The C-Suite Said Yes to the AI Platform and No to Everything the Platform Needs to Work

Quick answer

Enterprise AI projects fail at high rates, and the organizational conditions surrounding the tool account for more of those failures than the technology itself does. The C-suite approves the platform. It does not always approve what the platform requires: a named owner, a governance decision, and protected time from the people whose process is being changed. In retail operations, where peak inventory cycles run on the same calendar as AI training schedules, the person left holding that gap is almost always the VP-level leader who was already over-functioning before the rollout was announced.

TL;DR

The data on enterprise AI outcomes in 2025 and 2026 converges on the same finding, regardless of the research institution doing the measuring.

• Most enterprise AI investments fall short of intended business value

  RAND Corporation's analysis of more than 2,400 enterprise AI initiatives found this figure, at roughly twice the failure rate of conventional IT projects. Of that 80.3%, one-third are abandoned before production, 28% reach production and deliver no value, and 18% run but never recoup their costs. (RAND Corporation, 2025)

• Organizational and leadership conditions account for the majority of AI project failures, ahead of technology factors

  Across multiple research bodies reviewing AI deployments in 2024 and 2025, the recurring findings point to the same gaps: projects launch without measurable success criteria, without adequate investment in the data and process foundations the tool requires, and without sustained C-suite sponsorship past the first quarter. The technology itself is rarely where the project first breaks down. The surrounding organizational conditions break down first. (McKinsey Global Survey, 2024; IBM Institute for Business Value, 2024)

• Comprehensive AI governance frameworks remain rare across enterprises

  Seventy-two percent of organizations had adopted AI in at least one business function as of 2024, up from 55% the year before, according to McKinsey's annual State of AI survey. (McKinsey, 2024) Only 21% of those organizations have a mature governance framework to match. (Deloitte, 2026) The distance between deployment and governance is where implementations go to die.

• Most AI projects stall before they scale past the pilot stage (IBM Institute for Business Value, 2024)

  The announcement is not the commitment. Executive energy concentrates at launch and dissipates during implementation, exactly when the VP-level leader needs decisions, budget adjustments, and air cover the most. (RAND / McKinsey synthesis, 2025-2026)

• Knowledge workers lose the equivalent of 51 working days per year to technology friction rather than skilled work (WalkMe State of Digital Adoption, 2026)

  Knowledge workers lose the equivalent of 51 working days per year to technology friction rather than skilled work, equal to 7.9 hours a week, per person, burned navigating tools that were deployed without the training, workflow redesign, or protected adoption time those tools require. (WalkMe State of Digital Adoption, 2026)

• Fewer than one in three retail AI initiatives reach full deployment

  Retail underperforms the cross-industry average on AI ROI while simultaneously facing the tightest implementation window: peak season inventory cycles, seasonal labor constraints, and a C-suite that wants production-ready AI before Q4. (RAND synthesis, 2025-2026)

9:15am Monday: what she already knows that the room does not

The CTO finishes the all-hands. The room applauds. The platform has been approved, the vendor selected, and the timeline shared with the board. She is already walking back to her desk doing the math that no one on stage did. Her team's names are on the project plan. No one has been given dedicated hours. The change management budget is zero. Training is scheduled for a week that lands inside the peak inventory cycle. And the governance question, who owns this, who makes the call when the tool surfaces a recommendation the merchandising team disputes, who handles the escalation when the process the AI is running against turns out to be undocumented, has not been asked, let alone answered. She opens her laptop and starts drafting a project plan she was never asked to write. This is where the implementation actually begins. At her desk. With her own time. On top of everything else she was already carrying. IBM's Institute for Business Value found in 2024 that the majority of AI projects stall or fail to scale past the pilot stage, and cited leadership and organizational factors as the primary cause in most of those cases. (IBM Institute for Business Value, 2024) Leadership did not fail to care. Leadership failed to fund the conditions. And someone else picked up the tab.

