When the Business Owns the AI Roadmap and IT Owns the Risk, Nobody Owns the Outcome

Quick answer

I have seen this pattern across enough retail and consumer engagements to say it plainly: the business function sponsors the AI roadmap, IT carries the compliance and data governance obligations, and nobody owns the outcome. The gap between them lands on whoever sits in the middle, typically a VP or SVP of operations, merchandising, or supply chain. Her job becomes translation: upward to the C-suite explaining delays she was not informed of, downward to her team explaining why the timeline they were given no longer holds. That translation work is invisible, it accumulates, and it is a consistent driver of the burnout pattern that stalls implementations before they reach production.

TL;DR

Six findings from current research that frame the rest of this piece:

• AI proofs of concept fail at high rates, and the failure is rarely technical.

  RAND Corporation's 2025 analysis of more than 2,400 enterprise AI initiatives puts the failure rate at 80.3%, roughly twice the rate of comparable non-AI IT projects.

• In most retail organizations, the person accountable for AI results does not own the AI strategy.

  Deloitte's 2026 State of AI in Retail and CPG survey of 200 executives found that 54% of AI strategy ownership sits with technology leaders, not the business leaders accountable for delivering results from it (Deloitte, 2026).

• Governance concerns, not model capability, are what delay deployments.

  AvePoint's State of AI 2026 Report found that 86.9% of organizations delayed generative AI deployments by nearly six months. Data security and governance concerns were the leading cause (AvePoint, 2026).

• Fewer than one in ten organizations have embedded AI governance into their development pipelines.

  Trustmarque's AI Governance Index 2025, based on a survey of 507 senior UK IT decision-makers, found that fewer than one in ten organizations have integrated AI risk and compliance reviews directly into development pipelines, despite 93% using AI in some capacity (Trustmarque, 2025).

• Manager engagement has dropped sharply during the AI transition, with the steepest decline of any employee group.

  Gallup's 2026 State of the Global Workplace report found that manager engagement fell from 31% in 2022 to 22% in 2025, the steepest decline of any employee group, accelerated by the load AI transition has placed on the middle layer (Gallup, 2026).

• Retail AI conviction is running well ahead of organizational capability.

  Deloitte's 2026 State of AI in Retail and CPG survey defines this as the sector's 'say-do gap': wide AI adoption never exceeds 36% outside of IT, even as leadership names AI a top strategic priority (Deloitte, 2026).

What the structural conflict actually looks like on a Tuesday afternoon

She is writing her fourth explanatory email of the week. This one goes upward, to the SVP, explaining why the supplier integration the business committed to in Q3 planning is delayed. The reason: IT flagged a data governance issue she was not told about until last Thursday. She has already written the version that goes downward, to her team, explaining why the timeline they were given no longer holds. Neither email names what actually happened. She told herself Monday she would schedule a one-on-one with the IT director and ask directly why the flag came so late. She did not schedule it. She wrote the emails instead. That substitution is worth sitting with, because it is not a time management problem. She chose the emails. Emails are concrete, completable, and proof that she handled something. The one-on-one is a different kind of work: open-ended, possibly uncomfortable, and with no guarantee that the IT director will tell her what she actually needs to know. So she produced four artifacts that demonstrate competence and deferred the one conversation that could change the structure she is operating inside. The rest of this brief is about what that structure looks like, what it costs, and what the research says about how rarely it gets fixed. But the Tuesday at 3:15 PM is the thread. Every data point below is something she is living inside, right now, because nobody resolved the conflict before Q3 planning closed.

Why retail concentrates the problem

Retail data is fragmented by design. Point-of-sale systems, e-commerce platforms, distributor portals, warehouse management systems, and supplier reconciliation tools were built at different times, often by different vendors, and frequently in incompatible formats. Integrating AI into that environment requires architectural changes. For large retailers, those changes take twelve to twenty-four months (Harvard Business Review, as cited in retail AI implementation analyses, 2026). Deloitte's 2026 Retail Industry Global Outlook, drawn from 330 retail executives surveyed in late 2025, found that 44% of respondents say legacy systems are actively slowing innovation. The slowdown lives in governance: who decides which systems can be touched, at what pace, and who carries the compliance exposure when data moves across them. The Deloitte 2026 State of AI in Retail and CPG survey of 200 executives found that 54% of AI strategy ownership sits with technology leaders rather than the P&L owners who have to deliver business results from it (Deloitte, 2026). When the person accountable for the outcome does not own the strategy, and the person who owns the strategy does not carry the P&L consequence, the natural point of resolution is whoever is fielding questions from both directions. That is what she is on a Tuesday at 3:15 PM. The natural point of resolution. The one who handles it. And because she handles it, the structure that put her there never has to change.

