You Don't Have an AI Strategy. You Have a Vendor Portfolio.
.jpg)
78% of organisations now use AI in at least one function. Fewer than one in three are scaling it enterprise-wide. McKinsey calls this the "AI theater" problem. This article explains why — and what the architecture fix requires.
- The data: McKinsey (2025) finds 78% of organisations use AI in at least one function, yet fewer than one in three scale it enterprise-wide. The gap is not technology. It is architecture.
- The cost direction: Analyst estimates suggest platform AI can add materially to existing contract costs at enterprise scale — typically 25–50% uplifts depending on edition, tier, and consumption. Exact figures vary by contract and must be validated in your negotiation.
- The architecture gap: Vendor AI is primarily optimised to deliver value within the vendor's own ecosystem. Each tool operates in its own data boundary — making cross-functional AI performance invisible and governance fragmented.
- The solution: An independent AI orchestration layer — sitting above ITSM, CRM, HR, and Finance — with centralised governance, shared context, and LLM portability your vendors cannot unilaterally change.
- The important nuance: Embedded vendor AI delivers real productivity value within its domain. The strategic concern is the absence of an independent governance layer above it — not the features themselves.
01 — The Scene Every Vendor Has an AI Story. That Is Not a Strategy.
It is Q4 2025. You are in an executive briefing with ServiceNow. The deck opens on slide three: "Now Assist: Your AI-Powered Enterprise." The demo is polished. The use cases are real — ticket summarisation, intelligent routing, knowledge article generation. The account executive explains that the Pro Plus upgrade is the path forward.
Three weeks earlier, Salesforce presented the same architecture with a different logo. Agentforce. Einstein. Revenue Intelligence. Stacked on top of your existing CRM contract.
This is the defining commercial pattern of enterprise AI in 2025. Every platform vendor has shipped an AI product. Every AI product requires a tier upgrade. Every tier upgrade creates a pricing commitment that compounds at renewal. The demos are impressive. The cumulative bill requires scrutiny.
McKinsey describes this pattern directly: organisations are excellent at "doing AI projects" and systematically poor at turning them into a new operating baseline. The three blockers — fragmented data, unredesigned workflows, and the absence of scaling governance — are not technology failures. They are architecture failures.
The question no vendor EBC answers: what happens to enterprise AI performance when all of these platforms operate in isolation from each other?
02 — The Orchestration Deficit AI Silos Are Not an AI Strategy
Modern enterprise architecture does not run on a single application. It runs on the continuous orchestration across applications — ITSM managing service delivery, CRM managing customer relationships, HR managing workforce operations, Finance managing capital allocation. These systems interact constantly. An incident in ITSM has downstream effects in Finance. A customer escalation in CRM should trigger workflows in ITSM. A workforce change in HR should update access policies in Security.
When AI exists only inside each of those individual platforms, what you have assembled is not enterprise intelligence — it is automation by silo. Each tool can optimise its own narrow domain. None can see across the boundary. None share context. None operate under a common governance framework.
Business performance is not measured application by application. It is measured across the entire orchestration of systems that power your operations. An AI footprint that cannot observe, govern, or optimise across that orchestration is a collection of expensive features — not a strategy.
Forrester's State of AI 2025 puts specific numbers to this: while over 70% of firms have AI in production, few are measuring its financial impact and business leaders remain fragmented in their governance approach. The deployment breadth is real. The strategic depth is not.
A note on vendor capability: It is worth acknowledging that vendors are not standing still on interoperability. ServiceNow, Salesforce, and others are investing in integration capabilities and cross-platform workflows. The concern is not that embedded AI lacks value — it delivers genuine tactical productivity in its domain. The concern is structural: vendor AI is primarily optimised to create value within the vendor's ecosystem, not to serve as an enterprise-wide governance or orchestration layer. That architectural gap requires an independent answer.
What AI Silos Cost Operationally
| Operational Area | Vendor-Embedded AI Reality | Independent Layer Target |
|---|---|---|
| Performance Measurement | KPIs per platform; cross-functional AI ROI invisible | Unified observability across all execution layers |
| Governance & Compliance | Each vendor's AI operates under its own compliance posture | Single enterprise-wide AI safety and audit framework |
| Data Context | Each tool operates within its own data boundary; no shared truth | Centralised knowledge accessible to all AI agents |
| Cost Predictability | Consumption overages and tier escalation at renewal | Committed, vendor-agnostic AI spend under enterprise control |
| Model Flexibility | Locked to the vendor's LLM choices and release cadence | Model portability: swap or combine models as the market evolves |
| Risk Exposure | Automated decisions cascading across unmonitored systems | Governed agent actions with observable escalation paths |
03 — The Commercial Mechanics What Vendor AI Actually Costs at Enterprise Scale
Enterprise AI features are increasingly structured not as utility pricing, but as tier-based architectural commitments that embed AI cost into the renewal baseline. Understanding the commercial mechanics matters — not because the features lack value, but because the pricing structure shapes your architectural options at renewal.