Two decisions that almost always get collapsed into one

Approving a tool is a procurement decision. It requires a vendor, a contract, a line item in the technology budget, and a go-live date. The C-suite is very good at this decision. It is visible, it has a deliverable, and it produces an announcement. Approving the organizational change required to use the tool is a different decision entirely. It requires naming an owner with actual authority over the process being changed. It requires a governance decision: who decides when the AI recommendation is wrong, who holds the data quality standard, who can pause the rollout if the first training cohort reveals that the underlying process is too undocumented to automate. It requires protected time, carved out of the schedules of the people whose work is being changed, before the change happens, not after the tool is already in production. These two decisions require different things from the room above her. Procurement requires a budget approval. Organizational change requires sustained executive attention, accountability transfer, and a willingness to absorb the short-term productivity dip that every implementation creates. C-suite leaders are more than twice as likely to say employee readiness is a barrier to AI adoption as they are to examine their own role in creating the conditions for adoption, according to McKinsey's Superagency in the Workplace report. The data shows employees are not the bottleneck. What is missing is a structural mechanism to convert ready employees into named, supported, accountable owners of the change. (McKinsey, 2025) The two decisions get collapsed into one because the second one is harder to make and harder to announce.

What a named owner actually means, and why committees are not the answer

When no single person owns the AI implementation, every decision becomes a meeting. Every escalation becomes a coordination problem. Every question about the tool from a confused team member lands on whoever picks up the phone first, which is usually the person who was already answering every other question. PwC's 2025 AI Business Survey found that while responsible AI has moved closer to the business in many organizations, governance structures frequently remain committee-based and slow to resolve operational decisions, creating a gap between the accountability organizations claim on paper and the authority any single person actually holds when the implementation hits a wall. (PwC, 2025) Maturing organizations resolve this by naming the split of responsibilities so that the person running the implementation has the authority to move, and a clear path for what happens when she cannot. Only 21% of companies have a mature governance model for autonomous AI agents, even as adoption accelerates. (Deloitte, 2026) In retail operations, where an AI recommendation about inventory replenishment or workforce scheduling has a direct P&L consequence within 24 to 72 hours, the absence of a named decision-maker is a risk the VP-level leader absorbs personally, in real time, while also trying to hit her own performance targets. A governance decision names a person, defines their authority, and specifies the escalation path. Without those three things, the implementation does not stall. It runs, badly, on the judgment and bandwidth of whoever is closest to it.

Protected time: the condition everyone agrees is necessary and no one provides

Every implementation framework recommends it. Dedicate time for the people whose process is being changed to learn the tool, redesign their workflow around it, and surface the friction before it becomes resistance. Eight to sixteen hours for roles with significant workflow changes is a standard recommendation across AI change management literature. (Opsio, 2026) In practice, those hours come from somewhere, and in a lean retail operations team during a peak inventory cycle, they do not exist. The people whose process is being changed are already at capacity. They are buried under the work the tool was brought in to help them with, and they have no protected space to learn it. Workers who cannot see how a tool fits their real tasks default to the workflow they trust. (WalkMe, 2026) A rational response to an implementation launched without the organizational conditions that would make adoption possible. Organizations that pair executive sponsorship and role-specific workflow guidance with dedicated adoption channels see tool activation rates roughly double those of organizations that skip those conditions, within the same 90-day window. (Brainstorm analysis, 2026) The tool is identical. The organizational setup differs. That gap is the cost of treating protected time as optional. In retail, the timing problem is acute. Q4 planning begins in August. Peak inventory cycles run September through January. Any AI platform announced in Q3 with a Q4 go-live is asking the operations team to learn a new system during the highest-stakes, lowest-slack period of the year. The people most affected by that decision are the ones whose names are already on the project plan.

