Every enterprise AI project follows the same arc. The excitement phase comes first — this model can reason, this agent can execute, this system can automate. Then comes the deployment phase, where the AI meets reality. And reality, in every enterprise we have worked with, looks like this: the AI is ready. The data is not.
SAP just made the most expensive acknowledgment of this reality in enterprise software history.
Last week, SAP announced it would acquire Reltio, a master data management company that specialises in unifying, cleansing, and harmonising data across complex, multi-vendor enterprise environments. Terms were not disclosed, but the strategic significance was made explicit: Reltio will become a core capability within SAP Business Data Cloud, the platform integral to SAP's AI-first strategy.
SAP's executive board member Muhammad Alam framed the acquisition in terms that every enterprise AI team will recognise: “Enterprise AI needs trusted context that is open and interoperable across the heterogeneous IT landscapes our customers run.”
Trusted context. Not a better model. Not a faster chip. Not a more sophisticated agent framework. Trusted data — clean, unified, consistent, and accessible across systems — is what SAP identified as the missing piece in its entire AI-first strategy.
This is the most important enterprise AI story of the week. Not because of the acquisition itself, but because of what it confirms about where enterprise AI projects actually fail.
The Data Problem That 48% of Enterprises Cannot Solve
Nvidia's 2026 State of AI survey asked 3,200 enterprise respondents to identify their top barrier to scaling AI. The answer was not model capability, not compute costs, not regulatory uncertainty, not talent shortages.
It was data. 48% cited insufficient data or data-related issues as their primary barrier.
The specific data challenges that enterprises face are consistent across every industry and region we work in.
Data fragmentation. The average enterprise runs hundreds of applications, each with its own data format, its own schema, and its own definition of basic entities like “customer,” “product,” and “order.” A customer record in the CRM does not match the customer record in the ERP, which does not match the customer record in the billing system. The same customer appears as three different entities across three systems.
Data inconsistency. When the same entity exists in multiple systems with different attributes — different addresses, different contact information, different account statuses — no AI system can determine which version is correct without human intervention. AI agents that make decisions based on inconsistent data make inconsistent decisions.
Data duplication. Duplicate records accumulate over years of mergers, migrations, and manual data entry. An enterprise with 500,000 customer records may have 100,000 duplicates — records that represent the same customer but cannot be automatically matched because the data varies slightly across systems.
Data staleness. Enterprise data degrades over time. Addresses change. Contact information expires. Organisational structures evolve. Data that was accurate when entered becomes misleading if not continuously maintained.
Data inaccessibility. Critical data often resides in systems that AI agents cannot access — legacy databases, proprietary formats, on-premises systems without APIs, spreadsheets stored on individual machines. The data exists but the AI cannot reach it.
Each of these problems individually degrades AI performance. Together, they explain why 48% of enterprises identify data as their top barrier — and why the most sophisticated AI model in the world delivers mediocre results when fed mediocre data.
Why This Matters for AI Agents Specifically
The data quality problem becomes exponentially worse in the agentic AI era.
A traditional AI system processes a query and returns a response. If the data is imperfect, the response may be imperfect — but the scope of impact is limited to that single interaction.
An AI agent operates differently. It receives a goal, breaks it into steps, accesses multiple systems, makes decisions, and executes actions — often without human review of each step. A procurement agent that assesses supplier risk needs to access supplier data from the ERP, contract data from the legal system, financial data from the accounting platform, and performance data from the operations database. If any of those data sources contain inconsistencies, the agent's risk assessment is compromised — and its autonomous actions may be based on flawed analysis.
SAP's announcement specifically cited this agentic use case: “Low latency delivery and support for the Model Context Protocol enable real-time, multiagent workflows across SAP and non-SAP environments, allowing AI agents, such as a procurement agent, to assess supplier risk and trigger actions almost instantly using trusted, real-time data.”
The key phrase is “trusted, real-time data.” Not “data.” Trusted data — data that has been unified, cleansed, deduplicated, and validated as the authoritative record across systems. That distinction is the entire value proposition of the acquisition.
