Most organizations discover their AI problem only after deployment. The model worked perfectly in testing. Demonstrations impressed stakeholders. The proof of concept delivered promised results. Then the organization attempted to scale the initiative across actual business operations, and everything ground to a halt.
This scenario plays out repeatedly across industries with depressing consistency. Research from RAND Corporation reveals that over 80% of AI projects fail—double the failure rate of traditional IT initiatives.[1] Gartner projects that 30% of generative AI programs will be abandoned post-proof-of-concept by end of 2025, citing inadequate data quality, insufficient risk controls, mounting costs, and ambiguous business value.[2] S&P Global's 2025 research shows 42% of companies abandoned most AI initiatives during the year, a sharp increase from 17% in 2024.[3]
Executives naturally search for technical explanations: inadequate models, insufficient computing power, poor vendor choices. But investigation after investigation reveals a more fundamental cause. Organizations lack the semantic infrastructure necessary to make AI accurate and reliable across the complexity of actual enterprise operations.
The Foundation Determines Everything
Information architecture—the semantic structures that establish meaning within enterprise content—forms the foundation on which every successful AI deployment must rest. This isn't a supporting concern or a nice-to-have enhancement. It's the prerequisite. Without it, AI initiatives produce impressive demonstrations that collapse under operational pressure.
This article examines what that foundation consists of, why it determines AI performance in production, how organizations identify the highest-value opportunities for building it, and what questions executives must answer before investing further in AI expansion.
Reframing the Enterprise AI Challenge
The enterprise AI discussion remains excessively focused on models. But generating language isn't the difficult part. The difficult part is consistently surfacing the correct knowledge—authoritative, current, and applicable—to the moment of need. Most AI failures in production aren't intelligence failures. They're application failures: incorrect source, outdated version, wrong context, ambiguous authority, absent traceability.
In business terms, the objective isn't generating responses. The objective is producing outcomes: reduced escalations, accelerated cycle times, maintained compliance, eliminated rework, decreased audit friction. Those outcomes depend on whether the system can answer the question that actually matters: not "what seems likely," but "what applies here, now, for this customer, product, and jurisdiction, based on authoritative sources."
When knowledge exists in fragments—multiple definitions, conflicting policies, unclear ownership, inadequate versioning—generative AI doesn't resolve the problem. It amplifies it. Organizations get speed without control: fluent output generating operational drag.
Why Model-Centric Thinking Fails at Scale
Benchmarks measure general capability. Enterprises operate within constraints. The gap becomes immediately visible in production: the system responds quickly, but subject matter experts spend time correcting its output. Citations appear credible but fail under scrutiny. Minor variations in question phrasing completely flip recommendations.
This explains why AI risk programs are evolving beyond model safety into system safety, encompassing governance, traceability, and managed lifecycle. When you cannot reconstruct how a system reached its conclusion, you lack a scalable enterprise capability. You have a liability.
Models are commodities. Application creates differentiation. Enterprises succeed when they make meaning executable: shared language, stable identities, explicit relationships, documented provenance, lifecycle governance. That semantic backbone transforms any capable model into an asset that survives audit, regulation, and operational edge cases.
Defining Semantic Architecture
Large language models retrieve content based on linguistic similarity, not semantic comprehension. An LLM doesn't understand what your terminology means, how your concepts relate, or which rules apply in specific business contexts. It matches language patterns.
Consider what occurs when your product taxonomy labels the same item three different ways across three different systems. An LLM doesn't resolve that inconsistency. It inherits it. The model may retrieve obsolete documentation, pull instructions for incorrect product variants, or combine content from incompatible regulatory jurisdictions. The outcome is answers that sound authoritative but create liability, undermine trust, and prevent adoption.
Semantic architecture is the structured foundation preventing these failures. It comprises four interconnected components, each serving a specific function in grounding AI for enterprise application.
Figure 1: The Four Components of Semantic Architecture (Source: Earley Information Science)
Component One: Taxonomies as Organization Systems
A taxonomy exceeds simple hierarchy. It's a controlled system for naming things, clustering related concepts, separating distinct concepts, and enforcing consistency. For AI systems, taxonomy determines which content belongs together and which must remain separate. It governs how variants cluster and how retrieval paths narrow.
In field service environments, taxonomy distinguishes product family from specific model, generation from variant, subsystem from component, failure mode from symptom. Without these distinctions, retrieval-augmented generation systems confuse procedures across similar equipment versions, delivering instructions that are close but operationally wrong.
Component Two: Ontologies as Relationship Logic
Ontology describes the semantic structure of the enterprise: how concepts relate in ways reflecting real operational meaning. Ontologies define "see also" relationships, cause-and-effect chains, dependency sequences, and rules for contextual interpretation. This product addresses this problem; this troubleshooting guide applies to this error code; these are conceptually related items, curated by human subject matter experts.
