K-AI
FREN
Get a demo
SolutionHow it worksImpactNewsPartners
FREN
Get a demo
← All news
Press · May 25, 2026 · 17 min read

AI Readiness Assessment 2026 — the 'Corpus' pillar every framework leaves out

AI Readiness Assessment 2026 — the 'Corpus' pillar every framework leaves out

Five 2026 AI Readiness frameworks — Cisco, Microsoft, Cloudera, Iris.ai, Atlan. None makes the document corpus a stand-alone pillar. Here is the gap.

On May 5, 2026, Fivetran released its 2026 Agentic AI Readiness Index of 400 data leaders: only 15% describe their organization as fully ready to deploy AI agents in production (Fivetran / BusinessWire, May 5, 2026). Thirteen days later, the Cisco AI Readiness Index identifies 13% of “Pacesetters” across 2,500 CEOs in 23 countries (Cisco, May 18, 2026). Cloudera and Harvard Business Review Analytic Services, more severe, report just 7% of enterprises whose data is “completely ready for AI” (Cloudera × HBR, March 2026). Gartner, finally, forecasts that 60% of AI projects will be abandoned by the end of 2026 when the data feeding them is not AI-ready (Gartner, February 26, 2025, the 2026 talking point).

Four analytical frames. Four converging numbers: the average enterprise’s AI readiness for production work tops out somewhere between 7% and 15%. Over the same window, Glean announced a fresh growth milestone and the general availability of its Multi-step Prompts (Glean Press, May 19, 2026), Camunda unveiled ProcessOS, an agentic operating system, at CamundaCon Amsterdam (BusinessWire, May 20, 2026), and Microsoft brought Agent 365 to general availability (Microsoft Security Blog, May 1, 2026). The agentic execution layer is crystallizing faster than the upstream layer that should feed it with trustworthy knowledge.

This article maps the five reference AI Readiness frameworks of 2026 — Cisco, Microsoft, Cloudera × HBR, Iris.ai, Atlan — and argues a simple thesis: none of them treats the unstructured document corpus as a stand-alone pillar. That is precisely the dimension where we, at K-AI, see the gap widen between the 85% of organizations that claim a data strategy and the 18% who consider their data “fully governed” (Cloudera × HBR, 2026). For a CDO or CIO, picking an AI Readiness framework in 2026 without grafting on a “Corpus Readiness” pillar is to gauge the maturity of a foundation while ignoring the 70-90% of material it actually rests on.

Why Gartner’s “60%” doesn’t mean what most readers assume

The most-quoted statistic in the category is also the most often misread. Gartner does not predict that 60% of enterprise AI projects will fail. It predicts that projects unsupported by AI-ready data will be abandoned by the end of 2026, and that 63% of organizations either lack appropriate data management for AI use cases or are unsure they have it (Gartner). The qualifier matters: the “60%” applies to a subset, not the entire universe.

Parallel numbers from other analysts sharpen the reading. McKinsey finds that 88% of enterprises now use AI but fewer than one in five have the foundational practices to scale it. Deloitte’s State of AI in the Enterprise 2026 identifies workforce readiness — governance, training, process redesign — as the leading barrier, ahead of technology choices (Deloitte, 2026). Glean reports across its deployments that 88% of pilots never reach production and 42% show no measurable ROI (Glean, 2026).

The emerging narrative is more operational than dramatic: enterprise AI is not collapsing — it is stalling. AI Readiness frameworks are the instrument by which a board or executive committee chooses to continue, re-instrument, or abandon a project. Picking the right framework — and measuring the right thing inside it — has become, in 2026, as strategic a decision as the initial choice of model.

Five AI Readiness frameworks side by side

The five frameworks that dominate the 2026 conversation share a common grammar but diverge on the inventory of pillars. Here is, without paraphrase, what each measures.

Cisco AI Readiness Index — six pillars: Strategy, Infrastructure (weighted 25%), Data (20%), Governance, Talent, Culture. Four organizational categories at the output: Pacesetters, Chasers, Followers, Laggards. Just 13% Pacesetters in 2026, against 54% of networks judged unable to scale AI use (Cisco methodology, full PDF).