ServiceNow's Now Assist is structured across licensing tiers that require minimum upgrade commitments before AI capabilities unlock. Industry analyst estimates suggest that, for a mid-to-large enterprise, the AI-inclusive tier can represent a 25–50% increase over existing platform costs, depending on fulfiller count, edition, and consumption thresholds. Redress Compliance, 2025 ServiceNow does not publish granular AI-specific pricing publicly; figures in analyst commentary and customer discussions vary materially based on negotiated terms.
Salesforce operates a similar architecture. The base Enterprise tier provides CRM functionality. Adding Einstein Conversation Insights, Revenue Intelligence, and Agentforce layers increases per-user cost substantially. Analyst and customer-reported estimates for fully-stacked configurations range widely, with some enterprise implementations significantly exceeding $300 per user per month before implementation costs. Oliv AI, 2025 As with ServiceNow, actual contract pricing is negotiated and will vary.
An honest caveat on pricing figures: Enterprise SaaS pricing is negotiated, not listed. Figures cited in analyst commentary — including those referenced in this article — represent reported ranges from customer and analyst sources, not vendor-published rates. CIOs should treat directional cost signals as a prompt for their own procurement analysis, not as universally applicable benchmarks. What is structurally consistent across vendors is the direction: AI is being moved from optional feature to tier-inclusive baseline, making it harder to isolate and negotiate independently at renewal.
The strategic risk is not the upfront cost of AI features. It is the trajectory of renewal pricing as AI becomes embedded in core workflows. When critical processes depend on vendor AI outputs, the switching cost at renewal grows — regardless of whether the feature value justifies the pricing increase. Negotiating leverage decreases in proportion to operational dependency. The time to design for optionality is before that dependency forms.
Structured Q&A:
Vendor AI features automate tasks within a single platform's data boundary and are primarily optimised to deliver value within the vendor's ecosystem. An enterprise AI strategy orchestrates intelligence across all platforms, governs AI decisions uniformly, and remains independent of any single vendor's pricing model or LLM roadmap. The former is a product. The latter is an architectural capability.
Yes — within its domain. ServiceNow's AI features can meaningfully accelerate ITSM workflows; Salesforce's AI can improve sales team efficiency. The strategic concern is not whether these features deliver value in isolation. It is whether an organisation has an independent governance and orchestration layer above them. Without that layer, enterprise-wide AI performance remains invisible, ungovernable, and architecturally tied to vendor pricing decisions.
Lock-in tends to develop gradually. An enterprise adopts platform AI at one licensing tier; workflows become dependent on AI outputs; and at renewal the vendor positions AI-inclusive packaging as the primary upgrade path. Migration becomes operationally costly because AI is embedded in critical processes while data and governance remain inside the vendor's ecosystem. The cost of lock-in is not always visible until the first renewal cycle.
An AI orchestration layer is an independent control plane that sits above existing ITSM, CRM, HR, and Finance platforms. It manages multi-model routing, centralised governance, cross-platform observability, and shared operational context — without modifying underlying systems. It gives enterprise leaders visibility and control over AI performance across the entire business, not just within individual tools. Gartner and Forrester both identify enterprise-wide agent governance as a critical infrastructure gap for 2025–2026.
04 — The Architecture What Genuine AI Independence Requires
Building an independent AI layer does not mean replacing your existing platforms. ServiceNow, Salesforce, Workday — these remain your execution infrastructure. The shift is architectural: adding an independent governance and orchestration layer above them — one your vendors do not own, cannot unilaterally modify, and cannot monetise against you at renewal.
Gartner projects that by 2028, 33% of enterprise software applications will include some form of autonomous AI capability, up from under 1% in 2024, with 15% of day-to-day business decisions made by AI agents. Forrester and IDC similarly identify 2026 as the transition year for multi-agent system deployment at enterprise scale. When AI agents operate across 15 or more platforms simultaneously, an organisation without a cross-platform governance layer is not running AI at scale — it is running AI at risk.
McKinsey's 2025 data reinforces the urgency: while 62% of organisations are experimenting with AI agents, only 39% report measurable profit impact. The gap between deployment and value is not a model problem. It is a governance and orchestration problem.
On the complexity of independent layers: Building and governing an enterprise AI orchestration layer is not a trivial undertaking. It carries its own operational costs, integration complexity, and organisational change requirements. The argument is not that independence is free — it is that the cost of independence is finite and known, while the cost of architectural dependency compounds silently at every renewal cycle.
Five Capabilities That Define the Independent AI Layer
- Centralised Governance. A unified framework for data privacy, model bias auditing, and compliance reporting across all models and platforms — not dependent on any individual vendor's compliance posture or roadmap.
