Ensuring that a data platform is AI-ready means understanding the bottlenecks that may stand in the way of an AI-enabled environment. Often, they’re not related to the technology, but the processes and foundations that are necessary to support a relentless surge of AI initiatives.
AI is a work in progress, and there is a lot that still needs to be done to prepare a platform or environment for the AI wave now crashing on enterprise shores. Data comes in from many sources and is constantly subject to change and review. It tends to be highly dispersed and siloed across various systems and documents within typical organizations.
As a result, the professionals working with data—data engineers, administrators, analysts, and scientists—need to spend an inordinate amount of time vetting, resolving, and rationalizing their available data. The key is to build and maintain a data environment that is forged in governance, provides scalability, and ensures future-proofing.
The challenge is that many organizations’ data foundations are not ready for this new world. Most enterprises report half or fewer of their AI initiatives succeeding, a survey of 259 data executives by Unisphere Research and Radiant Advisors (sponsored by Quest) finds. The root causes include data and the gaps in quality, access, trust, and lineage that continue to limit outcomes and prevent AI from scaling.
Another challenge is the fact that while enterprise AI initiatives are being well-funded, architecture lags behind, the aforementioned survey finds. “[O]rganizations approve a significant AI budget, then spend 12 to 18 months discovering that the budget alone cannot manufacture the conditions for AI to operate at scale,” according to John O’Brien, CEO of Radiant Advisors and author of the report. “If your organization is committing $1M or more to 2026 AI initiatives and has not yet built the underlying readiness stack, the most important architectural decision … is to fund the stack before scaling the deployment,” the Unisphere/Radiant report states. “The data is clear that the alternative is to join the 28% reporting more than $1M in losses on failed AI initiatives.”
What actions should be taken now to ensure that an organization’s data environment is ready for the rigors of AI? Advice and analysis from McKinsey point to the essential practices that can help prepare for this new world. “None of this happens by magic,” according to the article. “Underneath, AI systems pull data apart and put it back together again and again—across documents, systems, prompts, and workflows,” states McKinsey. “Scaling AI means ensuring that all this data is treated as truth, so agents can act responsibly and users can trust outputs. Getting to this truth is anything but easy.”
Platforms must be capable of supporting complex workloads that accommodate structured data, unstructured data, real-time processing, vector search, and semantic search. They must enable large language models and agents “to use data consistently, especially where more deterministic behavior is required,” McKinsey advises. Let’s look at some key practices needed to achieve this.
Self-assess across three key layers—governance discipline, data foundation, and operating architecture. “AI readiness is not a score on one survey question,” the Unisphere/Radiant report observes. “It is the integrated state of three architectural layers. An organization can rate itself highly on the data foundation and still underperform if its governance discipline remains informal. The diagnostic question is which layer your organization has not yet operationalized, and the answer is the highest-return architectural investment available. …”
Build on the weakest layer, not the strongest. Successful data environments for AI implementations “favor breadth across the stack over depth in any single layer,” Radiant’s O’Brien states. Organizations with just one strong layer of the three key layers—governance, foundation, and architecture—“only achieve modest outcomes. Organizations with three layers at operational status achieve compound outcomes. If your data foundation is strong but governance is not yet operational, the … priority is governance, not deeper data foundation work.”
Work toward operational status, not completed frameworks. The single biggest jump to AI success “is from designed to operational, not from informal to enterprise-wide,” the Unisphere/Radiant study finds. “Operational semantic definitions outperform designed semantic governance. Operational orchestration outperforms planned orchestration. Cross the operational threshold for each layer, then deepen.”
Ensure an AI-ready architecture. “Traditional data architectures were designed for stable data sets and predictable workflows,” the McKinsey article states. AI systems “dynamically retrieve and generate information across many systems in real time.”
Put observability in place. With AI, any visibility that now governs existing systems and data needs to extend further. “It must make the assembly of context and the production of outputs observable, not just the movement of data,” according to McKinsey.
Emphasize quality in data. Traditional data quality controls are not enough when AI enters the picture. “Extracted objects, chunks, and embeddings become the artifacts that influence outputs,” McKinsey states. “Quality must therefore be maintained not only at ingestion but also across extraction, chunking, retrieval, and generation. A document may be accurate in full yet produce incorrect answers if outdated or incomplete fragments are retrieved.”
Take metadata management to the next level. With AI, data environments “need to move beyond managing files and folders to managing reusable data objects with fine-grained schemas,” observes McKinsey. “For example, in an audio transcript, these objects might include speakers, clauses, time stamps, and attributes. Enterprise graphs must anchor unstructured artifacts to structured enterprise data and core entities such as customers, contracts, products, assets, policies, and transactions.”
Take a deeper look at data lineage. While organizations have been focusing on data lineage more intensely in recent years, AI makes the process more complex. “It must track which version of a document was indexed, how it was segmented, which chunks were retrieved, how prompts were constructed, and which tools were invoked,” according to McKinsey. “Without this artifact-level traceability, the organization cannot explain how an answer was produced, assess the impact of updating a document, or confidently manage change.”
Revisit platform and tooling architectures. Standardization needs to be the rule here, across all functions. “If each team builds that pipeline independently, duplication multiplies,” McKinsey asserts.
Standardize the integration of unstructured data. It’s essential that a well-functioning and AI-ready data platform brings unstructured data into the fold. “For example, every contract PDF is converted into the approved schema, with named entities, sensitivity tags, and lineage captured before any model can use it,” McKinsey states.
Establish shared, reusable foundation services. “Core capabilities such as governed retrieval layers, reusable tool and skill services, runtime policy enforcement across prompts and embeddings, artifact-level lineage tracking, and observability must be built once and reused across applications,” the McKinsey article urges.
The most successful AI initiatives are not the ones with the biggest budgets, but rather the ones that have developed processes and solutions that meet the enhanced foundation and governance required for AI. This data foundation minimizes the bottlenecks surrounding AI efforts while providing AI systems and agents with context, traceability, and a well-functioning operational environment.