The Accountability Year

THE BRIEF

Companies are doubling their AI budgets in 2026. A BCG survey of 2,360 executives finds organizations plan to raise AI spending to 1.7% of revenue this year, up from 0.8% in 2025 — a doubling in twelve months. At the same time, PwC's 29th Annual Global CEO Survey of 4,454 chief executives across 95 countries found that 56% of CEOs reported neither revenue increases nor cost decreases from AI in the past year. Budgets are going up. Outcomes are not keeping pace. The gap between those two facts is where executive credibility lives and dies in 2026.

The board is not going to make this easier. A BCG survey of 625 business leaders — 351 CEOs and 274 board members — found that 61% of CEOs say their boards are "rushing" AI implementation, pushing for speed their CEOs believe the organization isn't ready to execute. Boards are reading the same headlines you are. They're not reading the implementation reality you're living. This tension is not going away; it's escalating. CEOs who cannot produce a credible answer to "what are we getting for this?" before the next board meeting are in a vulnerable position.

The data is beginning to separate winners from everyone else. PwC's survey found that only 12% of CEOs reported both revenue growth and cost reductions from AI — a group PwC calls the "vanguard." What distinguishes them is not budget size or model selection. It's that they embedded AI into high-complexity operations — demand generation, decision-making workflows, autonomous forecasting — rather than using it to summarize emails and generate first drafts. The 12% are not more sophisticated technologists. They made a different organizational bet.

The workforce math is starting to show up in real numbers. A working paper from the National Bureau of Economic Research, based on a survey of approximately 750 U.S. CFOs conducted by Duke University and the Federal Reserve Banks of Atlanta and Richmond, found 44% of CFOs planning AI-related workforce reductions in 2026 — translating to an aggregate 0.4% headcount cut, roughly 502,000 jobs economy-wide. That's nine times the AI-attributed layoffs recorded in 2025. The cuts are concentrated in large firms and routine-task roles. The CFOs not planning reductions are, in many cases, the ones who haven't yet figured out what to cut — not the ones getting more from the same headcount.

Block Inc. ran the experiment at scale. In February 2026, CEO Jack Dorsey described the company's decision to cut nearly half its workforce — approximately 4,000 employees — as a "deliberate and bold embrace of AI." CFO Amrita Ahuja characterized it as an "18-month leap" and called AI-driven job cuts "an inevitability for companies." The outcome, as of the company's guidance update: gross profit per employee is projected to roughly double to approximately $2 million in 2026, against a $12.2 billion gross profit target and a 54% increase in adjusted operating income. Block's stock rose 24% on the news. The market is already pricing the difference between organizations that are restructuring around AI and those that are adding AI on top of existing structures.

M&A is now a capability acquisition strategy, not a growth strategy. EY's January 2026 CEO Outlook survey found 53% of CEOs planning acquisitions for digitalization and productivity — up five percentage points year-over-year. Global M&A volume hit a record $1.2 trillion in Q1 2026, with AI consolidation as the primary driver, according to Wedbush. The companies being acquired are not revenue-stage businesses. They are proprietary data assets, AI agent frameworks, and vertical-specific models that would take years to build internally. If your organization doesn't have the capability it needs, the market is offering it — at a premium, and for a limited window before the best targets are gone.

---

---

---

THE SIGNAL

The tension that defines this moment: Boards want speed. CEOs know the organization isn't ready. And 56% of CEOs have nothing to show for last year's investment.

This is not a technology problem. Every major AI capability is accessible via API. Foundation models are commodities. The tools are not the constraint. The constraint is whether your organization is structured to extract value from them — and most are not.

The companies winning in 2026 share a specific pattern. They didn't ask "how do we use AI?" They asked "what work, if restructured around AI, would produce a materially different outcome?" Block cut half its workforce not because it had the best AI tools in fintech, but because it identified the organizational structure that AI made possible and built to it. Microsoft's AI revenue run rate reached $37 billion, up 123% year-over-year, according to Q3 FY2026 earnings — not because Microsoft has better models than its competitors, but because it has distribution, integration depth, and enterprise relationships that turned capability into contract value.

The companies losing are following a different pattern. They ran pilots. The pilots succeeded on their own terms. The pilots didn't scale because scaling required changing how departments worked, who owned what, and what middle management did — and none of that changed. Gartner's April 2026 CEO survey found 80% of CEOs expect AI to force high or medium operational capability overhauls. That number tells you something important: the overhaul is acknowledged. What it doesn't tell you is whether the acknowledgment is driving action or just producing more planning.

The board-CEO divide makes this worse. BCG's survey found that boards — particularly those with lower AI confidence — are twice as likely to push for faster adoption. They're responding to competitive anxiety and analyst pressure, not operational reality. The CEO who goes along with board-driven acceleration without the organizational foundation to support it is setting up the next failure. The CEO who pushes back without a credible alternative is losing board confidence. The right position is neither capitulation nor resistance — it's a clear, metrics-backed answer to what success looks like, when it arrives, and what needs to change internally to get there. That answer does not currently exist in most organizations.