- Cross-Platform Intelligence. The ability to synthesise operational signals from ITSM, CRM, HR, Finance, and Security into a coherent business picture — not four isolated dashboards operating on separate data boundaries.
- Unified AI Observability. A consolidated view of model consumption, API latency, costs, error rates, and autonomous decision outcomes across every AI implementation in the enterprise.
- Shared Operational Context. A single authoritative data layer that all AI agents reference — eliminating contradictory outputs, duplicated pipelines, and knowledge silos.
- Model Portability. The architectural capacity to swap or combine underlying models as the market evolves, without renegotiating vendor contracts or rebuilding platform integrations.
In a 2025 survey by a16z of 100 enterprise CIOs, 37% report running five or more LLMs in production — driven not solely by lock-in avoidance but by genuine model differentiation across use cases. Enterprises are already operating multi-model environments. The question is whether they have the governance infrastructure to manage them.
05 — The Decision Framework Five Diagnostic Questions for Technology Leaders
Before committing to the next platform AI upgrade cycle, the following questions provide a structured way to assess architectural exposure. These are not vendor evaluation criteria. They are enterprise architecture health questions.
| Diagnostic Question | 🔴 Exposure Indicator | 🎯 Target Architecture |
|---|---|---|
| Who governs AI decisions across all platforms? | Each platform manages its own AI compliance independently | A unified enterprise AI governance framework sits above all platforms |
| Can you measure aggregate AI ROI across systems? | Reporting exists only within individual platform dashboards | Cross-platform observability with consolidated financial reporting |
| What is your position if your primary AI vendor changes pricing at renewal? | Material budget impact with limited architectural alternatives | Modular design allows model and vendor substitution without rebuilding |
| Do AI agents across different platforms share operational context? | Each tool operates within its own knowledge boundary | Shared context layer referenced by all agents across the estate |
| Who owns your AI capability roadmap? | Your vendor's product release cycle determines your capabilities | Internal architecture decisions determine your AI trajectory |
More than two exposure indicators is a meaningful architectural signal. It does not necessarily mean immediate action — the right response depends on contract timing, operational maturity, and organisational readiness. But it is a signal that should inform your next renewal negotiation and investment cycle.
"Who owns the intelligence strategy of our enterprise — us, or our vendors?"
If the honest answer is a distributed web of platform roadmaps and consumption-based licence tiers, the enterprise has not adopted an AI strategy. It has outsourced one.
Vendor-embedded AI delivers real tactical value within its domain. That is worth acknowledging clearly, not hedging away. ServiceNow can meaningfully improve ITSM throughput. Salesforce's AI can accelerate pipeline management. These are legitimate productivity gains.
The strategic question is a different one: whether accumulating these features across a fragmented vendor estate constitutes a strategy — and whether the governance, observability, and cost structures that result are acceptable as AI consumption pricing escalates at every renewal cycle for the next decade.
McKinsey identifies it precisely: many local wins, little systemic reinforcement. The path from AI experimentation to compounding enterprise advantage runs through governance infrastructure, workflow redesign, and architectural independence — not through the next vendor tier upgrade.
The window to build that architecture independently is open. It closes incrementally with each contract renewal that deepens operational dependency. The organisations that will hold strategic AI advantage in 2028 are not those that bought the most AI features. They are those that built the layer that owns the intelligence.
- McKinsey & Company (2025): The State of AI — Global Survey 2025. Enterprise AI adoption, pilot-to-production gap, "AI theater" pattern, agent deployment data.
- McKinsey & Company (2024): Survey finding that 78% of firms use AI but only 18% have enterprise-wide AI governance councils.
- Forrester Research (2025): State of AI Survey 2025 — AI in production, governance gaps, business leader fragmentation.
- Gartner (2024–2025): Agentic AI forecast — 33% of enterprise software applications to include autonomous AI by 2028, up from <1% in 2024.
- F5 Networks / Business Wire (2025): State of AI Application Strategy Report — 77% of enterprises face security and governance hurdles; only 2% qualify as highly AI-ready.
- Andreessen Horowitz / a16z (2025): Survey of 100 enterprise CIOs — multi-LLM production deployments, model differentiation patterns.
- Redress Compliance (2025): ServiceNow Now Assist AI Strategy — tier structure, uplift ranges, procurement mechanics. Note: figures are analyst estimates; enterprise contracts vary.
- Oliv AI (2025): Salesforce AI Pricing — stack analysis and TCO commentary. Note: figures are analyst estimates; enterprise contracts vary.
- IDC (2025–2026): Worldwide AI and Automation Predictions — agent adoption trajectory and governance infrastructure requirements.
Is Your AI Strategy Vendor-Owned?
Datoin helps enterprise organisations move from fragmented AI pilots to production-ready, vendor-agnostic AI architectures. If your AI strategy currently lives inside vendor contracts, it's time for a different conversation.
Talk to Datoin →