80% of workers use unapproved AI tools, and 47% of that shadow AI usage runs through personal accounts, entirely outside the enterprise’s view; the average enterprise logs 223 data policy violations per month related to AI¹⁷, ²⁷

• 83% of CFOs plan to increase AI spending by 15% or more over the next two years, while AI costs have surged 108% from 2025; yet only 15% of AI decision-makers report an EBITDA lift from the spend¹⁸, ²⁸

The Accountability Vacuum. The organization created the title. Chief AI Officer. Head of AI Strategy. VP of AI Transformation. The titles multiplied. The accountability did not. No one owns the full surface area of what AI is doing inside the organization. The CIO owns the infrastructure. The CDO owns the data. The CHRO owns the workforce. The CFO owns the budget. And the gaps between those domains, the places where AI decisions fall between the chairs, are where the most consequential failures are forming. The accountability vacuum is not a missing person. It is a missing structure.

• 76% of organizations now have a Chief AI Officer, up from 26% in 2025. Titles surged. Accountability did not follow¹⁹

• COOs are discovering governance gaps that CFOs are not funding and CIOs/CTOs are not surfacing²⁰

• Only 1 in 5 organizations has tested a response plan for AI failures²¹

• AI systems are generating outputs that no one in the organization was authorized to act on²²

If you recognized your organization in any of those descriptions, you are not alone. Most do. The Constraint Map is the diagnostic that makes these visible, names them, and provides the remediation patterns that resolve them structurally rather than episodically.

04 · What our living practice delivers.

What our living practice delivers is not a report, a roadmap, or a platform license. It is not a fixed engagement that ends with a deck and a handoff. It is a continuous practice of AI preparedness that wraps around how your organization actually operates, and stays there. What changes as the practice runs is what your executive team can point to. Three outcomes mark that progression, and each one builds on what the last made possible.

Readiness Assessment. The first outcome is clarity about what is actually blocking you. The work begins with your business, not our instruments: we wrap AI literacy around how the function actually operates, in context, so the executive team sees its own operation differently before anything is diagnosed. From that footing, the Constraint Map℠ names the specific impediments operating between your AI strategy and its results, with recognition signals you can verify against your own experience and remediation patterns that direct the work that follows. The assessment locates the organization against Customer Core™, the ideal state the practice curates toward, and identifies where AI investment will close a gap rather than inherit and amplify one. Two weeks of focused work with the right people produces an operating picture, an evaluation framework, and a conversation guide. The executive team enters Q3 planning able to say what to fund, what to pause, and what to stop, in terms the board can follow.

Governed Corpus. The second outcome is an asset the organization did not have before. Through GenGov™, the practice curates and codifies what a function already knows into a governed corpus your AI systems can operate on. The corpus arrives in two forms at once: a narrative your people read and audit, and a machine-readable reference any agent you deploy can act on, both governed against the same source. This is the work of the following nine to twelve weeks. It is the difference between governance that exists as a policy document and governance that operates where decisions are made.

Capability Transfer. The third outcome is ownership. The corpus transfers to your team, and it transfers with H2AI™, the discipline that keeps AI accountable to human judgment after the practitioners step back. H2AI™ arrives as three things that are easy to confuse and important to separate. An assessment establishes where human-to-AI intermediation is actually required, the decisions consequential enough to demand a named human in the path. A playbook defines how your people stage that judgment against AI execution, decision by decision. A standing office, structured like a PMO, runs the intermediation as institutional practice rather than as consulting that leaves when the engagement does. Throughout, your people are not trained in the abstract. They build competence by running the cycle itself, which is what separates durable capability from the awareness that training programs usually produce.

What these outcomes build toward is Customer Core™, realized. Not a measurement report appended to the work, but the shape the organization takes once the practice has landed: an operating model organized around value streams rather than functional verticals, where each layer of AI investment connects to the next through the Seven Levels that carry raw data to earnings quality. That connection is the unbreakable thread from data to earnings, and it is the proof a CFO can audit. An organization that runs the practice through to that state is not better at coordinating AI initiatives. It has stopped needing to coordinate them, because they were built connected from the start.

The output is preparedness. Not the appearance of it. The kind that holds when the next wave arrives, because the methodology is already running and the organization already knows how to use it.