The governance gap in numbers

The governance problem is not a perception gap. It has a measurable structure. AvePoint's State of AI 2026 Report found that 86.9% of organizations delayed generative AI deployments by nearly six months on average, and 86% delayed AI agent deployments by a comparable period. The cause in both cases was data security and data management concerns, not model capability. More than half of organizations worldwide reported delaying or canceling more than six AI projects in the past year because of governance, compliance, or regulatory issues (Security Brief Asia, citing AvePoint research, 2025). Canceled rollouts rose from 31.7% to 40.7% year over year in the same study. That figure includes projects that cleared Q-planning, received budget approval, were communicated to the team, and then stalled when IT raised a flag that the business function was not positioned to address. The Q3 commitment was made before anyone asked IT what was in its governance queue. Trustmarque's AI Governance Index 2025, based on 507 senior UK IT decision-makers, found that only 7% of organizations have fully embedded AI governance, even though 93% are using AI in some capacity (Trustmarque, 2025). IBM's 2025 Cost of a Data Breach Report found that 97% of organizations that experienced AI-related security breaches lacked proper AI access controls, and 63% had no AI governance policy at the time of the incident. Grant Thornton's 2026 AI Impact Survey added the executive layer: 78% of business executives lack strong confidence that they could pass an independent AI governance audit within 90 days. They are scaling AI. They are not governing it. And the flag that surfaces at the VP level on a Thursday afternoon is the physical form that gap takes.

The translator role and what it costs

Harvard Business Review's June 2026 analysis of AI adoption and middle management identified what it called a triple burden: managing AI experimentation, maintaining delivery, and developing people, all simultaneously, all with unchanged or increased delivery pressure and no formal support structure. The translating VP is living that triple burden in a specific form. She is managing the technical requirements of an AI integration she does not own. She is managing the team expectations of a timeline that has already moved. She is managing C-suite confidence in a project whose delay she has to explain without naming the governance failure that caused it. Each of those tasks requires a different register, a different level of disclosure, and a different emotional position. Writing four emails in that mode in a single week accumulates as depletion, not workload. Gallup's 2026 State of the Global Workplace report found that manager engagement fell from 31% in 2022 to 22% in 2025, a nine-point drop that was the steepest decline of any employee group (Gallup, 2026). Gartner predicts that in 2026, 20% of organizations will use AI to flatten their structure, eliminating more than half of current middle management positions. The person absorbing the structural gap between business and IT is doing so at the moment her role is most exposed and least supported. Translation work accumulates precisely because it is invisible. It does not appear on a project plan. It does not have a budget code. It does not surface in a steering committee. It surfaces in an email drafted on a Tuesday at 3:15 PM that nobody asked her to write, and that she will finish before she goes home, because that is what she does. The sentence I want her to have at that moment is this one: writing the email postpones the problem. She finishes it, closes her laptop, and the governance gap is still there in the morning.

Why the C-suite does not see it

Larridin's State of Enterprise AI 2026 report, based on a survey of 365 senior leaders at companies with 1,000 or more employees, asked executives, VPs, and directors whether they had visibility into AI use in their organization. 92.4% of executives believed they did. 84.7% of VPs believed they did. 76.3% of directors believed they did (Larridin, 2026). Each level had less visibility than the one above it believed it had. Deloitte's 2026 Deloitte AI Institute State of AI in the Enterprise report, based on a survey of more than 3,000 executives across 24 countries, found that 84% of companies have not redesigned roles around AI, and only 21% have a mature AI governance model. Meanwhile, approximately 75% plan to deploy agentic AI within two years. The gap between deployment velocity and governance maturity is not closing. Writer's 2026 enterprise AI adoption research found that 53% of business leaders feel IT teams are not delivering real value with generative AI, and cite growing tension between IT and other lines of business. From the other direction, 58% of executives admit that many fellow leaders lack the fundamental knowledge to make strategic decisions about AI. Both readings are accurate, and both are operating from genuine information gaps. Grant Thornton's 2026 survey found that organizations with fully integrated AI are nearly four times more likely to report revenue growth than those still piloting: 58% versus 15%. The accountability difference between those two groups is decision rights: the leading organizations can show how their AI makes decisions, who owns the outcomes, and what happens when something goes wrong. The SVP who receives the fourth email this week does not have that picture. She receives a well-written explanation of a delay. She forwards it to her next level. The structural gap that produced the delay remains in place for the next integration, the next Q-planning cycle, the next Thursday flag.

