AI Readiness Is Not a Model Problem. It's a Context Problem.
SAP's acquisition of Reltio signals a shift: enterprise AI readiness is now a data governance and unified-context infrastructure story — not a model selection one.
SAP just said the quiet part out loud
On March 26, 2026, SAP announced it would acquire Reltio, a master data management platform. The stated purpose: help customers "make their SAP and non-SAP enterprise data AI-ready."
That phrase is doing a lot of work. It is worth unpacking.
SAP's rationale was direct. "AI cannot reach its full potential when data is fragmented across business units, platforms and domains without connection or context," said Muhammad Alam, SAP's head of product and engineering. The acquisition will strengthen SAP Business Data Cloud and, in SAP's framing, supply the trusted data context that its AI agents — Joule and Joule Agents — need to actually work.
This is not just an MDM story. It is a signal that the enterprise software market has reached a conclusion: vague AI readiness talk has run its course, and the next phase requires real infrastructure.
Why the language is changing
For the past three years, "AI readiness" meant something fuzzy. Did you have a vendor contract? An AI policy? Had you run a pilot? Most large enterprises could check those boxes.
The results were not impressive. MIT's latest State of AI in Business research found 95% of companies report their generative AI initiatives are falling short of expectations (MIT/MLQ). Deloitte's 2026 enterprise AI report found a similar gap: organizations say their strategy is AI-ready, but confidence falls apart when the question turns to infrastructure, data quality, and risk management (Deloitte).
What the early copilot wave papered over is now impossible to ignore: the problem was never the model. The problem is what the model reads.
McKinsey data shows that while 88% of organizations are using AI in at least one business function, only about one-third have begun actually scaling AI programs (McKinsey). The bottleneck is not enthusiasm. It is operational foundation.
The market is now converging on a more concrete requirement set: unify fragmented data, enforce governance, maintain freshness across sources, and give AI systems a coherent operating context instead of another disconnected pile of records. SAP buying Reltio is one proof point. Informatica deepening its Microsoft Fabric integration is another. The pattern is consistent.
What "trusted context" actually means
Trusted context is not a technology. It is a set of conditions that have to hold before AI can reliably act on enterprise data.
Those conditions include:
Consistency. Revenue means the same thing to sales, finance, and delivery. Customer health scores use the same methodology. When different teams operate from different definitions, AI systems inherit that fragmentation and amplify it.
Freshness. The data reflects reality as of now, not as of six months ago when someone last ran a reconciliation job. Stale records cause stale answers. Stale answers cause bad decisions.
Shared definitions. "Golden record" is the MDM term — one authoritative version of a customer, product, or supplier that resolves all duplicates and conflicts. Without it, an AI agent processing supplier risk might see the same company three times and draw three different conclusions.
Governance. Someone owns the data. Changes are tracked. Decisions are auditable. In regulated environments, this is not optional — it is the prerequisite for deploying AI at all.
Reltio's platform does this for structured enterprise records using AI-based entity resolution. It creates a single, consistent view across SAP and non-SAP applications, with MCP support for real-time multi-agent workflows.
That is real infrastructure work. It is also half the picture.
Why agents make the problem more urgent
The first generation of enterprise AI tools were copilots. They surfaced suggestions, drafted text, summarized meetings. Humans reviewed the output before anything happened. Context failures produced embarrassing answers, not operational disasters.
Agents are different. They take actions. A procurement agent assessing supplier risk does not draft a memo for a human to review — it triggers a workflow. A customer service agent does not suggest a response — it sends one. An HR agent does not flag a policy question — it routes the employee to a conclusion.
When AI starts acting instead of assisting, fragmented context becomes an operational liability. As the agentic AI failure rate data shows, most agent failures trace back to knowledge and context problems, not model failures.
The math is uncomfortable. 43% of chief operations officers identify data quality as their most significant data priority, and more than a quarter of organizations estimate they lose over $5 million annually due to poor data quality (IBM Institute for Business Value). Autonomous systems operating on that data will not contain the damage — they will scale it.
The missing layer: unstructured enterprise knowledge
Here is where the MDM story stops being complete.
Reltio governs structured records: customer entities, product catalogs, supplier data, employee records. These are the data types that live in ERP systems and CRMs. They are important. They are also a fraction of what AI agents actually operate on in a real enterprise.
Real enterprise work depends on a different corpus: policies, SOPs, contracts, support documentation, compliance guides, onboarding materials, scanned PDFs from legacy systems, internal knowledge bases that no one has touched in 18 months. This is the unstructured layer. It is rarely governed with anything resembling the rigor applied to structured data. It contradicts itself routinely. It decays silently.
An agent built on perfect customer records but operating on a stale returns policy will still give customers the wrong answer. An agent with clean supplier data but outdated contract terms will still create compliance exposure. Master data discipline does not close that gap.
This is a problem we have tracked consistently: unstructured knowledge drifts, and most organizations have no mechanism to detect or correct that drift before agents act on it.
The enterprise AI context problem is bilateral. Structured records and unstructured knowledge both need governance. Most current infrastructure investments are addressing one side.
Platforms like Mojar AI approach this from the unstructured side — contradiction detection across documents, feedback-driven auto-remediation, and active knowledge maintenance so that what agents retrieve is current, consistent, and attributable. It is not an MDM play. It is the complementary layer that MDM leaves unaddressed.
What enterprises should do next
The SAP-Reltio announcement is useful because it names the actual problem clearly. Enterprise AI has a context problem. The market is starting to build real infrastructure around it.
For organizations mapping their AI readiness posture, the honest checklist is two-sided:
On the structured side: is your master data unified and consistently defined? Do your AI systems have one source of truth for customers, products, and suppliers, or are they reconciling conflicts on the fly?
On the unstructured side: are your policies, SOPs, and operational documents current? Do agents retrieving that content get consistent answers, or do contradictory versions exist across teams? Is there any mechanism to detect when that content has decayed or drifted out of alignment with reality?
Most organizations that run this audit will find they are further from readiness than their pilot results suggested.
The copilot era offered a useful illusion: because humans reviewed AI outputs, context failures stayed contained. The agentic era removes that buffer. Trusted context — governed, unified, fresh, and attributable — is the infrastructure requirement that comes next. The vendors are already building for it. The question is whether enterprises will build for it too, or wait until agents in production surface the gaps at scale.
Sources: SAP press release, MIT/MLQ State of AI in Business 2025, Deloitte 2026 Enterprise AI Report, McKinsey State of AI, IBM Institute for Business Value, Unite.AI — AI ROI, Data Wellness and Human Trust, InfoWorld — The starkly uneven reality of enterprise AI adoption
Frequently Asked Questions
AI-ready data is enterprise data that is unified, consistent, governed, and trustworthy enough for AI systems to act on it reliably. That means resolving duplicate records, enforcing shared definitions, maintaining freshness across sources, and ensuring that structured records and unstructured documents tell the same story.
Copilots surface suggestions that humans review before acting. Agents execute autonomously. When an agent routes work, generates a customer-facing output, or triggers a business process based on fragmented or stale context, there is no human in the loop to catch the error before it lands.
It solves part of it. MDM disciplines structured records: customers, products, suppliers. But real enterprise work also depends on policies, SOPs, contracts, and support documents — content that decays, contradicts itself, and is rarely governed with the same rigor. Agents operating on clean ERP data but stale policy documents still fail.