Accenture's GenAI revenue reached $2.7 billion in fiscal year 2025, up from $0.9 billion the year prior. McKinsey QuantumBlack fields approximately 1,400 data scientists. Deloitte, BCG X, and IBM Consulting each employ thousands of AI practitioners across hundreds of engagements simultaneously.
These are real capabilities. The question is not whether the major consulting firms can do AI work. They can, and at significant scale. The question is whether the incentive structure that makes them capable of that scale is compatible with the kind of production accountability — the contractual commitment to a specific outcome, in a defined timeline, under a commercial model that transfers risk — that enterprise buyers in regulated industries need to unlock production AI.
The answer is structural, and it is no.
I. The Utilization Economics of the Consulting Pyramid
The major consulting firms operate on a pyramid model. Senior partners and directors sell and scope engagements. The work is delivered by a graduated pyramid of managers, consultants, and analysts, each billing at a rate that reflects their position in the hierarchy rather than the marginal contribution of their specific labour to the client's outcome. The pyramid economics work because the client is buying coverage — access to the firm's brand, bench, and global delivery infrastructure — not specifically the senior expertise that sold the engagement.
This model has served professional services well for decades, and it generates the scale that allows firms like Accenture to book $5.9 billion in GenAI engagements in a single year.
It is fundamentally incompatible with genuine fixed-outcome AI delivery for three reasons.
Reason 1: Utilization targets and outcome commitments are in direct conflict. A consulting firm that commits to a fixed outcome — this AI system will be in production in 90 days, delivering X EBITDA impact, under an SLA — has made a commercial commitment that is bounded by the outcome. If the outcome is achieved efficiently, in fewer hours than the engagement budget, the margin on those hours is high. If the engagement runs long, the margin is compressed. Every hour over budget is a direct P&L impact on a fixed-fee contract.
The pyramid model, by contrast, derives its economics from utilization: consultants billed at rate, leveraged against partner cost, producing a margin that increases with headcount deployment. The commercial incentive of a utilization-based model runs directly against the incentive of a fixed-outcome model. A firm with utilization targets cannot simultaneously offer genuine fixed downside on an outcome commitment — the incentive to deploy headcount to protect utilization will override the incentive to deliver efficiently against the outcome.
This is not a criticism of the consulting firms' intention. It is a description of their economics. The incentive structure is the constraint.
Reason 2: The advisory-to-build conflict creates accountability diffusion. The major consulting firms are typically engaged for the strategic and advisory phase of an AI programme — the readiness assessment, the use case prioritisation, the operating model recommendation — and then separately contracted for the implementation phase, often with a different delivery team or a system integrator sub-contractor.
This creates a structural accountability gap. The team that designed the strategy is not the team that implements it. The team that implements it is compensated for delivery, not for outcome. The team that monitors the outcome is frequently a third party — the client's internal operations team or a managed services provider. When the production system underperforms, the accountability question has no clear answer: the strategists point to the implementation, the implementers point to the strategy, and the managed services team points to both.
Genuine production accountability requires a single party that designed, built, deployed, and maintains the system — contractually committed to its performance throughout the entire lifecycle. The advisory-to-build separation that characterises most major consulting engagements makes this accountability structure impossible to establish.
The team that designed the AI strategy is not the team that builds it. The team that builds it is not compensated for the outcome. When production underperforms, the accountability has no address. This is the structural consequence of the advisory-to-build separation — not a failure of execution, but a failure of commercial design.
Reason 3: The brand-insurance premium prices in zero performance risk. Part of what enterprises pay for when they engage McKinsey or Accenture is brand insurance: the implicit protection of having engaged a name firm, which reduces the personal career risk of the executive sponsor if the programme fails. This is a real and rational purchase — in large, regulated enterprises, career-risk management is a legitimate input to vendor selection.
The price of brand insurance, however, is that the vendor's commercial exposure to failure is minimal. McKinsey's fee is earned whether or not the operating model recommendation converts to a production AI system. Accenture's implementation fee is earned whether or not the delivered system achieves the projected EBITDA. The premium paid for brand insurance is not a premium paid for outcome accountability — it is a premium paid for the diffusion of accountability, distributed across the firm's reputation rather than concentrated in the commercial terms of the engagement.
II. What Genuine Production Accountability Requires
The alternative to brand-insurance consulting is not cheaper consulting. It is a structurally different commercial design.
Genuine production accountability has four commercial and operational requirements that distinguish it from the incumbent model.
Fixed scope with a defined production outcome. The engagement specification describes what will be delivered — an AI operating model in production for a specific workflow — and what production means: performance thresholds, SLA commitments, governance architecture, and EBITDA attribution methodology. The specification is agreed before the build begins and is not renegotiated in response to scope creep or delivery difficulty. If additional scope is required to achieve the specified outcome, the implementation party carries that cost.