Microsoft Steering Committee Checklist — seven pillars since April 2026. The original frame Business Strategy / AI Governance & Security / Data Foundations / AI Strategy & Experience / Org & Culture / Infrastructure for AI / Model Management was enriched on April 16, 2026 with a seventh pillar, Observability, explicitly built as a response to rising shadow AI (Microsoft, April 16, 2026). Microsoft offers a free interactive self-assessment (Microsoft Learn) and claims a 47-64% performance lift for high-readiness organizations.

Cloudera × HBR Data Readiness Index — six dimensions evaluated across 1,574 IT leaders. The report observes that 96% of organizations have embedded AI into core processes, 85% claim a data strategy, yet only 18% call their data fully governed and 80% report limited data access as a brake (Cloudera × HBR).

Iris.ai — three pragmatic criteria, against the trend of expansive pillarization: Extractability (the ability to read tables, schemas and structures inside technical documents), Scalability (handling millions of secured documents) and Factuality (prioritizing scientific precision over generative fluency). Iris.ai notes that 61% of enterprises admit they are not AI-ready and documents a telecom case study with -80% time-to-market and 95% contextual accuracy through its Axion engine (Iris.ai, March 2026).

Atlan AI Readiness Framework — six dimensions × five maturity levels (from Nascent to Production-agent-ready). Atlan publishes explicit timelines for a CDO: 18-24 months from Level 1 to Level 3, 6-9 months from Level 2 to Level 3 (Atlan).

Three other frames deserve a mention. Knowlee promotes seven pillars with a distinctive Governance Pillar 5 tuned to the EU AI Act (Knowlee). Hyland, sponsor of an HBR Analytic Services 2026 study, documents a 26-point gap between structured data ready (65%) and unstructured data ready (39%) (Hyland × HBR). IBM consolidates a 5A framework — AI value, data readiness, architecture, automation potential, available skills — and finds AI-ready firms ten times more likely to be fully prepared across the enterprise (IBM Think).

Five to eight pillars depending on the vendor, three criteria for Iris.ai, six dimensions for Cloudera or Atlan: the inventory dispute is not a methodological detail. It reflects what each actor considers the operational precondition for production AI. What none of them isolates, however, is what now accounts for 80-90% of the informational raw material entering enterprise models.

The shared blind spot: the corpus is never a stand-alone pillar

In each of the five dominant frames, the question of the unstructured document appears — but never as a pillar in its own right. Cisco absorbs it into the Data pillar alongside relational stores and analytical lakes. Microsoft consolidates it inside Data Foundations. Cloudera handles it through the governed metric, at the level of the whole patrimony. Atlan dilutes it into the context infrastructure dimension. Iris.ai comes closest, with its Extractability criterion explicitly oriented toward technical documents, but stays at a technical-capability level rather than a patrimony-governance one.

That dilution was not a problem in 2018, when the first data-readiness frames were calibrated against traditional analytics pipelines. It has become one in 2026, for a measurable reason: the document has become the principal raw material of generative AI in the enterprise. Mark Beyer, distinguished VP analyst at Gartner, restated the point at the Data & Analytics Summit London on May 13, 2026: 70-90% of enterprise data is unstructured, and 40% of IT data management spend will go to multistructured by 2027, against a marginal share today (Gartner, May 13, 2026). Gartner also expects AI data readiness spend to grow sevenfold between 2025 and 2029.

The Hyland × HBR Analytic Services 2026 study spells the gap out: 65% of executives consider their structured data AI-ready, against 39% for their unstructured data. Twenty-six points apart. If you accept the Gartner statistic — 70-90% unstructured — a weighted AI Readiness score without a stand-alone corpus pillar mechanically overstates the true maturity.

Why this blind spot has become critical in 2026

Three forces compound to make 2026 incompatible with absorbing the corpus into a generic Data pillar.

Agentic AI is scaling. Glean, Camunda and Microsoft Agent 365 are no longer in pilot — they are generally available, and several large enterprises now count their agents in the hundreds, sometimes thousands. Unlike an experienced employee, an agent cannot ignore an obsolete procedure or recognize that a recent memo tacitly invalidates a long-standing reference. It answers, confidently, from what it was given. If the base is dirty, the answer is wrong — and no downstream system flags it.