Shadow AI is already resolving the governance problem, badly

When the formal governance process slows an integration to six months or kills it outright, the people who need the output find a workaround. Microsoft's Work Trend Index found that 78% of AI users are bringing their own tools to work (Microsoft, 2025). Writer's 2026 enterprise AI research found that 67% of executives believe their company has already suffered a data leak or breach due to unapproved AI tools (Writer, 2026). In retail, where supplier data, consumer transaction records, and pricing architecture sit in the same environment, the exposure from shadow AI is immediate. IBM's 2026 report found that shadow AI breaches averaged $5.39 million each, roughly $400,000 above the global breach mean. Organizations requiring IT approval before deploying AI tools decreased from 45% to 38% year over year in the same research cycle, meaning governance controls are regressing as exposure grows. The VP in the middle often knows this is happening on her team. She has seen the spreadsheet exports, the ChatGPT summaries of supplier reconciliation data, the ad hoc tools her analysts have started using because the formal integration is still in IT's queue. She has not escalated it because escalating it would slow the work further, add another governance conversation to her already full plate, and require her to explain to the C-suite why her team is using tools that were never approved. So the rationalization is clean: it is working, nobody is complaining, and the report came in on time. She forwarded it with 'looks good.' She did not ask where that number came from. That moment, the approved report and the question she did not ask, and the missed one-on-one with the IT director and the four emails she wrote instead, are the same mechanism. She produces the artifact that demonstrates competence and defers the conversation that could change what she is managing. She does it at 8:47 AM on Monday and at 3:15 PM on Tuesday and on the Thursday she learned about the governance flag and decided to process it by email rather than by phone. Each instance is small. The accumulation is not.

What good ownership structure looks like

Apotheker et al. (2025), referenced in Innovative Human Capital's analysis of the widening AI value gap, found that approximately 40% of future-built companies explicitly embed shared ownership of AI into governance structures, compared with 19% of stagnating firms. These organizations are 1.5 times as likely to adopt a collaborative model with clear decision rights across both business and IT. Deloitte's 2026 State of AI in Retail and CPG survey names the prescription directly: put AI strategy in the hands of the P&L owners accountable for the outcomes, redirect spend toward the use cases that move revenue and margin, and treat scaling, not piloting, as the real measure of progress. In practice, the structural requirement is simpler than most governance frameworks make it. Someone needs to own the outcome. That person needs the authority to resolve a data governance flag without writing four emails to figure out whose queue it sits in. And she needs to know about the flag before the integration is already committed to a Q3 timeline. That last condition is the one she can act on before any organizational redesign happens. Knowing about the flag before Q3 planning closes requires a relationship with the IT director that currently does not exist, because she has been managing the gap by email rather than by conversation. Supply chain AI adoption illustrates how far most organizations are from that condition. ABI Research's 2025 survey of 490 supply chain professionals found that 94% of companies plan to use AI for decision support within two years. Gartner found that only 23% of supply chain organizations have a formal AI strategy. The distance between intent and documented decision rights is where the integrations die and the explanatory emails are written.

What could go wrong

Naming the honest risks in this frame:

• Assigning a single owner does not fix a fragmented data architecture.

  Clear accountability helps. It does not resolve the twelve-to-twenty-four months of integration work required when core systems predate modern API standards. Ownership without infrastructure investment produces a clearer failure, not a faster success.

• Centralizing governance in retail can concentrate risk rather than reduce it.

  Deloitte's 2026 State of AI in Retail and CPG survey found that retail is consolidating around centralized AI governance (43%), while CPG has not converged. Centralization speeds decision-making but also means a single governance misjudgment affects the entire portfolio.

