Strategic Insights
3 Foundational Decisions That Separate the 6% Generating EBIT from AI
Ownership, portfolio discipline, and embedded governance—the operating-model choices that compound into enterprise-scale returns.
Ownership, portfolio discipline, and embedded governance—the operating-model choices that compound into enterprise-scale returns.
Enterprise AI adoption is near-universal, but only 6% of organizations generate 5% or more of EBIT from it. The difference comes down to three operating-model decisions—not technology, talent, or budget, but three choices—CEO-level ownership, portfolio concentration, and governance embedded from Month 0—that create a compounding advantage no amount of experimentation replicates. This brief decomposes those decisions, presents the evidence behind each, and provides a 90-day path from reading to action.
The 6% do not run more pilots, adopt faster models, or hire more data scientists. They make three operating-model decisions that the remaining 94% avoid: they place AI under CEO-level ownership with P&L accountability, they impose portfolio discipline that concentrates resources on the highest-value use cases, and they invest in shared capabilities that eliminate the duplication trap.
Each decision is necessary. None is sufficient alone. Together, they create a compounding advantage that no amount of experimentation can replicate. The decisions are mutually exclusive in scope and collectively exhaustive in coverage—they address who decides, what gets built, and how it operates.
88% of organizations now use AI in at least one business function, up ten points year-over-year (McKinsey, 2025). Investment is accelerating. Executive attention is high. The technology works. By every surface-level indicator, AI has arrived.
Yet the returns remain concentrated. Only 25% of AI initiatives deliver expected ROI. Only 16% have scaled enterprise-wide. Most investments require two to four years to show satisfactory returns—and many never do. The pattern is consistent across industries: adoption is high, scale is low, and value capture is narrow.
When researchers decompose the 6%, one variable dominates. 90% of organizations achieving the highest ROI from AI have it driven directly by the CEO or C-suite. Only 10% of all organizations structure ownership this way. The majority locate AI in innovation labs, IT departments, or centers of excellence, where accountability diffuses and use cases proliferate based on technical novelty rather than business necessity.
| AI Ownership Model | % of Organizations | ROI Performance |
|---|---|---|
| CEO / C-suite with P&L authority | ~10% | Highest ROI cohort (90% of top performers) |
| CIO / CTO (technology function) | ~35% | Moderate; limited to IT-adjacent use cases |
| Innovation lab / CoE | ~40% | Low; fragmented portfolios, pilot stall |
| Decentralized / BU-led | ~15% | Inconsistent; duplication across units |
The table makes the pattern visible: ownership location functions as a performance predictor, not merely a governance detail.
Most executive teams attribute slow AI progress to technical barriers—data quality, model readiness, integration complexity. The evidence points elsewhere. The organizations stuck in pilot have the same access to models, cloud infrastructure, and vendor partnerships as those scaling successfully.
The real barriers are organizational: unclear ownership that lets initiatives drift, overfull portfolios that spread resources across dozens of low-impact use cases, and governance applied as a post-build review rather than a design constraint. These are operating-model problems. They cannot be solved with better technology.
65% of organizations remain in pilot mode not because their AI does not work, but because they have not made the decisions that allow it to compound. The cost of deferring those decisions rises each quarter as competitors who made them earlier pull further ahead.
The three decisions have different time horizons. Ownership is a Q1 decision. Portfolio is a Q1–Q2 execution. Governance embedding requires Q2–Q3 design and build. Organizations that wait past Q2 to begin the sequence will not have compounding capability by year-end.
When AI reports to the CEO, it operates under the same standards as other strategic initiatives—clear value ownership, resource authority, and integration with capital allocation. When it reports to IT, it operates as a service function: responsive to requests, evaluated on technical metrics, disconnected from revenue accountability.
The difference is material. CEO-owned AI portfolios receive direct budget allocation tied to business outcomes. IT-owned portfolios compete for budget against infrastructure, security, and operational maintenance. The result: CEO-owned portfolios concentrate on three to five high-impact use cases. IT-owned portfolios accumulate dozens of experiments with unclear business sponsors.
90% of organizations achieving the highest ROI have AI driven directly by the CEO or C-suite. Only 10% of all organizations structure ownership this way. Among the bottom quartile of ROI performers, AI ownership is distributed across IT (42%), innovation labs (31%), and individual business units (27%).
Design implication: Place AI under a C-suite executive with direct P&L authority by end of Q1. Not a dotted line. Not a committee. A single owner who can allocate resources, eliminate underperforming initiatives, and be measured on AI-driven EBIT within 12 months.
Organizations arrive with long inventories of AI use cases. Multiple teams champion initiatives based on local incentives. Product managers want prediction. Operations wants automation. Marketing wants personalization. HR wants analytics. Without discipline, every function builds its own pipeline, its own data connections, and its own governance workarounds.