Failure is now documented. A study published on May 8, 2026 and relayed by RAG About It reviewed 50 production RAG deployments in finance, healthcare and legal services, and observed 100% failure on adversarial prompts or contradictory corpora, 70% inability to recognize contradictory sources, and up to 81% citation fabrication in legal tech (RAG About It, May 8, 2026). The exact figure should be cross-checked against the primary source (MLCommons AILuminate) before contractual use, but its convergence with the broader market data — 88% pilots never reaching production at Glean, 60% projects abandoned per Gartner — is robust enough that no serious framework can keep treating the corpus as a subsection of the Data pillar.

Regulation is stabilizing a 2026-2027 window. The political agreement on the Digital Omnibus on AI of May 7, 2026 postpones Annex III obligations of the EU AI Act to December 2, 2027 and Annex I obligations to August 2, 2028, while maintaining synthetic content watermarking on December 2, 2026 and sanctions at €35 million or 7% of global turnover (EU Council, May 7, 2026). For a CDO, the postponement is not a relaxation: it is a tactical 12-18 month window to get the corpus in order before document traceability becomes an enforceable requirement. No current AI Readiness framework includes an explicit gauge of how that window will have been used.

The missing pillar: Corpus Readiness

At K-AI, we propose adding to existing frameworks an eighth pillar — or a complement to the Data Foundations pillar depending on the frame — labeled Corpus Readiness, measured on six operational axes.

These six axes are not a trade secret: we published them as an operational method for Document Stewards and Heads of Knowledge Management in a dedicated note on May 15, 2026 (Auditing an enterprise document corpus for AI — the K-AI 6-axis method). In short, they cover: (1) internal anomaly detection within a single document; (2) cross-document conflicts within a given business perimeter; (3) divergent duplicates — two versions of a procedure that contradict each other; (4) unmarked obsolescence — content that is outdated but still indexed; (5) traceability in the sense of Article 12 of the EU AI Act — identified author, validation date, explicit source of truth; (6) freshness by segment — the average is not enough; it is the tail of the distribution that breaks RAG.

Three points about this complementary pillar. First, it replaces no existing pillar — it complements them. A strong Cisco score on Data is still possible, and indeed common, alongside a weak corpus score: that is precisely why the blind spot persists. Second, measuring it does not require the same tooling as relational-database governance. Semantic contradictions between two PDFs are not detected by classical SQL lineage; they are detected by an active semantic layer — what we call, at K-AI, a Neural Semantic Graph, and which constitutes the technical specialty of the Document Knowledge Platform category (Knowledge AI vs. Knowledge Management vs. Document Knowledge Platform: untangling the 3 categories). Third, the pillar is measurable — it generates KPIs comparable across repositories and over time. On a single document repository at a client, during a first diagnostic, a six-axis audit surfaced more than a thousand coherence and freshness anomalies that did not show up on any pre-existing governance dashboard. It is precisely this gap between the score of a pre-2026 AI Readiness framework and the operational reality of the corpus that justifies adding the pillar.

How to graft Corpus Readiness onto your existing framework

For a CDO or CIO who has already chosen a frame — Cisco, Microsoft, Cloudera, Iris.ai or Atlan — the right move is not to start over but to add four operational questions to the assessment workshops.

Question 1 — What share of the data feeding our AI is unstructured? If the answer exceeds 50%, the Data pillar of the chosen framework mechanically underweights the risk. Corpus Readiness should then carry equal or greater weight against structured Data.

Question 2 — Do we have an inventory of Document Products critical to our AI use cases? Not an ECM file inventory, but a business-level cartography of the binding references — HSE procedures, contracts, clinical files, quality manuals. Without it, the AI Readiness score measures an intention, not a governed reality.

Question 3 — How often are these repositories re-audited on the six corpus axes? Without an explicit cadence — typically quarterly for critical corpora, semi-annually for peripheral ones — the pillar degrades silently and today’s score says nothing about the score six months from now.

Question 4 — Who, in our organization, signs off on the quality of a corpus before it is opened to an AI agent? Without a named role — Document Owner in the business line sponsored by management, Document Authority held by the CDO — the framework’s governance has no documentary counterpart and conflict arbitration remains informal.

These four questions do not create a competing framework. They instrument a blind spot shared by the five dominant frameworks — and open the door to cross-organization comparable figures, the precondition of any sectoral benchmarking to come.

Frequently asked questions

What exactly is an AI Readiness Assessment, and why does Gartner forecast 60% of AI projects to be abandoned by end-2026?