Risk-transfer commercial structure. The implementation partner carries a defined portion of the delivery risk. This takes several forms: a fixed fee with no cost escalation for delivery overruns; SLA-based credits tied to specific, measurable performance metrics; a capped outcome kicker — a share of verified, controllable EBITDA improvement — that aligns the implementer's return with the client's benefit; and indemnification against production failures within the implementer's scope of control.
This structure is commercially viable for an implementation firm only if the delivery economics support it: the cost of delivery is predictable (not driven by scope discovery), the outcome is attributable to the implementation rather than to exogenous market factors, and the firm has the operational capability to deliver within the specified scope.
Continuity across the full lifecycle. The same team that scopes the diagnostic also builds the production system and maintains it in production. This continuity is not primarily about knowledge management (though it helps). It is about accountability: the team that wrote the governance architecture is responsible for its compliance in production; the team that made the throughput projections is responsible for the throughput that materialises.
Senior-only delivery. The accountability model is not compatible with the pyramid leverage structure. If the production outcome is contractually committed, the risk of junior-to-senior knowledge gaps, ramp time, and supervised-but-imperfect delivery is carried by the implementation firm, not the client. The only commercially rational response is to staff engagements with the senior expertise required to execute the specification reliably — not to leverage senior oversight against junior delivery.
A correctly structured production accountability engagement looks nothing like a consulting engagement. It is closer to a construction contract: fixed scope, fixed price, defined outcome, risk-transfer terms, and a single accountable party across design, build, and post-delivery performance.
III. Why Hyperscalers Face the Same Problem From a Different Angle
The consulting firms face the accountability problem through utilization economics. The hyperscalers — AWS ProServe, Google Cloud Professional Services, Azure Consulting — face it through a different structural conflict: their primary commercial interest is compute consumption, not outcome accountability.
Hyperscaler professional services are, in aggregate, a customer acquisition and expansion motion for the cloud business. The objective of a ProServe engagement is to land and expand the cloud footprint of the account, and the P&L of the professional services unit is subordinate to the compute and platform revenue it generates. This creates an incentive to build solutions that maximise cloud resource consumption over solutions that maximise client EBITDA.
More specifically, hyperscaler ProServe cannot offer genuine multi-cloud or cloud-independent architecture. Recommending that a client self-host a model, or recommending that an architecture run primarily on a competitor's cloud, is outside the commercial scope of a hyperscaler engagement — not because the technical recommendation would be wrong, but because it would not serve the primary commercial objective.
For enterprises in regulated industries — where data residency requirements, multi-cloud resilience mandates, and vendor-concentration risk management frequently require cloud-neutral architecture — this is a material limitation. A production AI operating model designed to maximise AWS consumption may not be the right architecture for a financial services institution facing DORA operational resilience requirements or for a pharmaceutical firm with data localisation obligations in multiple jurisdictions.
The hyperscalers' ProServe units are capable of excellent technical work. The conflict is not in the quality of the engineers. It is in the commercial incentive that governs the engagement design.
IV. The Proof Gap and Who Fills It
The practical consequence of the incumbent accountability gap is a trust deficit in the market that the consulting firms' brand does not fully repair.
Enterprise buyers know that Accenture has delivered AI projects at hundreds of clients. They cannot easily access, verify, or reference the EBITDA outcomes of those projects. The reference calls available through a major consulting engagement are typically with sponsors who approve of the relationship, not with CFOs who can verify specific, quantified EBITDA results. The brand insurance that incumbents provide is insurance against project failure; it is not a guarantee of financial outcome.
The vendors filling the proof gap are those investing specifically in the production accountability architecture: building small cohorts of referenceable, deeply documented EBITDA outcomes in specific verticals; structuring commercial models that transfer defined portions of delivery risk; and investing in the governance certifications that make a small firm accessible to the procurement and risk functions that regulate enterprise vendor selection.
This is a different competitive advantage than scale. It is not available to every firm — it requires genuine specialisation (depth in specific workflow verticals), genuine governance investment (SOC 2, SR 11-7-aligned model risk practice, EU AI Act-compliant architecture), and genuine commercial discipline (declining engagements where the outcome cannot be committed).
The asymmetry is structural: an incumbent's utilization economics make genuine fixed downside a dilutive exception; a production-accountable firm's economics require it as the standard.
V. What to Evaluate When Selecting an Enterprise AI Implementation Partner
The vendor selection process for a production AI engagement should evaluate different criteria than the process for an advisory engagement. The following framework identifies the attributes that predict production accountability, as distinct from advisory quality or technical capability.