For AI, ontology is essential because it expresses the logic connecting concepts, maintains related ideas in connection, and prevents unrelated ideas from merging. In insurance contexts, ontology clarifies that claims relate to coverage, coverage depends on policy type, policy type varies by jurisdiction, and jurisdiction determines regulatory requirements. These relationships control retrieval and prevent model confusion. The ontology is the knowledge structure for the enterprise.
Component Three: Metadata as Retrieval Engine
Metadata drives retrieval more than any other factor. Well-designed metadata ensures AI never retrieves outdated content, unapproved drafts, content for incorrect product variants, or contradictory definitions. When a manufacturing procedure is tagged with model, variant, revision, and applicability attributes, AI retrieves precisely the correct instructions rather than everything linguistically similar.
Without metadata, retrieval accuracy cannot be governed. Without governed retrieval, every additional use case multiplies the risk of incorrect answers at scale.
Component Four: Knowledge Graphs and Controlled Vocabularies
Knowledge graphs connect these components to operational enterprise data. They link structured data to unstructured content, connecting the "what happened" from analytics to "why did it happen" from customer analysis. They map categories to content and data, connect controlled vocabulary terms across operational systems, and encode the pathways AI follows during information retrieval. Controlled vocabularies ensure that synonyms, abbreviations, and department-specific variations all resolve to identical canonical meaning.
When "deviation," "nonconformance," and "exception" carry different meanings in quality environments, vocabulary governance prevents AI from treating them as interchangeable. This matters enormously in regulated industries where language precision directly affects compliance.
Building Application Architecture Around Models
Applying a model isn't connecting it to a document repository. It's constructing a decision system around it, one that assembles context and enforces constraints. At minimum, applied enterprise AI must accomplish five things consistently.
Entity identification: Correctly identify the specific customer, product, policy, asset, or configuration—not a look-alike. This is entity resolution, foundational to reliable context assembly.
Source authority selection: Approved policies and procedures outrank training materials, drafts, and tribal knowledge. The system must enforce source precedence.
Applicability determination: Jurisdiction, effective date, product variant, customer segment, role, and channel all affect which information applies. These dimensions must be computable, not guessed.
Evidence provision: Cite the governing clause, version, and provenance enabling rapid human validation. In high-consequence environments, provenance isn't optional.
Safe failure: Refuse or escalate when evidence is insufficient, rather than improvising. Reliable systems know when they don't know.
These five capabilities are architectural. They sit above the model. They determine whether the model's general capability translates into specific business performance.
Enterprise Information Metabolism: The Competitive Advantage Engine
Every business process is a continuous loop of sensing, interpreting, and acting. Organizations capture data and signals from operations (sense), apply semantic structure and domain knowledge (interpret), and deliver insights enabling decisions (act). Your organization's speed and competitive position depend on how rapidly and reliably these loops execute.
Figure 2: The Enterprise Information Metabolism (Source: Earley Information Science)
LLMs and agentic AI can dramatically accelerate this metabolism, but only when the information they process is structured, trustworthy, and findable. Without semantic architecture, AI amplifies fragmentation rather than resolving it.
At every cycle point, friction slows flow. Data scattered across siloed systems means days spent chasing correct sources. Research confirms the pattern at granular levels: enterprise employees spend an average of three hours daily searching for information needed to perform their jobs, with 47% reporting that fragmented knowledge remains their biggest productivity obstacle.[4] The same item described differently across multiple ERPs means no shared vocabulary for AI to leverage. Manual handoffs and approval bottlenecks slow every decision. Teams routinely spend two to three days locating data sources and verifying accuracy before beginning analysis.
Figure 3: The Fractal Nature of Friction in the Enterprise Information Metabolism (Source: Earley Information Science)
You cannot layer AI atop friction and expect friction to disappear. AI accelerates whatever it touches, including dysfunction.
Identifying Information Leverage Points
The strategic question isn't where to apply AI broadly. It's where to find information leverage points: specific friction points where one targeted fix generates disproportionate downstream impact.
When you identify a bottleneck in the information metabolism, the approach follows three steps. First, map information flows and locate bottlenecks with largest downstream effects. Second, design targeted interventions: standardize vocabulary, automate handoffs, deploy agents where processes are well understood. Third, measure. Baseline before, measure after. If the intervention works, scale it. If not, that signals the need to examine underlying information architecture more deeply.
Agentic AI can eliminate entire manual steps from processes, but only when the process is understood and semantic infrastructure supports it. The foundations make leverage possible.
The Economic Reality: Knowledge Costs Compound
Model capability will continue improving, and inference costs tend to fall. But knowledge costs compound when meaning isn't managed: rework loops, escalations, audit remediation, duplicated assistants, and integration debt accumulate across the enterprise.