05 · The inflection point.

Every executive who needs this practice already knows they need it. Not in the language of AI preparedness or governance methodology. In the language of their own exposure. The CFO who cannot explain why the last major initiative did not deliver what was modeled. The CHRO who is fielding AI questions from every function and has no coherent answer that spans all of them. The CEO who is hearing the board ask harder questions about AI strategy every quarter while the internal answer gets less specific, not more. The CIO who is being asked to govern infrastructure decisions that were made before anyone understood what governing them required.

The inflection point is the moment when that private awareness becomes an actionable recognition. When the cost of waiting becomes more visible than the cost of starting. It is not a dramatic moment. It is usually quiet. A question that could not be answered. A number that could not be traced. A conversation that revealed how thin the architecture underneath the confidence actually is.

The shift after this realization changes the entire corporate posture. The executive team goes from playing defense in front of an increasingly skeptical Board of Directors to presenting an audited posture of risk control and capital efficiency. When the path to AI preparedness is structurally proven, board reporting transitions from an exercise in damage control to a strategic demonstration of market advantage.

The diagnostic conversation is designed to reach that moment precisely. It applies the Constraint Map’s recognition signals to the executive’s own function and surfaces a specific, named constraint the executive can verify against data they already possess. The gap between what the executive believes is governed and what actually is becomes visible within the conversation itself. That is not a sales technique. It is the same diagnostic methodology the practice applies at every subsequent stage.

What happens after the inflection point is not a crisis. It is a decision. The organization that reaches that moment with a practice already in the room has something no platform, no policy, and no single-discipline engagement can provide: a path that is already designed for them, already current with the research, already built in context with how they operate, and already proven by the same methodology applied to the practice itself.

The Q2 Imperative is not a sales argument. It is a timing observation. The organizations that will be prepared for what arrives in Q3 and into 2027 are making that decision now. The window between systematic preparation and episodic reaction is not permanently open. It closes one quarter at a time, and the cost of the wrong side of it compounds with every quarter it goes unaddressed.

Q3 opens July 1. That is when VP-level leaders revisit headcount and commit the AI initiatives that will run in Q4. The organizations that enter Q3 with a methodology already in place allocate from clarity. The ones that enter without one allocate from vendor pressure, competitive anxiety, and instinct.

Our readiness assessment exists for the executive who is ready to stop coordinating pieces and start building something that holds. The two-week engagement runs in June. It produces the operating picture, the evaluation framework, and the conversation guide that allow Q3 to open from a position of readiness rather than exposure. June is the window for that work.

Contact us below to begin your engagement. Our thirty-minute diagnostic scoping conversation will identify an initial constraint specific to your function and your operating context, serving as the basis for every decision that follows.

michael@magruder.co · doug@magruder.co

Magruder & Company · magruder.co · customercore.co

Sources

1. McKinsey & Company, The State of AI, 2025 edition. Figures cited: 88% of organizations deploying AI in at least one business function; 6% qualifying as high performers (defined as deriving 5% or more of EBIT from AI).

2. PwC, 29th Annual Global CEO Survey, published 2026. Figures cited: 12% of CEOs reporting AI has delivered both cost and revenue benefits; 56% reporting no significant financial benefit.

3. Writer Inc. and Harris Poll, 2026 survey of approximately 900 CEOs. Figure cited: 75% of executives admitting their company’s AI strategy is “more for show” than actual internal guidance.

4. MIT Project NANDA, The GenAI Divide: State of AI in Business 2025, July 2025. Figure cited: 95% of companies failing to achieve measurable financial impact from generative AI. Methodology: analysis of 300+ public initiatives, 52 organizational interviews, and surveys of 153 senior leaders.

5. Gartner, CEO Survey, April 2026. Figure cited: Only 27% of organizations have fully embedded an AI strategy across business units.

6. Writer Inc. and Harris Poll, 2026 survey of approximately 900 CEOs. Figures cited: 73% of CEOs reporting stress or anxiety; 64% fearing job loss. Same instrument as endnote 3.

7. DataCamp and YouGov, 2026 survey of 500+ enterprise leaders. Figures cited: 82% providing AI training; 59% still reporting an AI skills gap.