Criterion 1: Commercial model structure
Does the vendor offer a fixed-scope, fixed-fee engagement for the production build phase — or is the commercial structure time-and-materials, requiring the client to absorb scope and delivery risk? Is there an SLA commitment with defined credits tied to measurable production metrics? Is there an outcome-linked component (even a small one) that aligns the vendor's return with the client's EBITDA?
A vendor unwilling to commit commercially to what they are delivering is communicating something about their confidence in delivering it.
Criterion 2: Referenceable EBITDA outcomes in your specific vertical
Not references in general. Not project-completion references. Specific, quantified EBITDA outcomes — throughput improvement, loss-rate reduction, speed-to-decision gain — in engagements comparable to yours: same industry, same workflow category, same regulatory environment, similar scale. With CFO-level contacts available for verification calls.
A vendor who cannot provide three such references either has not done it before or has done it but cannot verify the outcomes. Neither is a production-accountability vendor.
Criterion 3: Governance documentation available before contract signature
The vendor should be able to produce, before any implementation agreement, documentation of their governance methodology: how HITL escalation is designed, what their drift monitoring architecture looks like, how they produce SR 11-7-compliant model documentation, and what their incident response procedure is. This documentation should be specific — not a general description of their "AI governance framework," but the actual artifacts that will be produced for your engagement.
A vendor who defers governance documentation to the post-contract phase has not built governance architecture into their delivery model. They are describing governance as a deliverable they will figure out. That is governance-by-retrofit.
Criterion 4: Continuity across the engagement lifecycle
Is the same team — specifically the same senior architects and engagement lead — responsible for the diagnostic, the build, and the post-deployment assurance? Or does the vendor hand off between phases? Handoffs between phases are where accountability diffuses. A vendor who commits to a single accountable team across the full engagement lifecycle is structurally capable of the accountability commitment. One who does not is not.
Criterion 5: Certifications and procurement-readiness documentation
SOC 2 certification (Type I minimum, Type II in progress or completed) for security and operational controls. Industry-specific compliance documentation where relevant: SR 11-7 model risk methodology for financial services, HIPAA controls for healthcare. A procurement-ready vendor risk management package (security questionnaire, data handling addendum, MSA template). These are table stakes for regulated-industry procurement, not differentiators — but their absence is a disqualifier.
VI. The Competitive Window That Benefits Procurement Discipline
The current market structure creates an unusual opportunity for enterprise buyers willing to apply procurement discipline to their AI partner selection: the pool of vendors with genuine production accountability credentials is small, the commercial terms available from that pool are more favourable than they will be once the pool expands, and the reference density being built by that pool today will compound into a wider moat over the next 24–36 months.
The incumbent consulting firms are actively investing to close the accountability gap — through acquisition of boutique AI firms, through productising fixed-scope offerings, and through internal governance infrastructure investment. That investment will narrow the differentiation between accountability-first vendors and brand-insurance vendors over time.
For buyers who move now — who commission a production-accountable engagement in 2026 and become reference cases for the next buyer — the compounding benefit is not only the EBITDA from the first deployment. It is the organisational capability that accrues from having a production AI operating model: the governance infrastructure, the measurement architecture, the operating model patterns, and the institutional knowledge that makes every subsequent AI deployment cheaper and faster than the first.
The enterprises deferring to incumbent consulting — waiting for the brand-insurance firms to develop a genuinely accountability-first model — are deferring to a structure that is under no economic pressure to change, because their clients keep buying the current one.
Conclusion: Accountability Is a Commercial Design, Not a Promise
The consulting firms are not dishonest when they describe their AI engagements as outcome-oriented. They intend to produce good outcomes. The limitation is not intent — it is commercial design.
An engagement structure that bills by the hour cannot be genuinely committed to the outcome. An engagement that separates advisory from implementation cannot assign unified accountability for what the implementation delivers. An engagement that offers brand insurance as the primary value driver has priced the risk diffusion rather than the risk transfer.
These are not failures that individual partners or delivery leads can overcome through determination. They are structural consequences of the commercial models the firms are built on.
Genuine production accountability requires a commercial design that aligns the vendor's economics with the client's EBITDA. It requires a team that stays across the full lifecycle. It requires the governance investment that allows a specific, credentialed outcome to be committed. And it requires the discipline to decline engagements where those commitments cannot be made — because an uncommittable engagement is, ultimately, not an accountable one.
That commercial design exists in the market. It is not the default offering of the scale consultancies. It is the defining characteristic of the firms that have chosen production accountability over scale.
See what an accountable engagement structure looks like.
ExecuteML builds production-grade AI operating models under a fixed-scope, outcome-aligned commercial model — Diagnostic Blueprint to build to ongoing assurance, with a single accountable team across the full engagement.
- Fixed-scope build, priced before commitment
- One accountable team across the full engagement
- Outcome-aligned commercial terms, not time-and-materials