Yes, Data is the New Oil. But only if it’s Liquid
Organizations have never had more data. Yet most AI initiatives fail. An MIT study points to one of the key reasons: the information exists, but it isn’t liquid.
By David Salzer, CEO and Founder of Cambium, a Matrix Group software company developing AI systems, web solutions, and applications for leading organizations across industries.
More Data Does Not Necessarily Mean Better Data
Over the past decade, organizations have invested heavily in accumulating data under the assumption that the more information they had, the greater its value. The AI era is fundamentally changing that assumption. Models do not simply need access to vast amounts of information; they need information that is relevant, current, and accessible. The question is no longer how much data an organization has, but how relevant, high-quality, and accessible that data is across systems and processes.
Consider a simple example. A manufacturing company wants to plan next year’s budget. It has budget data dating back to 1978. Everything is accessible, everything is available, and all of it can be fed into an AI system to provide endless context. In practice, however, market conditions from the late 1970s are of little relevance to the business reality of the coming year. In this case, data from the last three years, combined with current market analysis, is far more valuable than forty years of historical data.
In the age of AI, the value of information is no longer measured solely by its volume. It is measured by the ability to use it at the right time, in the right context, and in a way that generates meaningful insights.
What is Data Liquidity?
Data Liquidity describes how easily an organization can access its data, combine it from multiple sources, move it across systems, and reuse it quickly and flexibly. Similar to financial liquidity, where an asset is considered liquid if it can be easily converted into cash, data is considered liquid when it can be activated in real time, connected to different contexts, and transformed into value without technological or organizational barriers. The goal is not merely to possess data, but to ensure that it is available, connected, current, and ready to support analytics, decision-making, and AI systems.
Organizations with high levels of Data Liquidity achieve stronger business performance, including improvements of approximately 14% in metrics such as time-to-market and customer experience (MIT research)
When it comes to Agentic AI, this challenge becomes even more critical. The data exists and sits in an expensive, sophisticated data warehouse. But if it takes two weeks to extract, cleanse, and prepare that data for use, it is not liquid. In such a situation, autonomous agents begin to hallucinate, make mistakes, and generate decisions based on incomplete or irrelevant information.
Recent studies from 2024 and 2025 show that an organization’s ability to generate value from AI depends directly on its level of Data Liquidity, meaning its ability to reuse and integrate data across systems. A 2025 study by MIT Center for Information Systems Research found that organizations with higher Data Liquidity are more successful in deploying AI initiatives and achieve better business outcomes, including roughly a 14% improvement in metrics such as time-to-market and customer experience. At the same time, MIT’s 2025 report on GenAI in organizations found that approximately 95% of AI initiatives fail to generate measurable business value. One of the primary reasons is the disconnect between AI systems and the underlying data infrastructure, as well as the inability to move and integrate data effectively. In other words, the gap between organizations that succeed with AI and those that do not is driven less by the models themselves and more by the ability to make data liquid, accessible, and connected so it can continuously and reliably power AI systems.
Three Key Levers for Increasing Data Liquidity
The first is architecture. When the CRM identifies a customer with one identifier, the accounting system uses another, and application logs rely on a third, the data exists but does not truly connect. The solution begins with a single decision: a unified identifier for every entity across all organizational systems. Easy to say, much harder to implement in a large, mature enterprise.
The second is data preparation. Cleansing, tagging, and rich metadata should be in place before AI teams begin their work. Data scientists often spend a substantial portion of their time on data wrangling simply because the information was never prepared properly for use.
The third is access management. Organizations need to move from a “submit a request and wait for approval” model to Policy-Based Access, where authorized users can obtain access automatically. In many cases, bureaucratic bottlenecks are just as limiting as technological ones.
Data Liquidity Creates Business Value. But It Also Increases Risk.
High levels of data liquidity have a downside that cannot be ignored. The freer information flows, the larger the attack surface becomes. In an environment where data moves easily, an attacker who gains access can move laterally much faster and extract information before the breach is detected.
There is also a regulatory challenge. Regulations such as the General Data Protection Regulation (GDPR) require data minimization and the right to be forgotten, both of which become significantly harder to enforce when information is replicated across dozens of systems.
Alongside this comes the classic Garbage In, Garbage Out problem at scale. When incorrect data flows automatically without human oversight, errors spread immediately across the models and systems that depend on it. Financial markets have already experienced flash crashes triggered by algorithms reacting simultaneously to the same faulty signal.
The answer, therefore, is not maximum liquidity but managed liquidity. Data should move quickly where it creates immediate business value, while guardrails and deliberate delays are maintained in sensitive areas. Approaches such as Data Mesh give ownership of data to the teams that produce it and treat data as a product with internal customers. Data Fabric, by contrast, creates a virtualization layer that connects distributed information without physically moving it.
Caterpillar increased service revenues by billions of dollars through predictive platforms built on operational data liquidity. Fidelity Investments created an internal data marketplace that transformed the way information flows between business units.
Ultimately, the question architecture leaders should ask themselves is simple: How long does it take today to extract a critical piece of information and prepare it for use? If the answer is measured in weeks, your data may be available. But it is not yet liquid.
