AI is generating big hopes and even bigger investments. But many initiatives aren’t making it past the pilot stage. A widely quoted study by MIT reported in late August that 95% of AI initiatives stall before moving beyond the pilot stage. Bain’s research finds that a third or more pilots move on to production, depending on the use case. But both studies find that most use cases don’t advance past the pilot.

The fundamental gap is not necessarily in the capabilities of AI models but rather in deficiencies regarding how they’re deployed. Many companies have yet to invest—or are just beginning to invest—in critical enablers for AI value realization, including end-to-end process redesign, disciplined AI governance, solid change management, executive commitment, and an effective data strategy.

Weaknesses in managing organizational data—including poor data quality, inconsistency, weak compliance, and insufficient accessibility—have dogged deployment of digital initiatives since well before the AI age. Pilots often succeed because they’re built on offline, nonproduction data sets that have been manually cleaned. But when it comes time to scale those pilots across the enterprise, underlying data issues quickly resurface, slowing or even halting progress.

With the advent of AI foundational models, there were hopes that AI would become so sophisticated and capable at handling messy and unstructured data that managing and governing data quality would be a thing of the past. That may still happen in the future, but we’re far from that today. While AI can assist with discrete elements (such as identifying quality issues or helping flag inconsistencies), the basic rule of “garbage in, garbage out” remains a feature of AI as much as any other digital solution.

The hard work of building a strong data foundation matters, and it’s more important than ever.

AI has made data more valuable but also more complex. Generative AI makes use of structured and unstructured data, including audio, images, and video. Most organizations haven’t historically governed unstructured data, resulting in some significant data quality challenges. For example, information retrieval in contact centers, particularly in complex enterprise environments, often run into issues with outdated or conflicting sources of information for the same prompt, resulting in inaccurate answers from AI.

As organizations deploy agentic AI, this foundation becomes nonnegotiable. These agents don’t just analyze data; they act on it, powering workflows, making decisions, and handling customer tasks autonomously. Without reliable, well-governed data as a single source of truth, agentic AI risks acting on flawed inputs, undermining both performance and trust.

The principles of good data strategy and governance are well established, with clear best practices for how to develop a robust strategy within both centralized and decentralized organizations. Now is the time to reinvigorate and enhance those principles and practices to ensure successful AI deployment (see Figure 1).

Figure 1

The legacy barriers holding back data strategy

As data becomes more integral to business performance, underlying challenges such as fragmentation, complexity, and misalignment become harder to ignore. Solving them requires a shift from legacy thinking to enterprise-wide data strategy and ownership. But many organizations are still stuck among a range of roadblocks.

• AI efforts frequently launch as standalone initiatives, only to realize that the data demands far exceed what’s needed for traditional reporting. Without a coordinated data strategy, progress stalls.
• Data lakes have focused on collecting large volumes of data, but many have become complex, monolithic platforms in which data quality is hard to manage and useful data sets are difficult to find and use effectively.
• Ownership is often unclear, defaulting to system administrators or data platform teams. Without business-aligned ownership, governance lacks direction and fails to connect to real-world needs.
• Governance may be informal or limited to a few core data sets. But today’s analytics and AI initiatives require broader, more robust governance that is jointly managed by both the business and data technology.

Data strategy foundations for scaling AI

AI is only as strong as the data behind it. Leading organizations treat data like the strategic asset it is—prioritizing value, establishing clear ownership, and building the architecture and governance needed to turn high-quality data into a lasting source of competitive advantage. Successful transformations share several important principles.

• Prioritization tied to value: Clarify which data assets create a competitive advantage, and develop data initiatives that fuel strategic success.
• Data product model: Build data products—namely, curated data sets built for specific purposes—to enable those high-value data initiatives. Data products should be discoverable and interoperable so that analysts and engineers can use them as building blocks that can be transformed and combined to unlock new sources of value.
• Ownership and accountability: Organize high-value data into domains, and assign owners who are accountable for making sure that the data remains high quality and accessible. Assign ownership to data products as well, especially those data products built on data from multiple domains. Apply ownership pragmatically; not all domains or data assets need to be governed with the same level of rigor.
• Enterprise alignment: Coordinate governance across teams and business units to unlock initiatives that draw from disparate data sources. Develop enterprise policies on how data will be documented and shared, and create mechanisms to agree on critical data definitions. Establish and agree on decision rights and escalations to resolve any conflicts between teams.
• Investment focused as needs evolve: As tech teams capture, curate, and conform data, emerging use cases should guide where resources go. Data management is continuous, and teams must stay agile, shifting focus to the assets that matter most.
• Governance processes to improve data quality: Embed good governance into workflows, and ensure a regular cadence of meetings among business and technology leaders to review data quality and decide how to continue improving it. Encourage frontline teams to flag quality concerns, and incorporate user feedback into a continuous loop. Business and technology teams work together to develop and manage robust policies, standards, and processes to ensure data quality, security, and accessibility.
• Data architecture: A forward-looking data architecture will support more dynamic data products and harness an expanding range of data types. It will equip agents to act smarter, surface sharper insights, deliver real-time recommendations and predictions, and enable more personalized, multichannel engagement with both employees and customers. Built-in governance and trust, along with the flexibility to scale, will ensure that data flows securely and seamlessly across the enterprise, turning high-quality, accessible data into a true engine of innovation and growth.

One North American utility company showed how strengthening data foundations can improve the ability to extract value from analytics and improve efficiency. The utility had struggled with its fragmented ownership of data, inconsistent quality, and limited documentation. To turn things around, it began by mapping data maturity across 12 dimensions, developing a unified taxonomy and launching pilots to document key data assets and lineage (i.e., where it’s created and how it moves across its life cycle). A first phase closed critical data gaps across more than 20 business-critical use cases. A second phase operationalized governance by embedding stewardship into workflows and scaling lineage and quality tracking across domains. These initiatives delivered real results—specifically, a 20% to 25% gain in efficiency over the first year—helping to recover about $10 million from billing discrepancies and improving accuracy in forecasting customer load.

Successful AI depends on data strategy

A robust data strategy, governance, and operating model are no longer just nice to have; they’re core enablers of AI value realization. To deliver on the promise of AI, every stage of the data life cycle—from capture and processing to AI enablement and end-user engagement—requires intentional design, governance, and active stewardship of curated data products.

This doesn’t happen organically. It calls for deliberate modernization—evolving team capabilities, building organizational readiness, strengthening collaboration between business and tech, and upgrading both architecture and technology. A strong data strategy lays out these goals clearly, and it charts a pragmatic, actionable roadmap to reaching them.