AI’s rapid evolution—from deterministic to generative to agentic—has transformed how enterprises need to handle data. Unlike deterministic AI, which depends on carefully curated data models, generative AI’s reliance on massive unstructured datasets has already put traditional data frameworks to the test. Now that agentic AI can autonomously drive business outcomes, enterprises are facing a crisis: the data governance models they relied on in the past are no longer sufficient and are, in some cases, completely out of sync with the reality of business needs.
The problems go beyond quality and access. As AI agents take actions that affect business operations, bad data doesn’t just lead to bad insights—it leads to bad decisions. This is particularly crucial given that agentic AI engages with data in fundamentally different ways than traditional business intelligence systems or human analysts.
The 3 Stages of AI Evolution
Whether you are a market leader in a data-driven culture or an organization that has struggled to get your data right for years, the way we manage data is ready to evolve. Figure 1 shows the three stages of evolution and ways organizations should adapt.
Figure 1: AI's Evolution: From Answers to Actions and the Breaking Point of Enterprise Data
AI has evolved from systems that use clean data to solve specific problems, to generative AI that uses massive amounts of data but creates unreliable answers. Now with agentic AI, systems actually take real business actions based on data. This makes data quality critical. Bad data leads directly to costly mistakes and operational problems.
Why Traditional Data Governance Will Not Work
Historically, enterprise data governance has followed a "curate first, use second" approach ensuring data was clean, structured and verified before AI could act on it. Right now, many organizations are instinctively doubling down on past governance models, hoping stricter policies will control the chaos. However, this tactic will not solve the problem. Yesterday’s approach to data is just too rigid for today’s real-time AI needs.
Traditional governance is designed for structured data. AI agents need multi modal, unstructured data which is governed just with as much rigor and with similar focus on quality standards as structured data. However, the very nature of unstructured data means it cannot effectively be governed with predefined, static and human centric processes.
Traditional data architecture frameworks shield source application data from use. However, agentic AI-powered solutions need raw data that comes directly from source application data in real time. Often, these kind of data pipelines bypass governed data architectures altogether and reach right into the raw back-end applications data, operating outside established data engineering and governance practices.
Instead of focusing on cleaning up data after the fact, enterprises must build proactive data validation systems that can assess data reliability before AI agents act on it. To be successful with agentic AI, enterprises need new governance models that prioritize real-time accuracy, adaptability and dynamic validation methods.
Fixing Data Also Means Challenging How Data Projects Are Delivered
Currently, the majority of data projects follow one of two delivery models:
- Boil-the-ocean approach: Lofty goals of “getting our data right” often lead to long multi-year projects. Why these usually fail is because governance becomes the point of contention that ultimately diminishes the value of the project. They also leave little room for rapid innovation and are incapable of moving at the speed of business.
- By-pass-the-mess approach: This approach completely ignores existing structures and creates completely new pipelines designed around a singular need. This approach is often adopted by frustrated leaders who operate in silos and are the culprit behind the “shadow IT” syndrome, which creates organizational problems and strategic misalignment.
Inevitably, for companies that struggle with data, there is an ongoing shift from one approach to the other, often tied to external pressures. Today, CEOs everywhere are asking why they need to wait for value when a proof of concept has shown them how an agent can take “messy” data and work magic with it in months. Organizations need a new approach to sourcing data services that reflect the reality of needing to move and evolve at the speed of business.
Three Solutions to the Data Crisis
The following three solutions …
- Manage data dynamically or fall behind: To prevent AI from breaking enterprise data (and the decisions that depend on it), companies must shift their data approach from static control to dynamic validation. Frankly, for most organizations that have struggled with extracting value from data for years, this conclusion is nothing new. However, the emergence of agentic AI has also brought new tools for data management, such as real-time AI-driven data quality, governance and performance management. These are especially welcome for managing multimodal data where manual governance and data quality processes are not feasible.
- Prepare for data sitting inside applications to become your key data products of tomorrow. Most data engineering teams today are focused on managing data inside well-defined, structured platforms, often supported by what is referred to as Medallion architecture, where raw data is imported and staged in a series of environments that eventually lead to a high-quality semantic layer. However, for agents to truly drive change, AI needs to interact directly with back-end application data. This means we are looking at the future where “writing back” data updates or corrections to applications and enforcing data governance at the source becomes essential. Preparing for this eventuality today means that, as agentic AI matures, dynamic governance of application data becomes a competitive advantage tomorrow.
- Redefine how your provider does data delivery. If your data provider is telling you that the way to get value from data is to wait for a multi-year project to succeed, it’s time to rethink your strategy and perhaps your partnership. Instead, consider shifting your approach to a model of two speeds that operate in parallel:
- Optimization mode for stable, predictable data needs that require highly governed and stable processes. It’s important to note that not everything can be optimized, so look critically at your core data as a valuable and expensive resource. Identify data that has proven, defined ROI and draw a line in the sand. Set realistic goals and ensure there is a clear visible link to business value.
- Innovation mode for exploratory, business driven, adaptive data needs that are not bound by existing architecture or optimization constraints. These bursts of energy will be expensive and sometimes will break things and will rely on data that has yet to prove its value. However, as long as innovation remains strategic, it can be optimized down the road
Figure 2: Fixing the Data Crisis: A Framework for the AI Era
To succeed, your AI strategy must support both modes independently and create a clear path for successful innovations to become part of the core operating model.
AI Is Only as Good as the Data It Uses
When data was first recognized as a source of competitive advantage, early adopters moved swiftly, reaping rewards that left others scrambling to catch up. Today, AI is driving a huge upheaval, but it also presents an opportunity for those that have struggled to win with data to finally make a dent in their challenges, survive and become leaders.
Transform your data from a liability into your greatest AI asset. ISG helps organizations tackle the challenges in logical steps with an honest third-party perspective on both your internal challenges and your partner implementation approach. Contact us to find out how we can get started.
About the authors
Loren Absher