The retail-specific compounding factor

Active AI deployment in retail reached 58% in early 2026, a 16-point jump in a single year, according to NVIDIA's State of AI in Retail and CPG report. (NVIDIA, 2026) Nearly 68% of retailers expect to deploy agentic AI for key operational activities within the next 12 to 24 months. (Deloitte, 2026 Global Retail Industry Outlook) A large majority of retail executives also expect AI-driven personalization capabilities to be in place within the next year, as part of the same Deloitte survey findings. (Deloitte, 2026 Global Retail Industry Outlook) The pressure to deploy is quarterly. Board-driven. And it lands on an operations function that is already running a lean team through a peak cycle. Roughly three in four retail AI initiatives do not deliver what they promised, by the RAND synthesis of implementation outcomes across the sector. (RAND synthesis, 2025-2026) The investments that succeed share a consistent set of conditions: clear pre-approval metrics, sustained executive sponsorship throughout the implementation, protected time for adoption, and a named owner with authority over the process being changed. (RAND / MIT Project NANDA synthesis, 2025-2026) None of those conditions are technology requirements. Every one of them is an organizational decision the C-suite makes once, at launch, or fails to make, and then delegates to someone who does not have the authority to require them. The VP-level operations leader in retail is managing a peak inventory cycle, a platform rollout, a team that has zero protected hours, and a governance structure that does not exist yet. She is also the person most likely to absorb the fallout when the first training session lands on a week her team cannot attend.

The under-appreciated point: shadow AI is the symptom, not the problem

70% of knowledge workers are already using AI tools outside official company policy. (LinkedIn / Microsoft Work Trend Index, 2025) One in five organizations experienced a breach involving shadow AI in 2025, at an average additional cost of $670,000 per incident. (IBM Cost of a Data Breach, 2025) The standard response is a governance crackdown: policy documents, usage restrictions, audit trails. That response treats shadow AI as a compliance failure. Shadow AI is also an information signal. When people reach for unauthorized tools, the authorized tool does not fit how they actually work. The process is too undocumented to automate cleanly. The training did not account for the real edge cases. The workflow the tool was built around exists on a slide deck, not in how the team actually runs the day. Shutting down shadow AI without addressing what drove people to it produces a temporary measurement improvement and a permanent trust problem. The team stops using the unauthorized tool and stops surfacing the friction that led them there, and the VP-level leader loses her most reliable signal about where the implementation is actually struggling. The contrarian position here: shadow AI inside a retail operations team is often more useful as a diagnostic than as a compliance problem. What tools did they reach for? What process did they apply them to? Those answers tell you where the official implementation failed to meet the team where they are.

The burnout lens: who carries the gap between funded and sanctioned

There is a specific kind of exhaustion that comes from being responsible for outcomes you do not have the authority to produce. The VP-level leader owns the AI implementation result. She does not own the governance decision. She does not own the ability to carve protected time out of her team's calendar without pushback from the C-suite that added the project without removing anything else. She does not own the escalation path when the tool surfaces a recommendation the CFO disagrees with. She owns the project plan she wrote on her own time. She owns the questions she fields when the training lands wrong. She owns the gap between what was funded and what was actually sanctioned. This is the specific mechanism through which AI implementations produce burnout in the people running them. The volume of work was already high. What accumulates is the weight of decisions that are hers to manage but not hers to make, over and over, across a timeline that keeps shifting because the executive sponsor has moved on to the next announcement. 51% of organizations reported at least one negative consequence from AI use, yet only 28% said their CEO takes direct responsibility for AI governance, and just 17% said their board does. (McKinsey State of AI, 2025) The consequences land on the operations layer. The accountability sits nowhere. The person standing closest to both is the one who was already over-functioning before the platform was approved. Bracing for a month like this one, the way she has been bracing for it since 9:15 this morning, is a choice. She made it before the month started. The C-suite made its choice at the vendor selection meeting. Neither of those choices was inevitable.

What could go wrong

• The governance conversation gets deferred to implementation

  Governance decisions made during a live implementation are made under pressure, without the clarity that comes from deciding before anything is running. The questions that feel theoretical at procurement become urgent at week six, and the VP-level leader answers them alone.

• Protected time gets approved on paper and canceled in practice

  A training block on the calendar does not survive a peak inventory week. If the protected time is not defended by the person above the VP-level leader, it disappears, and adoption data becomes meaningless because adoption never actually happened.

