Richard Clough is the EY Global Chief Data Officer, mobilizing the AI Ready Data Strategy and AI Ready Data Governance.
Over half of tech executives expect the majority of AI deployments to be autonomous within two years, according to EY research. But this level of autonomy calls for a radical rethink of how we harness data and power agentic AI.
Data alone can’t deliver the promise of agentic AI, yet many businesses still manage their data estates like static repositories rather than dynamic assets. To fully harness agentic AI’s possibilities, organizations must adopt a mechanism that continuously refines, enriches and redistributes their data assets, turning potential energy into tangible business value.
I believe a key part of this solution is the AI-ready data (AIRD) consent flywheel. In this article, I’ll explain why this model is essential for the agentic future and what it looks like in practice.
Why Flywheels?
In the agentic era, the value and “surface area” of your data is determined by its availability, governance and discoverability. Humans might skip around, overlook assets or get fatigued. AI agents, however, are relentless: if you give them permission, they’ll search every corner of your data estate, every hour of every day.
Most organizations gather data for a single purpose, use it once and move on. In a flywheel model, every dataset is treated as a reusable asset, carefully organized and made accessible for a wide range of applications. A data consent flywheel is a feedback loop where the use of data by AI agents generates new outputs and new data, which is then fed back into the system to fuel continuous improvement. It’s a virtuous cycle, and it’s what can allow agentic AI systems to scale.
For example, as agents process financial transactions, they not only generate insights and recommendations, but create new structured data (analysis results, audit trails and even annotated feedback), which can be leveraged for future decision making, training and compliance.
How To Adopt A Flywheel Approach
When building the AIRD consent flywheel in practice, consider three key approaches:
1. The AI-Ready Data Factory
For data to be discoverable, trusted and easily usable by both humans and machines, it must be AI-ready data. AIRD is rich in business context, properly formatted, meticulously governed and enhanced with the metadata agents need to interpret and use it.
A central part of our strategy at EY is an “AIRD factory.” By aiming to industrialize and streamline the process of creating AIRD at scale, automated tools, rigorous governance frameworks and deep human experience can be brought together to cleanse, enrich and classify data.
It is responsible for curating and permissioning all data, whether it’s generated by agentic platforms, traditional business processes or third-party sources. Think of it as a reliable, just-in-time supply chain for high-quality data, building in quality and compliance from the start.
2. Model Training And Continuous Improvement
The second critical element is model training: infusing AIRD with deep domain knowledge and continuously improving performance.
For us, this means codifying the experience of thousands of EY professionals in areas like tax, consulting and audit, and baking that into our foundation models. The result is AI agents that answer questions quickly while providing advice that is relevant, accurate and grounded in real-world practice.
But model training isn’t a one-and-done exercise. In fact, the distinction between “training” and “operation” is blurring. Continuous improvement, through techniques like prompt engineering, supervised fine-tuning, reinforcement learning (RL) and model distillation, is now the norm. RL, specifically, allows models to iteratively try tasks, receive feedback and self-improve, much like a human apprentice learning on the job.
A crucial tool in this process is synthetic data, which is invaluable for training models while protecting sensitive information, bootstrapping new use cases and scaling solutions across the enterprise. RL methods naturally generate synthetic examples, and model distillation helps build smaller, efficient models based on the outputs of larger ones.
Synthetic data is no silver bullet, but it is a vital enabler for scaling, anonymization and supporting the diverse needs of agentic AI.
3. Agentic Platforms
Everything comes together on an engine that allows AI agents to analyze AIRD and apply it directly to business outcomes. An agentic platform is the cornerstone of any agentic-driven transformation, where models are hosted, agents are developed and intelligence/workflows are orchestrated to scale human capabilities and drive transformative outcomes for clients and internal teams alike.
But agentic platforms aren’t just about technology. They’re about embedding robust data access and permissioning mechanisms, making sure the right users (and agents) get access to the right data, at the right time and for the right purpose. This is where years of investment in governance, risk controls and process excellence can pay off.
At EY, our agentic platform strategy also feeds back into our AIRD factory with the use of data agents. These smart, autonomous helpers work behind the scenes to catalog assets, monitor quality, enforce governance policies, streamline flows and even translate business terms into query-ready formats. They flag risks, suggest improvements and keep the data estate healthy.
The Road Ahead
Through flywheels, data can be used to unleash the true potential of agentic AI. But the full power of the consented data flywheel, where AI agents generate new, valuable data that attracts even greater AI activity, is still emerging. For now, organizations must focus on the foundational work: getting their data cleaned up, enriched and ready for autonomous operation.
The agentic revolution is here, and the difference between organizations that harness this energy and those left behind will likely be their commitment to data readiness today. The sooner that organizations begin their AIRD consent flywheel strategy, the sooner they can turn data into the fuel that drives their business forward.
The views reflected in this article are the views of the author and do not necessarily reflect the views of the global EY organization or its member firms.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