The competitive stakes have a timeline. Gartner projects that 40% of enterprise applications will embed task-specific AI agents by the end of 2026, up from less than 5% at the end of 2025. That is not a gradual shift — it is a structural compression of the window in which organizations can catch up. The companies embedding agents into core workflows now are not just ahead on productivity. They are generating proprietary operational data at a rate their lagging competitors cannot replicate through spending alone. Data advantage compounds. Speed advantage does not.

The organizations that emerge from the 2026 trough as durable winners will have made a specific set of decisions: they chose one or two mission-critical workflows to restructure completely rather than dozens of workflows to augment partially; they built the measurement infrastructure to show P&L impact before the board asked for it; and they moved on workforce structure rather than waiting for more certainty. The ones still in the trough at the end of 2026 will have done the opposite — run more pilots, added more AI features to existing tools, and deferred the organizational decisions until the ROI was clearer. The ROI will not be clearer until after the organizational decisions are made. That is the loop most enterprises are trapped in.

---

---

THE DEEP DIVE

Thesis: The 56% of organizations reporting zero AI return are not failing because of their technology choices. They are failing because they never changed how work gets done.

The tools are not the variable. In 2026, every large enterprise has access to the same foundation models, the same agentic frameworks, the same cloud AI infrastructure. What separates the 12% who are generating measurable returns — PwC's "vanguard" — from the majority is not AI selection. It is organizational redesign.

Here is the decision framework every CEO needs to apply before the next budget cycle:

The Capacity vs. Tools Distinction

AI tools augment existing work. AI capacity restructures what work exists. An organization that deploys a writing assistant to its marketing team has added a tool. An organization that restructures its content operation — eliminating the role of first-draft producer, shifting human time entirely to strategy, editing, and distribution decisions — has built capacity. The first produces marginal efficiency. The second produces a structural cost advantage that compounds over time.

Most organizations are buying tools. The vanguard is building capacity. The way to tell the difference: if your AI investment could be removed tomorrow with three months of adjustment to return to prior operating norms, you have tools. If removing it would require a fundamental restructuring of how work is organized, you have capacity.

Why the Pilot Trap Persists

The pattern is consistent across industries. A pilot is scoped for success. The pilot succeeds. The pilot doesn't scale. The reason is nearly always the same: the pilot was designed to prove AI could do something, not to prove the organization could restructure around it. Scaling AI is not a technology problem — it is a change management, incentive, and workflow design problem. It requires changing what middle management does. It requires redefining what outputs are valued. It requires accepting a transition period where productivity temporarily declines before it improves. None of those things are in a pilot scope.

The NBER/Duke/Fed CFO survey is instructive here. The 44% of CFOs planning AI-related workforce reductions in 2026 are, in most cases, not the same companies as the 56% stuck at zero ROI. The workforce decisions and the ROI gap are two separate populations. The companies cutting headcount tied to AI have already made the organizational decisions. The companies still at zero ROI are largely still running pilots and waiting for clarity that will not arrive on its own.

The Four Failure Modes

Failure Mode 1: Tooling without redesign. AI is deployed to existing roles doing existing work. Per-task time decreases. Output volume increases. Headcount stays flat. Costs decrease marginally. Revenue impact is zero because the work being accelerated wasn't the bottleneck.

Failure Mode 2: Pilot sprawl. Multiple departments are running AI experiments simultaneously. Each reports success metrics that are real but not comparable and not aggregated into P&L impact. The board asks for a number. The answer is a collection of percentages that don't add up to a dollar figure. This is the most common failure mode in enterprise AI right now.

Failure Mode 3: Governance delay. Legal, compliance, and IT have concerns. Those concerns are real. The response is a governance framework that, by the time it is approved, has delayed deployment by nine to eighteen months. Competitors who moved with managed risk are already on the other side of the learning curve.

Failure Mode 4: Wrong bottleneck targeting. AI is deployed to reduce costs in functions where costs are not the primary driver of competitive position. Customer service efficiency is improved while product development cycles, where AI could compress time-to-market by months, are untouched. The effort is real. The strategic leverage is absent.

The Framework for Getting Out

The CEO question is not "where should we use AI?" It is: "What is the work, that if restructured around AI, would change our competitive position in the next twelve months?" That framing has a short list. It is never longer than three things. Pick the one with the largest P&L surface area and redesign it completely — not augment it. Assign it an owner with budget authority and a P&L accountability line. Set a six-month timeline. Define the measurement criteria before the work begins, not after.

Everything else is sequenced behind that. The organizations winning in 2026 are not running more AI initiatives than their competitors. They are running fewer, deeper ones.