The 6% generating meaningful EBIT select ruthlessly. They apply a two-axis scoring framework—Strategic Impact × Implementation Feasibility—to every proposed initiative. The top three to five get funded. Everything else stops. The savings from killed projects fund shared capabilities that accelerate the winners.
| Portfolio Behavior | Bottom 94% | Top 6% |
|---|---|---|
| Active AI initiatives | 15–30+ | 3–5 |
| Selection method | Team advocacy / CEO interest | Strategic scoring framework |
| Shared capabilities | Built per project | Built once, deployed across many |
| Elimination rate (underperformers) | <10% per year | 60–70% at scoring stage |
Organizations with disciplined portfolio management report 2.5× higher per-initiative ROI than those with broad, unfocused portfolios. The relationship is not linear—there is a threshold effect: once the portfolio exceeds eight active initiatives without shared infrastructure, marginal returns per initiative approach zero.
Design implication: Apply Strategic Impact × Implementation Feasibility scoring to every active and proposed initiative. Use the Starkey Model to identify the top 3–5. Redirect resources from everything else. Build shared data pipelines and orchestration layers as platform investments, not project expenses.
Most organizations treat governance as a review stage. Legal, compliance, and risk teams evaluate AI systems after they are built. The result: months-long approval queues, redesign cycles, and organizational resentment toward oversight functions.
The 6% embed governance into system design from the start. Before the first line of code, they convene Legal, Risk, Operations, and the business owner in a single working session. Explainability requirements, risk thresholds, data lineage standards, and integration points are defined as design constraints, not post-build checkpoints.
The difference compounds. Embedded governance eliminates the approval bottleneck entirely. Automated checks replace repeated sign-offs. Predefined thresholds reduce subjective judgment calls. Teams build faster because integration requirements are clear from day one. Operational resistance drops because stakeholders were in the room when decisions were made.
Organizations with embedded AI governance report 40% faster time-to-deployment compared to those using post-build review models. The improvement is driven by two factors: elimination of the redesign cycle (which adds 2–4 months to typical deployments) and reduced organizational friction during rollout. PwC’s Responsible AI Survey found that organizations with formal governance frameworks are 2.3× more likely to scale AI beyond pilot.
Design implication: Run a Month-0 governance session for the next AI initiative entering development. Define explainability, risk thresholds, and integration points before any build begins. Measure success by whether the initiative launches without a post-build review delay.
The Starkey Model evaluates AI initiatives along two axes: Organizational Value Potential and Implementation Feasibility. The simplified preview below shows representative criteria per axis, drawn from the full model’s custom evaluation framework. Use these as a first-pass filter for your current AI portfolio before engaging the complete diagnostic.
| Organizational Value Potential | Implementation Feasibility |
|---|---|
| Revenue impact or cost reduction measurable within 12 months | Named executive owner with decision authority |
| Strategic alignment with at least two of the Four Capability Bands | Organization can absorb the change within current capacity |
| Compounds with existing portfolio investments | Technical infrastructure exists or can be provisioned within one quarter |
| Addresses a capability gap visible at the executive level | Governance trail is documentable from pilot to production |
The full Starkey Model applies custom weighted evaluation criteria calibrated to your priorities, assembles scoring panels by demonstrated expertise, and produces documented portfolio governance decisions. Available as a standalone diagnostic.
Source: RBD. The Starkey Model, adapted from The Intelligence Organization™, Ch. 6.These criteria are derived from Band 04 (Adaptive Governance) of The Intelligence Organization. Each question must be answered affirmatively before an initiative advances from evaluation to pilot.
| # | Go / No-Go Criterion |
|---|---|
| 1 | Does this initiative have a named executive owner with decision authority (not a committee)? |
| 2 | Does it address a capability gap identified across at least two of the Four Capability Bands? |
| 3 | Can the organization absorb this initiative within current capacity without creating adoption debt? |
| 4 | Is there a documented governance trail from concept through pilot through production? |
| 5 | Does it compound with existing portfolio investments rather than creating a standalone dependency? |
Any initiative that fails two or more of these criteria should be deferred or eliminated. This is not austerity. It is portfolio governance that concentrates resources where organizational absorption capacity exists.
Source: RBD. Portfolio Governance Criteria, derived from The Intelligence Organization, Band 04: Adaptive Governance.In many enterprises, AI ownership sits with a CIO or CTO who holds both the organizational value mandate and the technical implementation remit. The instinct is to distribute decisions through committees. The Intelligence Organization model offers a different approach: polycentric authority with clear decision nodes.
| Distribute | Retain |
|---|---|
| Use case identification (functional teams closest to the work) | Portfolio governance (which initiatives advance, which are eliminated) |
| Adoption design (change management owns rollout) | Capital allocation (budget authority stays with the owner) |
| Fluency development (HR partners with functional leads) | Vendor architecture decisions (strategic, not tactical) |
| Integration testing (engineering validates operational fit) | Escalation authority (the final decision node when stakeholders disagree) |
The principle is not delegation. It is distributed authority weighted by proximity to the work, with governance retained where accountability must be singular. This avoids both the bottleneck of centralized control and the drift of committee-based diffusion.
Source: RBD. Polycentric Authority Model, derived from The Intelligence Organization.RBD. helps leadership teams make the three operating-model decisions that separate the 6% from the rest—using proprietary diagnostic frameworks developed from Fortune 1000 client work.