8. RapidScale (a Cox Business company), The Talent Gap, May 2026, survey of 259 IT professionals. Figures cited: 90% of executives say their AI efforts are effective; only 39% of technical teams report meaningful impact.

9. Anaconda and Forrester Research (primary source), replicated in a March 2026 survey of 650 enterprise technology leaders published by DigitalApplied. Figure cited: 88% of AI agent pilots never reaching production. URL: https://www.digitalapplied.com/blog/ai-agent-scaling-gap-march-2026-pilot-to-production.

10. DataCamp, 2026 enterprise survey. Figure cited: Only 35% have a mature, organization-wide AI upskilling program despite 82% offering some form of training.

11. Grant Thornton, 2026 survey of 950 C-suite leaders. Figure cited: 78% lacking confidence their organization could pass an independent AI governance audit within 90 days.

12. Agentic AI Institute, 2026. Figures cited: 72% in production with agentic AI; 60% lacking formal governance. URL: https://agenticaiinstitute.org/agentic-ai-enterprise-adoption-2026-governance-gap/.

13. Cloud Security Alliance, April 2026. Figure cited: 82% of enterprises have unknown AI agents running in their IT infrastructure.

14. Composite figure. Figure 1: Only 19% have fully implemented AI governance frameworks (McKinsey 2026). Figure 2: 67% believe their company has already suffered a data leak or breach from unapproved AI use (Writer 2026).

15. PwC, 29th Annual Global CEO Survey, published 2026. Figures repeated from endnote 2 in the context of Invisible Spend.

16. RGP, The AI Foundational Divide: From Ambition to Readiness, December 2025. Survey of 200 U.S. finance chiefs. Figure cited: Only 14% report clear, measurable impact from AI investments. URL: https://rgp.com/press/rgp-cfo-survey-shows-growing-divide-between-ai-ambition-and-ai-readiness/.

17. Composite figure. Figure 1: 80% of workers using unapproved AI tools (Cybersecurity Insiders, 2026). Figure 2: Average enterprise logging 223 data policy violations per month related to AI (Netskope, 2026).

18. Composite figure. Figure 1: 83% of CFOs planning to increase AI spending by 15% or more over the next two years (Bain & Company, 2026). Figure 2: AI costs surged 108% from 2025 (Zylo, 2026 SaaS Management Index).

19. IBM, 2026. Figures cited: 76% of organizations now have a Chief AI Officer, up from 26% in 2025.

20. Composite qualitative finding from Dataiku 2026 and Grant Thornton 2026. Claim: COOs are discovering governance gaps that CFOs are not funding and CIOs/CTOs are not surfacing.

21. Grant Thornton, 2026 (consistent with endnote 11 instrument). Figure cited: Only 1 in 5 organizations has tested a response plan for AI failures.

22. Composite qualitative finding (Grant Thornton and Dataiku 2026). Claim cited: AI systems are generating outputs that no one in the organization was authorized to act on.

23. S&P Global Market Intelligence (451 Research), 2025. Figure cited: share of companies abandoning the majority of their AI initiatives rose from 17% to 42% year over year.

24. Deloitte, Rethinking Operating Models for Humans and Agents, 2026. Figure cited: 84% of companies have not redesigned jobs to fit AI.

25. Drexel LeBow / Precisely, Data Integrity and AI Readiness, 2026. Figures cited: 87% report having necessary infrastructure while 42% name infrastructure as their biggest obstacle (the confidence-readiness paradox).

26. Capgemini Research Institute, The Rise of Agentic AI, 2025. Figure cited: only 2% of organizations have deployed AI agents at scale, against approximately 23% running agentic AI somewhere (McKinsey/Gartner). Corroborated by Camunda 2026 (11% of agentic use cases reached production).

27. ITPro / Netskope, 2026. Figure cited: 47% of shadow AI usage occurs via personal accounts.

28. Forrester, AI Predictions 2026. Figure cited: only 15% of AI decision-makers report an EBITDA lift from AI investment.

Magruder & Company · magruder.co · customercore.co

© 2026 Magruder & Company. All rights reserved.

Customer Core™, GenGov™, and H2AI™ are trademarks, and Constraint Map℠ is a service mark, of Magruder & Company.

v1.9.5 · 20260603