The analyst community reinforced this point. An Info-Tech Research analyst noted that SAP is moving from being a “system of record” to a “system of context” — providing AI agents not just with data, but with the unified, trusted context they need to make reliable decisions across the enterprise.
What Reltio Actually Does
Reltio's core capability is entity resolution — the process of identifying that records from different systems represent the same real-world entity and merging them into a single, authoritative record.
The platform ingests data from multiple sources, across SAP and non-SAP systems, in real time. It uses AI-based matching logic to identify related records — even when the data varies between systems. It merges those records into what the industry calls “golden records” — single, unified representations of each entity that combine the most accurate and current attributes from all source systems.
The result is a unified view of customers, products, suppliers, locations, and employees that any AI agent can access as the authoritative source. When a procurement agent queries supplier data, it gets one consistent record — not three conflicting records from three different systems.
Reltio also handles data governance — ensuring that data quality is maintained continuously, not just at the point of initial cleansing. As new data enters the enterprise through any channel, the platform validates, matches, and integrates it into the unified record in real time.
For enterprises operating across complex, multi-vendor IT environments — which describes virtually every enterprise in the Gulf and globally — this capability is foundational for AI deployment. You cannot build reliable AI agents on unreliable data. You cannot automate decisions if the data those decisions depend on is inconsistent across systems.
The Lesson for Every Enterprise
SAP is a $43 billion annual revenue company with 400,000 customers across every industry. Its entire AI-first strategy depends on data quality — and it concluded that it needed to acquire a specialised company to solve the problem because its existing capabilities were insufficient.
If SAP, with its resources and its position at the centre of enterprise data, determined that data readiness for AI requires a dedicated acquisition, the lesson for every enterprise is clear: data preparation is not a preliminary step that you complete before your AI project begins. It is a continuous, specialised capability that requires dedicated investment.
The enterprises that treat data quality as a one-time cleanup project before deploying AI will face the same problems that led SAP to make this acquisition: AI systems that underperform because the data they consume is fragmented, inconsistent, and untrustworthy.
The enterprises that invest in continuous data unification — maintaining golden records across systems, validating data in real time, and governing data quality as an ongoing operational function — will have AI agents that make reliable decisions because they operate on reliable data.
What to Do This Quarter
Audit your data readiness. Before evaluating AI models or agent frameworks, assess the quality of the data your AI systems will consume. How many systems contain customer data? Are those records consistent? How many duplicates exist? What percentage of records are current? Where is critical data inaccessible to your AI infrastructure?
Invest in data unification before AI deployment. The most common enterprise AI failure pattern is deploying sophisticated AI on top of poor data. Invest in master data management, entity resolution, and continuous data quality before — or at minimum, alongside — AI deployment.
Design for multi-source data from the start. Enterprise AI that only accesses data from a single system delivers a fraction of its potential value. Design your AI architecture to access unified data across all enterprise systems — ERP, CRM, billing, operations, compliance — through a single trusted source.
Treat data quality as an operational function, not a project. Data degrades continuously. Records become stale. New duplicates are created. Schemas change. Data quality must be maintained continuously through automated validation, matching, and governance — not through periodic cleanup projects.
Evaluate your ERP migration timeline. If your enterprise is planning or executing a migration to a modern ERP platform, factor AI data readiness into the migration strategy. The same data unification work that prepares data for a new ERP also prepares it for AI deployment. Doing both together is more efficient than doing them sequentially.
SAP just told the enterprise world that the foundation of its AI-first strategy is not a better model. It is better data. That is the most honest assessment of enterprise AI readiness that any major technology company has delivered in 2026.
“48% of enterprises cite data as their top barrier to scaling AI. SAP just acquired Reltio to solve that problem for its 400,000 customers. The message is unambiguous: enterprise AI does not fail because the models are not good enough. It fails because the data is not ready. Fragmented across systems. Inconsistent between records. Duplicated across databases. Stale from years without maintenance. The most sophisticated AI agent in the world makes poor decisions on poor data. Data readiness is not a preliminary step. It is the foundation that determines whether enterprise AI succeeds or fails.”