Making data products and agentic AI truly intelligent – insights from McKinsey

Despite enormous investment in AI, especially agentic AI, many organizations are still not seeing its full benefits at a joined-up, enterprise level, according to consultants McKinsey. While 80% of respondents to its latest AI survey said the technology had improved their individual productivity, only 37% said it had improved overall company earnings.  

Poor data foundations are one of the key obstacles to successful, scalable AI. A lack of access to trusted, understandable, relevant and contextualized data means that AI agents are unable to operate effectively. This prevents AI being used to transform end-to-end processes and ways of working, meaning that any benefits are piecemeal, rather than company-wide.

To overcome this issue, McKinsey recommends viewing AI adoption as part of a wider shift, from treating data as an asset to be analyzed, to instead seeing it as a reusable product that AI agents can consume and act on. This blog goes deeper into this concept, the benefits it brings, and the central importance of a data product marketplace to delivering seamless data access to humans and AI.

AI reinforces the importance of data to business performance. Without reliable, AI-ready data, AI models and agents will simply not scale effectively, as many organizations have found. This impacts the success and ROI of AI programs, undermining value.

McKinsey’s latest Global Survey on the state of AI highlights this gap. It found that 80% of individuals said AI had improved their performance, with around half believing it improved their own decision-making. Yet only about four in ten thought that AI has contributed positively to their organizations’ financial performance – a figure that is flat year-on-year, despite the increase in AI rollouts.

The reason for this lack of ROI? Today’s AI and intelligence-driven organization relies on specific data foundations for success. AI-ready data has to be:

• Reliable and high-quality
• Based on a set of common business definitions and semantics
• Built on lineage and traceability, from where data has come from to where it is used
• Subject to real-time quality monitoring
• Delivered through governed access
• Accessible through machine-usable APIs and interfaces 

That’s why even though nearly two-thirds of global organizations have experimented with AI agents, under 10% have increased their use to deliver tangible value. Eight in ten companies surveyed by McKinsey said data limitations were acting as a roadblock to scaling agentic AI.

vercoming these issues requires a shift in how data is handled, and how IT and data teams work with the business. Strategy and operations need to move from seeing data as a siloed cost center to an integral part of the organization and its objectives. Data teams have to collaborate with their business colleagues to co-create relevant, high-quality and reusable data products that both humans and AI can access and act on. According to McKinsey, this is driving six changes in how data leaders focus their programs:

1. Reusable data products matter more than individual datasets

To deliver value, data has to be easily accessible, understandable and trusted by all relevant users, human and AI. As McKinsey stresses, this is best realized by treating data as a product with ownership, quality standards, semantics, interfaces, and observability.

These data products are designed to scale data sharing, with a single product able to support multiple analytics, AI, and operational use cases rather than each team creating its own pipeline. By combining data products further requirements can be met, all while ensuring efficiency and a single, reliable data source.

2. Data products are becoming “intelligence products”

More advanced organizations are evolving beyond first generation data products that provide actionable information and insights. Instead, they are harnessing generative AI to deliver specific recommendations and predictions based on analysis of the data within the product and the user’s context and role. This can even be extended further to empower AI agents to monitor and take decisions based on changing data. For example, instead of providing a dashboard-based data product, a company could provide an agent that continuously analyzes customer behavior and takes approved actions. McKinsey calls this emerging model “intelligence-as-a-product.”

3. The context and semantic layers are becoming strategically important

AI agents lack the innate ability to understand the context around a particular piece of data, unless it is fully encoded through a context layer. By contrast humans, particularly experienced employees, have built up their understanding of the context around specific data and processes. Agents need to have this context added, through context and semantic layers, based on tools such as usage intelligence and business glossaries. These provide agents with a common understanding of entities, relationships and business rules instead of forcing every agent to interpret disconnected tables and documents independently.

4. Data products and agents are converging

Often AI is used effectively within different steps of a workflow, but not end-to-end. This piecemeal approach introduces what McKinsey calls a “coordination tax”, when handoffs between each step in a process require human intervention, decreasing efficiency and slowing operations. Taking a more holistic approach, either through single- or multi-agent workflows, will realize the enterprise-wide value that many companies are not currently seeing. At its heart, these end-to-end AI agent processes depend on interoperable, well-governed data products that bring together data, models, governance and business rules in a machine-readable form.

5. Proprietary data is becoming more valuable

Foundation AI models are now broadly available and accessible to all companies. However, while they are powerful, they are all essentially trained on the same data, and therefore don’t differentiate companies from their rivals. McKinsey’s advice is for companies to focus on surfacing and consuming their own, unique proprietary data, and using this to drive models and agents. McKinsey’s vision of the agentic organization describes a future in which companies continually capture customer behavior, product usage, sensor data, and other proprietary signals, then turn them into differentiated products and AI-enabled processes.

6. Data governance is shifting from static to continuous control

Traditional data governance was reactive and compliance-driven. It focused on cataloging and protecting data, rather than looking to make relevant information securely available to those that could benefit from it. This static picture has to change when delivering data to both humans and agentic AI. Governance has to be more flexible and continuous, monitoring not just which employee or AI agent accessed the data, but looking downstream at what they are then using the data for, and how they are combining it with other information. This more dynamic approach has to incorporate new features such as automated access controls, lineage, auditability, agent telemetry, guardrails, and continuous monitoring.

Building a data-first organization based on McKinsey’s best practices requires secure, seamless human and AI access to reliable, understandable, and trustworthy data products. However, simply creating data products does not drive engagement or usage. What is needed is to make them easily discoverable and consumable through a centralized collaboration space – the data product marketplace. 

Data product marketplaces go beyond traditional data catalogs by providing direct access to relevant data, based on granular permissions to ensure compliance. Lineage enables monitoring of how data has been consumed downstream by all, with users able to give feedback and message data product owners with any queries they might have. Humans benefit from an e-commerce style experience to build trust and help them easily connect to relevant data, while AI agents are guided to the right data products, based on which are consumed by relevant human users. The result is greater data usage, the ability to scale agentic AI, and greater ROI from existing data stack investments.

Embracing agentic AI at scale has the potential to transform business performance. Automation brings efficiency, intelligence delivers agility, and new processes enable innovation. However, AI agents require the right inputs if they are to scale across the business. As McKinsey’s research shows, many organizations are currently stuck in pilot phases, or are seeing limited enterprise-wide benefits from AI. Deploying the combination of data products and data marketplaces to make them discoverable and consumable provides the essential foundation for building the intelligent organization, empowering AI and employees with trusted, reliable and contextualized data in the right format for their changing requirements.