• Clarity about the role does not add capacity to fill it.

  Recognizing that she is absorbing a structural gap does not add hours to her week. Naming the problem without addressing the staffing or budget conditions that produced it adds awareness to depletion, not relief. This is worth saying directly, because a lot of governance advice stops at the diagnosis.

• Shadow AI does not disappear when formal governance improves. It relocates.

  Teams that have built workarounds to compensate for slow formal processes have also built habits, muscle memory, and informal efficiencies around those tools. A governance tightening without a functional alternative produces friction, not compliance.

• The escalation path for a governance flag is less reliable than the C-suite believes.

  Larridin's State of Enterprise AI 2026 found that 92.4% of executives believe they have visibility into AI use in their organization, while directors are operating with meaningfully less (Larridin, 2026). Flags get absorbed at the VP level because she has learned that surfacing them produces more work than managing them.

The J.Caresse point of view

I want to say this in my own voice, because it is something I know from the work, not from the research. Across years of operations and consulting work in retail and consumer environments, the pattern that shows up in every stalled AI implementation is a decision-rights vacuum that everyone has agreed, implicitly, to let someone in the middle fill. The filling is invisible because it looks like competence. She handles it. The report comes in. The meeting runs on time. Nobody asks what it cost her to make that happen, because the asking would require acknowledging that the structure put her there. The argument this month is that AI readiness begins with the person signing off on the implementation actually knowing what her team does to make the numbers look right. In the context of the business-IT conflict, that means knowing what governance flags are sitting in IT's queue, what workarounds the team has already built, and what the real timeline is, before the Q3 commitment is made. That knowledge requires a relationship with the IT director she did not make time to build, because she was writing explanatory emails instead. The four emails are symptoms. The missed one-on-one is the mechanism. The mechanism is the one she can actually change. An organizational redesign requires a sponsor, a budget cycle, and a steering committee. A one-on-one requires a thirty-minute block on a calendar. So here is the ask, and it is a small one: before she opens the next draft email explaining a delay, she opens her calendar instead. One slot, thirty minutes, with the IT director. No agenda beyond one question: what is in your governance queue that touches my roadmap, and when were you planning to tell me? That conversation will not fix the structure. It will tell her what she is actually managing, which is more than the four emails have done.

Key takeaways

What this research brief establishes, in plain terms:

• The business-IT conflict has a structural cause, and it lands on a specific person.

  Business functions own the roadmap. IT functions carry compliance obligations. When decision rights over the outcome are unassigned, the person closest to both absorbs the gap. In retail, that person is consistently a VP or SVP of operations, supply chain, or merchandising.

• Governance, not model readiness, is what delays AI deployments.

  AvePoint's 2026 research found that 86.9% of organizations delayed generative AI deployments by nearly six months, driven by data security and governance concerns (AvePoint, 2026). The tools are ready. The decision structure is not.

• The translation work is real work, and it accumulates as depletion.

  Writing explanations upward and translating decisions downward is invisible on every project plan. Harvard Business Review's June 2026 analysis named the triple burden middle managers are now carrying. Gallup's 2026 State of the Global Workplace documented the result: manager engagement fell from 31% in 2022 to 22% in 2025, the steepest drop of any employee group (Gallup, 2026).

• 54% of retail AI strategy ownership sits with tech leaders rather than the P&L owners who must deliver results from it (Deloitte, 2026).

  Deloitte's 2026 State of AI in Retail and CPG survey of 200 retail and CPG executives identifies this as the governance risk that matters more than which model an organization picks. Strategy owned by someone without P&L accountability produces commitments that cannot be delivered (Deloitte, 2026).

• Shadow AI is a current condition being managed at the VP level through rationalization.

  78% of AI users are already bringing their own tools to work (Microsoft, 2025). 67% of executives believe their company has already suffered a breach from unapproved tools (Writer, 2026). The rationalization is clean: the report came in on time. The question of where the number came from goes unasked.

• Organizations with named outcome ownership report revenue growth at four times the rate of those still piloting.

  Grant Thornton's 2026 survey found 58% versus 15% (Grant Thornton, 2026). The separation between those two groups is accountability structure: a named person, clear decision rights, and a known answer to what happens when something goes wrong.