Proprietary Data May Become the Most Valuable AI Moat

Artificial intelligence is becoming easier for companies to access. Leading models can be licensed through cloud platforms, open-source systems can be customized, and AI assistants are appearing inside common business software. This expanding availability creates an important question for investors: if competitors can access similar models, where will lasting AI advantages come from?

A new study published in Research Policy¹ suggests that the answer may lie in how effectively companies combine artificial intelligence with proprietary data, institutional knowledge, and human expertise. Its central argument is that AI should no longer be treated as a conventional software tool. It is becoming another learning agent within the organization.

This distinction matters because purchasing AI does not automatically make a company better at learning. Businesses still require relevant data, employees who understand the operating environment, technical teams capable of configuring models, and processes that convert AI-generated insights into action. Companies that assemble these capabilities early may develop advantages that become progressively harder to reproduce.

AI Is Changing How Companies Absorb Knowledge

Management researchers use the term absorptive capacity to describe a company’s ability to recognize valuable external knowledge, understand it, and apply it commercially. Traditionally, this process has been viewed as a human activity supported by research, communication, training, and experience.

The new paper argues that this framework is no longer sufficient. Modern AI systems can scan information, detect patterns, combine data from unrelated domains, and generate possible explanations or solutions. Rather than simply storing information for employees to retrieve, AI can influence what the organization searches for and how it interprets what it finds.

This creates what the researcher calls a dual learning architecture. Human learning remains guided by experience, context, association, and judgment. Machine learning operates through statistical relationships found across data. Each has different strengths and weaknesses, and neither is sufficient alone.

This supports a broader lesson emerging across the economy: information does not create value until an organization can interpret and act upon it. Recent Securities.io coverage of how AI is opening new opportunities for small businesses similarly found that AI capability and entrepreneurial capability are complementary rather than interchangeable.

Data Heritage Could Become a Compounding Corporate Asset

One of the paper’s most useful concepts is data heritage. This refers to the volume, variety, quality, and history of data accumulated by a company.

Traditional databases are largely passive. They preserve information until a person or predefined program retrieves and analyzes it. AI makes accumulated data more active because new information can change how the system identifies patterns and generates future knowledge.

A manufacturer, for example, may possess years of maintenance records, production measurements, defect reports, equipment logs, and technician observations. A competitor might purchase comparable AI software, but it cannot instantly reproduce that operating history. The incumbent’s data can help its models diagnose failures, optimize production, and identify relationships that were not apparent when the records were created.

The process can become self-reinforcing. Better data helps an AI system identify which additional information would be valuable. The company then collects better information, improving subsequent analysis and guiding further collection. The paper describes this deliberate investment in sourcing, curating, integrating, and managing data as data absorption effort.

The resulting advantage is not simply having more data. Poorly structured, irrelevant, or biased data can weaken a system. The moat comes from the combination of:

• Proprietary and relevant information
• Reliable collection and governance processes
• Employees who understand the operating context
• Technical expertise capable of adapting the AI

This is consistent with the broader shift toward hybrid data strategies that combine proprietary corporate information with larger external datasets. Access to a powerful model may become common, but access to the right internal context will remain uneven.

How AI Can Preserve Institutional Knowledge

The study also introduces the idea of knowledge virtualization. Employees often accumulate expertise that is difficult to write down. An experienced technician may detect a mechanical problem from an unusual vibration, while a risk analyst may recognize a suspicious application without being able to reduce the judgment to a simple rule.

Companies traditionally preserve this tacit knowledge through training, mentorship, routines, and shared experience. These methods are valuable but expensive, slow to scale, and vulnerable to employee turnover.

AI creates another possible channel. Models trained on decisions, outcomes, documents, sensor readings, and historical cases can encode portions of experiential knowledge into mathematical representations. This does not mean that a model literally understands everything an experienced employee knows. It means the organization may be able to preserve and redeploy patterns derived from that expertise.

That could turn knowledge previously associated with one employee or team into a reusable corporate asset. New personnel might interact with an AI system that reflects parts of the organization’s historical decision-making, while experienced employees could use its output to identify overlooked evidence.

