By Pratik Jain, Senior Director of Technology at Kyvos Insights
Despite all the AI euphoria, Mint quotes Nandan Nilekani having said that “The AI deployment gap in large enterprise clients is real and closing that gap is where the work is.”
Enterprises expect agentic AI to augment or replace many business processes within the next two years, yet only 7% say their data is completely ready for AI adoption. According to a new study from Cloudera and Harvard Business Review Analytic Services, more than 70% report significant challenges in preparing data for AI initiatives.
This is hardly surprising. Enterprise data architectures were built to store, process and move data efficiently. Their design objectives never included providing the complete business context and baked-in governance that AI systems require to reason, make decisions and act autonomously. Context, in this sense, combines semantics that define what data means, relationships that describe how entities across systems connect and an ontology that structures this knowledge so AI can reason over it.
As AI adoption accelerates, organizations are discovering a widening gap between what intelligent systems need and what existing data foundations can deliver. Consequently, many AI initiatives struggle to progress from experimentation to reliable enterprise-scale deployment.
Before organizations can build a foundation for enterprise AI, they must first understand where and why AI breaks down in production.
Why Enterprise AI Fails
Enterprise data does not carry business meaning by default. This gap was bridged by analysts and domain experts, who applied business knowledge and selected the correct definitions before decisions were made. They understood that revenue could have different definitions for finance and sales, knew which customer records were authoritative, and could reconcile inconsistencies across systems.
AI systems possess no such implicit understanding. They process data exactly as it is presented to them. This creates several compounding failure modes.
AI may apply its own interpretation of metrics such as revenue, margin or customer count when governed definitions do not exist. Different AI agents may answer the same question differently because they resolve against different sources or calculations.
AI also struggles to connect information spread across systems because the relationships between customers, products, transactions and support interactions are rarely defined in ways that machines can understand.
These failures compound at scale. AI generates query volumes far beyond anything human analysts produce, creating performance bottlenecks and inflating token usage.
What’s the Solution
Designing more powerful models does not solve this problem because the issue is not intelligence; it is the absence of governed business context. Organizations are unable to determine whether AI-generated outputs align with their business and are trustworthy. The result is that most projects die before they reach production. Gartner predicts that over 40% of agentic AI projects will be canceled by the end of 2027.
Enterprise AI, therefore, needs a layer that explicitly captures the context consistently and makes it available to every AI system interacting with enterprise data.
What an AI-Ready Semantic Layer Must Deliver
At its core, an AI-ready semantic layer equips AI systems to reason, operate and scale reliably across the enterprise. It ensures that every AI-generated output is reliable and explainable by centralizing business context and serving it consistently to AI agents, chatbots and analytics consumers.
In addition, governance must be enforced at the point of query, ensuring that security policies, access controls, lineage and auditability are applied uniformly across all applications. The layer should deliver interoperability across existing data sources and consuming applications.
An AI-ready semantic layer must improve efficiency by eliminating the need for AI systems to infer meaning from raw schemas. Context readiness improves first-response accuracy and reduces token consumption and the cost of AI workflows. It must also support any scale of data and any concurrency without latencies that create performance bottlenecks, allowing you to run AI at any scale.
These capabilities raise an important question: can existing semantic-layer architectures actually deliver them or were most designed for a bygone era of analytics?
Not All Semantic Layers Are Built for Enterprise AI
Many platforms exist that provide semantic consistency, but not all semantic layers are architected for enterprise AI. When evaluated against the demands of enterprise AI, those architectural differences determine what they can deliver in practice.
Vendors such as AtScale, Cube and dbt Labs help centralize definitions and reduce metric inconsistencies. However, they provide varying levels of governed business context, requiring AI systems to reconstruct meaning from metadata and raw schemas, increasing token usage and reducing efficiency.
Enterprise-scale performance also varies, making it difficult to sustain consistent performance as AI workloads grow. Their warehouse-dependent execution means AI systems must rely on the warehouse to resolve every query, increasing warehouse contention, latency and cloud compute costs.
Enterprise AI demands something more comprehensive: a universal semantic layer that combines governed business context, token economics, scale and enterprise-grade performance, together with cloud compute savings and AI token efficiency. This is the approach pioneered by Kyvos semantic layer.
Conclusion
Agentic AI promises significant gains in efficiency, automation and business innovation. Gartner predicts that by 2028, 15% of day-to-day work decisions will be made autonomously and one-third of enterprise applications will incorporate agentic AI.
Organizations that establish a governed, AI-ready semantic foundation today will be able to realize this value and extend AI reliably. In line with Gartner recommendations, rethinking workflows with agentic AI from the ground up is the ideal path to successful implementation.
ABOUT THE AUTHOR
Pratik Jain, Senior Director of Technology at Kyvos Insights, has over 20 years of experience in building high-performance analytics and AI platforms. He brings deep expertise in architecting and delivering scalable, enterprise-grade analytics products. As a Generative AI thought leader, he has driven innovation across multiple product suites and leads the UX/UI strategy at Kyvos. https://www.kyvosinsights.com/pratik-jain/