• The named owner has the title but not the authority

  Naming an owner without giving that person the authority to make process decisions, hold the governance standard, or say no to scope creep produces a coordinator, not an owner. The gap fills itself, the same way it always does, with whoever is already carrying the most.

• The implementation timeline ignores the retail calendar

  A Q3 announcement with a Q4 go-live asks a retail operations team to learn a new system during their highest-stakes, lowest-slack period. The operations leader sees this immediately. The room above her typically does not, because the sequencing looks clean on a slide and brutal on a calendar.

• Shadow AI gets treated as a compliance problem rather than a diagnostic

  Shutting down unauthorized tool use without investigating what drove it removes the clearest signal the implementation has about where the official rollout failed to meet the team where they are.

• The VP-level leader self-funds the gap until she cannot

  She writes the project plan she was not asked to write. She answers the governance questions she was not given authority to answer. She reschedules the training sessions. She covers. And then she is depleted, the implementation stalls anyway, and the C-suite asks what happened.

The J.Caresse point of view

I have sat in enough post-mortems on failed AI implementations to know what the problem report says: poor adoption, insufficient training, resistance from the team. That language is almost always accurate and almost always incomplete. It describes what happened at the team level without describing what was decided, or not decided, at the level above. The tool did not fail. The conditions were never created. And the person whose performance review will mention the rollout is the one who drafted the project plan at 9:15 on a Monday morning, before anyone asked her to. The three things that make an AI implementation survivable at the VP level are not complicated: a named owner with real authority, a governance decision made before anything is running, and protected time that is defended by the person above her when peak season tries to take it back. Those three conditions are not technology requirements. They are organizational commitments. They belong in the approval conversation, not in the retrospective. If you are walking into that approval conversation, or preparing to have it with the room above you, the AI Governance Readiness Checklist is the one-page diagnostic that tells you which of those conditions are missing before the project plan exists. Request it directly from me on LinkedIn.

Key takeaways

For the VP-level retail operations leader navigating an AI rollout that was approved without everything approval requires.

• The tool approval and the organizational approval are different decisions

  By the RAND Corporation's 2025 synthesis of enterprise AI outcomes, roughly four in five large AI projects fail to deliver their intended value, and the dominant cause traces to leadership decisions rather than technology performance. Approving the platform and approving the conditions the platform requires to work are two separate decisions, and organizations routinely make only the first one. (RAND Corporation, 2025)

• Governance is a named person with named authority, not a committee

  Only 21% of companies have a mature governance model even as deployment accelerates. (Deloitte, 2026) A governance committee that reviews every AI decision is a bottleneck. A named owner with the authority to make process decisions and hold the escalation path is the condition the implementation actually needs.

• Protected time is a structural commitment, not a calendar note

  A 27% tool activation gap exists between implementations with structured adoption support and those without it, on identical tooling. (Brainstorm analysis, 2026) Protected time that is not defended by the level above the implementation owner disappears the first time a peak cycle conflicts with the training block.

• Shadow AI inside your team is a diagnostic, not only a compliance failure

  70% of knowledge workers who use AI are bringing their own tools to work, outside official policy. (Microsoft/LinkedIn Work Trend Index, 2025) What they reached for and what they applied it to tells you exactly where the official implementation failed to fit how the work actually runs.

• The retail calendar and the AI implementation calendar are in direct conflict in Q3 and Q4

  Active retail AI deployment is up 16 points in a single year. (NVIDIA, 2026) The pace of deployment has not been matched by a renegotiation of peak-season timelines. A rollout announced in Q3 with a Q4 go-live is asking the operations team to learn under maximum load.

• The person carrying the gap between funded and sanctioned is already over-functioning

  AI projects routinely lose C-suite sponsorship before the implementation has reached the point where it needs real decisions. That fade happens at exactly the wrong moment. The VP-level leader fills the gap. The question is whether she fills it with authority and support behind her, or fills it alone.