However, there is a strategic tension. Tacit knowledge can be valuable precisely because competitors cannot easily copy it. Virtualizing that knowledge makes it more scalable internally, but it also raises security, governance, and ownership concerns. If sensitive training data, model weights, or workflows are exposed, the company could inadvertently transfer part of its operational advantage.

Human And Machine Judgment Must Remain Connected

The paper calls the effective coordination of domain knowledge and AI knowledge cognitive duality. This capability requires more than employing data scientists alongside experienced managers. The two groups must be able to understand and challenge one another.

Domain experts need enough AI literacy to recognize when a model is unreliable, operating outside its training environment, or confusing correlation with causation. Technical teams need sufficient domain familiarity to understand which outcomes matter and which constraints cannot be inferred from data alone.

Without that bridge, companies face two opposite risks. Employees may trust AI recommendations too readily, allowing errors or historical biases to spread at scale. Alternatively, they may reject useful findings because the system’s reasoning is unfamiliar or insufficiently explainable.

This helps explain why enterprise AI projects can produce very different results even when organizations use similar underlying models. Performance depends on the surrounding learning system. The World Economic Forum has likewise identified AI-ready data, employee skills, appropriate models, and responsible governance as connected requirements for scaling AI across an enterprise.

What The Study Found About AI-Driven Learning

The research combined a conceptual review with 40 interviews conducted across eight small and medium-sized AI-first firms in Italy. Participants held technical and managerial positions in healthcare, finance, creative content, data management, forecasting, and risk analysis.

The framework is valuable, but its limitations are important. This is a theory-building qualitative study rather than quantitative evidence that these capabilities increase profits or shareholder returns. None of the participating firms had 250 or more employees, and all were based in one country. Further research is needed to test the model across large enterprises, low-technology industries, and different economic environments.

ServiceNow Offers Exposure To Enterprise AI Workflows

For investors seeking exposure to this shift, ServiceNow provides a relevant example. Its platform connects enterprise data, workflows, employees, and increasingly autonomous AI agents. That positioning closely reflects the paper’s argument that the value of AI emerges from its integration with organizational processes rather than from the model alone.

ServiceNow has been expanding its Workflow Data Fabric, AI Control Tower, and autonomous analytics capabilities. The company describes its real-time enterprise data foundation as a way to provide governed operational context to AI systems. Securities.io has also examined the adjacent opportunity through its coverage of UiPath and the agentic automation market.

The investment case is not risk-free. Enterprise software competition is intense, AI features may become bundled commodities, and customers may resist the cost or complexity of broad platform deployments. ServiceNow must also demonstrate that its AI products generate measurable customer value rather than simply increasing product breadth.

Nevertheless, its position illustrates the paper’s central investment insight. As foundation models become more accessible, durable value may migrate toward companies that control workflow context, governed data access, and the systems through which human and machine decisions become operational actions.

The Real AI Divide May Widen Over Time

The emerging divide may not be between businesses that use AI and those that do not. It may be between companies that treat AI as an isolated tool and those that build an organizational learning system around it.

Proprietary data, accumulated expertise, model governance, and cross-functional coordination are all path-dependent. They take time to develop and improve through repeated use. Companies that begin building them early may gain a cumulative advantage, while late adopters cannot necessarily close the gap simply by buying the newest model.

For investors, this changes how AI readiness should be evaluated. Spending on software and computing infrastructure matters, but it reveals little about whether a company can translate AI into defensible value. The more revealing questions concern what proprietary information the business controls, whether it can preserve employee expertise, how closely technical teams work with domain specialists, and whether AI insights can reach real workflows.

The next generation of AI winners may therefore be defined less by who has access to intelligence and more by who knows how to absorb it.

References:

¹ Pedota, M. (2026). Minds and machines: Rethinking absorptive capacity in the age of artificial intelligence. Research Policy, 55, 105600. https://doi.org/10.1016/j.respol.2026.105600