McKinsey estimates that generative AI could add the equivalent of $2.6 trillion to $4.4 trillion annually to the global economy, with total economic benefits (including broader productivity gains) reaching $6.1 trillion to $7.9 trillion per year.[5] But that value isn't unlocked by purchasing smarter models. It's unlocked by operationalizing knowledge: making it reusable, governed, and applicable across contexts. The same semantic backbone that powers a customer service assistant can ground an internal compliance tool, an engineering knowledge base, and an analytics pipeline. That reuse is where ROI compounds.
McKinsey's research confirms this pattern: organizations reporting significant financial returns from AI are twice as likely to have redesigned end-to-end workflows before selecting modeling techniques.[6] Meanwhile, MIT's 2025 study on the "GenAI Divide" found that roughly 95% of enterprise generative AI pilots produce no measurable profit-and-loss impact.[7] The gap between potential and reality isn't a technology gap. It's an architecture gap.
Assessing and Building Readiness
You cannot improve what you don't measure. You cannot automate what you don't understand. Organizations that scale AI successfully build foundational capabilities in specific sequence. And they assess where they stand before investing in expansion.
Building Foundations Sequentially
The five foundational capabilities for enterprise AI—governance, information architecture, knowledge engineering, data and content management, and operational readiness—must be built in order. Each layer enables and stabilizes the next.
Governance comes first, and it doesn't require heavyweight processes. A lightweight decision-making framework suffices to start. One organization implemented governance by requiring every new initiative to identify which taxonomies and controlled vocabularies it would use. That single requirement prevented inconsistent terminology from proliferating across AI projects. That was a gating factor managed through governance processes.
Information architecture follows, providing the vocabulary your AI uses to understand your business. Knowledge engineering transforms documents into machine-ready knowledge. Data and content management ensures knowledge remains accurate, current, and retrievable. Operational readiness embeds AI into real workflows with real measurement.
Each layer is a minimum viable program, not a heavy initiative. The key is starting lightweight, building on what works, and investing where leverage is highest.
The AI Readiness Framework
The AI Readiness framework evaluates organizational readiness across four interconnected domains: Knowledge Readiness, Operational Readiness, Technical Readiness, and Governance. Across these domains, comprehensive assessment examines multiple factors through diagnostic questions, providing a complete view of where an organization stands and what must be built next.
Each factor is scored on a five-level maturity scale from ad hoc to optimized, revealing specific gaps preventing scaling and the sequence in which they should be addressed. In our experience, Knowledge Readiness is consistently the weakest domain, which precisely explains why so many AI initiatives stall when moving from controlled pilots into messy enterprise reality.
Critical Questions Before Scaling
Before approving expansion of AI initiatives, leaders should answer seven questions about the systems they're building.
First: What decisions will the system influence, and what is the cost of being wrong?
Second: What is the authoritative source of truth for this domain, and what outranks what?
Third: Do we have effective dates, versioning, and jurisdiction overlays that the system can enforce?
Fourth: Can we resolve identity for core entities (customer, product, provider, asset)?
Fifth: Can the system show evidence (clause, version, provenance) for every high-consequence answer?
Sixth: What happens when evidence is insufficient: does the system refuse, escalate, or improvise?
Seventh: How will we measure drift and reconstruct outcomes for audit and incident response?
If answers to these questions are unclear, the organization isn't ready to scale. It's ready to build the foundation making scaling possible.
From Speed to Impact
Enterprise AI performance isn't a function of model intelligence alone. It's a function of whether you can apply the model inside a disciplined knowledge environment, one that enforces authority, applicability, evidence, and lifecycle governance. That is information architecture as an operating model. And it explains why there is no AI without IA.
Organizations with faster information metabolism out-decide and out-execute slower ones. In regulated industries such as financial services and life sciences, incorrect answers create liability, not inconvenience. Ungoverned AI in these environments represents safety and legal risk. Even in less regulated sectors, the cost of lost trust and abandoned adoption quietly erodes every dollar invested in AI capability.
The organizations that build semantic foundations will be the ones that capture enterprise-scale AI value. Not because they chose the right model, but because they built the architecture that makes any model effective. The path from pilot to production runs through information architecture. It starts with governance, builds through semantic structure, and scales through measurement. The investment isn't optional. It's the prerequisite.
Notes
[1] Ryseff, J., De Bruhl, B., & Newberry, S.J., The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed, RAND Corporation, RR-A2680-1, 2024.
[2] Gartner, Inc., "Gartner Predicts 30% of Generative AI Projects Will Be Abandoned After Proof of Concept By End of 2025," July 29, 2024.
[3] S&P Global Market Intelligence, "The Big Picture 2025: Generative AI," 2025.
[4] Coveo, "The Search for Relevance: Can AI Connect Employees to What Matters?" EX Relevance Report, April 2025.
[5] McKinsey & Company, "The Economic Potential of Generative AI: The Next Productivity Frontier," June 2023.
[6] McKinsey Global Survey on AI, 2025.
[7] MIT NANDA Initiative, "The GenAI Divide: State of AI in Business 2025."
Note: This article was originally published on CustomerThink and has been revised for Earley.com.