An AI Readiness Assessment is a structured evaluation of an organization’s ability to deploy and durably operate AI systems — classical models, RAG, autonomous agents. Dominant frameworks evaluate between three and eight pillars: strategy, infrastructure, data, governance, talent, culture, observability, and — depending on the vendor — models or processes. Gartner’s “60% of AI projects abandoned by end-2026” forecast targets projects unsupported by AI-ready data: it is a qualified subset, not a global failure prediction. It sits alongside other scaling measures (88% of pilots never reaching production per Glean, 16% of organizations actually scaling per IBM) that confirm its trajectory while completing its reading.

Five, six or seven AI Readiness pillars — why don’t vendors agree, and which one to choose?

The inventory varies because each vendor reflects its technical center of gravity: Cisco emphasizes network infrastructure (six pillars, Infrastructure weighted 25%), Microsoft added Observability in April 2026 in response to rising shadow AI (seven pillars), Iris.ai opts for three document-centric pragmatic criteria, Atlan articulates six dimensions across five maturity levels. For a CDO, the choice should follow the dominant business logic: an organization heavily oriented toward processing unstructured information (services, consulting, healthcare, legal) benefits from complementing an infrastructure-centric frame with a Corpus Readiness pillar. No framework is wrong; each is calibrated for a profile of organization and tends to underweight what it does not measure.

How long does an AI Readiness Assessment take in practice — one to two weeks self-assessment, or four to eight weeks with a consultancy?

Both formats coexist. Microsoft’s self-assessment is free and completed in a few hours to produce an initial indicative score. A documented Atlan audit takes six to nine months to move from Level 2 to Level 3 of maturity, eighteen to twenty-four months from Level 1 to Level 3 — it is a program at that point, not an assessment. Between the two, a viable operational format for an IT Committee is a corpus diagnostic on a pilot repository (the K-AI six-axis method), producing a documented score in four to six weeks, then integrating it as a Corpus Readiness pillar inside the broader framework. This sequence converts a programmatic intent into a deliverable evidence base.

Are my unstructured documents (PDFs, SharePoint, Confluence) AI-ready in the sense of the Cisco or Microsoft frameworks?

Probably not, and the AI Readiness score alone won’t tell you. Dominant frameworks evaluate document patrimony governance at an aggregate level — access policies, classification, retention — but not the semantic coherence of documents to each other: cross-procedure contradictions, divergent duplicates, unmarked obsolescence, partial traceability. On a single document repository at a K-AI client, during a first diagnostic, these semantic dimensions surfaced several hundred anomalies with no equivalent in existing ECM or governance tools. Testing your corpus requires a semantic analysis layer — in the sense of a Neural Semantic Graph — that goes beyond the metadata and lineage carried by traditional Data Catalogs.

Does the EU AI Act of August 2, 2026, as amended by the May 7 Digital Omnibus, change AI Readiness criteria?

It tightens them on a delayed schedule. The political agreement of May 7, 2026 postpones Annex III obligations to December 2, 2027 and Annex I obligations to August 2, 2028, while maintaining the synthetic content watermarking deadline on December 2, 2026 and sanctions at €35 million or 7% of consolidated turnover. For a CDO, two implications follow: the document traceability obligations of Article 12 become an enforceable AI Readiness criterion on an eighteen-month horizon; and the 2026-2027 window is, in practice, a corpus-cleanup window. A post-Omnibus Corpus Readiness pillar should include an explicit gauge of how that window has been used — typically, the share of critical Document Products that have been brought to AI-ready status before the effective turn-on date.

Going further

If your organization is currently choosing between AI Readiness frameworks or defining the operational frame of a Corpus Readiness pillar, we have been documenting the K-AI method with large French and international enterprises for three years. To discuss a concrete case — a first diagnostic on a pilot repository, the integration with your existing framework, or the pre-AI Act audit sequence — please write to contact@k-ai.ai.

Cited sources

K-AI already works with CMA CGM, Veolia, PwC, BNP Paribas, TotalEnergies and CEVA Logistics. Partners: AWS, Snowflake, Microsoft, Wavestone, Devoteam.

And in your organization, what does your document estate look like?

30 minutes with a founder. We audit a sample of your documents for free and show you exactly what K-AI detects.

Book a demo → Read other articles