The Consequence

Gartner projects that by 2028, AI agents will participate in 90% of $15 trillion in B2B procurement interactions. The organizations that have embedded agentic AI into their commercial workflows by end-2026 will have operational data and customer interaction history that new entrants cannot replicate with a larger budget. Data advantage in AI compounds in a way that financial advantage alone does not. The window to build that advantage is this year. Not 2027.

The organizations still optimizing their governance frameworks in Q4 2026 will not be locked out of AI. They will be locked out of the data moats that early movers are building right now. That is a different and more durable problem.

---

---

THE PLAYBOOK

CEO / COO

• Stop measuring AI success by adoption metrics and start measuring it by P&L delta — because boards will stop accepting "X% of employees are using AI tools" as evidence of strategic progress within the next two quarters, and you need a different answer ready before they ask.

• Identify the one or two workflows where AI can restructure — not augment — your cost or revenue position, assign an owner with budget authority and a hard six-month deadline, and publicly commit to the outcome — because the accountability gap between AI spend and measurable return is where your board credibility is currently eroding, and internal commitment is the only mechanism that closes it.

• Build your board answer now: a one-page summary of what AI spending has produced in P&L terms, what the next twelve months will produce, and what organizational decisions are required to get there — because the 61% of boards already pushing for faster AI adoption are not going to get more patient, and the CEO who shows up to the next meeting with a coherent plan owns the room.

CFO

• Require every AI project above a materiality threshold to carry a committed P&L impact figure, a named owner, and a measurement date before it receives budget — because the NBER/Duke CFO survey shows 44% of your peers are already planning workforce reductions tied to AI outcomes, and organizations that cannot connect AI spend to financial metrics are falling behind peers who can.

• Model the workforce math explicitly: identify which roles in your organization are performing routine, high-volume, low-judgment work at above-market cost relative to AI alternatives, and bring that analysis to the CEO before the board asks for it — because the 502,000 projected AI-related job eliminations in 2026 are concentrated in exactly the cost centers that show up on your balance sheet, and your competitors' CFOs already have this model.

• Evaluate the M&A opportunity with urgency: the AI-native targets that carry defensible moats — proprietary datasets, vertical-specific models, embedded customer workflows — are trading at 8–35x revenue and are being acquired now; a deliberate decision not to pursue M&A is a legitimate strategic choice, but it needs to be deliberate and documented before the window closes.

Department Lead / AI Initiative Owner

• Kill the pilots that have been running for more than six months without a path to P&L impact — because pilot sprawl is the most common reason AI investments produce impressive internal metrics and zero board-visible outcomes, and every resource tied up in a stalled pilot is unavailable for the redesign work that actually moves the needle.

• Build the measurement framework before you start the next initiative, not after: define the baseline metric, the target metric, the timeline, and the dollar value of the improvement in the first week of the project — because your CFO will ask for this number and the answer cannot be "we'll know when we see it."

• Bring the organizational change ask to your CEO explicitly and early: the work that matters most requires changing roles, reporting structures, or incentive systems, and that cannot be approved by you — so the project plan needs to include the change management ask and the executive sponsor commitment as prerequisites, not dependencies to be resolved later.

---

---

THE NUMBERS

—

The callout:

—

---

WHAT'S NEXT + WHAT'S COMING

The forward signal this week is the Board-CEO divide hardening into a governance crisis. BCG's survey — released this month — found that boards less confident in their own AI knowledge are twice as likely to demand faster adoption, and more than half of CEOs believe boards are overestimating AI based on headline coverage rather than operational reality. This tension is appearing simultaneously in practitioner forums, investor calls, and trade press, which is the multi-channel pattern that precedes a structural shift. Watch for the first high-profile CEO departure attributed to board-level AI pressure — it is being discussed in executive circles and is the kind of event that resets board expectations across industries.

One thing to watch before next Tuesday: Q1 2026 earnings calls from Block (reporting May 7) and any CFO commentary that quantifies AI labor-cost outcomes in dollar terms, not productivity percentages. The first time a CFO says "AI reduced our COGS by $X million" on an earnings call — with attribution and methodology — the board conversation changes permanently.

M&A to watch: Google's $32B Wiz acquisition closed March 11, establishing multicloud AI security as a foundational enterprise layer. ServiceNow's acquisition of Moveworks (~$3B) and Workday's acquisition of Sana ($1.1B) signal that workflow AI is now an acquisition-stage market, not a build-or-evaluate-stage market. The window for acqui-hire and smaller vertical AI targets remains open but is compressing.

Upcoming: Block Q1 2026 earnings (May 7). OpenAI developer conference signals on enterprise agent pricing expected in May. EY's next CEO Outlook installment, expected Q2 2026, will likely quantify the gap between AI investment intentions and outcomes in more granular sector-level terms.

---

This report was produced with AI assistance and human editorial review.

---

This report was produced with AI assistance and human editorial review.