Skip to content
Fast Company Executive Board

The Fast Company Executive Board is a private, fee-based network of influential leaders, experts, executives, and entrepreneurs who share their insights with our audience.

Why trust is key with synthetic data in the agentic era

Three ways to gain trust in AI while innovating

[Photo: Getty Images]

Profile photo of Joe Depa

Global Chief Innovation Officer · EY

Global tech executives are racing to deploy autonomous agents over the next two years, but in doing so they face a balancing act: How do you leverage data in a way that maximizes trust and confidence in agentic AI?

Organizations hold enormous data wealth they’ll want to leverage while honoring privacy, compliance, and contractual agreements. It’s a challenge we know well at EY. Our data estate is significant—we handle data for over half the Fortune 500 companies, and process over 1 trillion lines of financial data annually. 

We’re also pioneering our own agentic AI program, while keeping human safeguards at the core. By deploying transformative technologies internally first, to test risk controls and human-AI collaboration, we ensure our technologies are ethical and safe before offering them to clients. We refer to this as the Client Zero approach.

Synthetic data is one technology we think will help achieve this. We’re using a secure layer of synthetic data as the foundation to build our first finance agents and we have been happy with the results. Gartner reports that 75% of businesses will generate synthetic customer data by next year, a sharp rise from under 5% in 2023. And there are plenty of other use cases as well. 

Here is how trust and synthetic data can drive agentic AI innovation in three key areas.

Human-centered, AI-ready data

AI must be trained on “AI-ready data” (AIRD)—data that is discoverable, trusted, and easily used by humans and agents. AIRD is high quality, rich in business context, properly formatted, and well-governed. The adage “garbage in, garbage out” still applies to agentic AI, but with higher stakes. With agents, it’s “garbage in, landfill out.”

If AI is trained on biased or incomplete data, the outcomes will reflect those same flaws. Synthetic data addresses this challenge, helping teams overcome data scarcity, privacy hurdles, and bias. It enables models to learn from realistic scenarios without exposing personal data. For example, a traffic management AI may lack edge-case scenarios like multi-car pileups. Synthetic data fills that gap, improving the safety and reliability of systems people depend on.

Synthetic data also allows for customization tailored to human needs and experiences. One inventory management firm trained its AI with synthetic data tailored to its operations, producing better insights for its workforce.

Our digital finance copilot at EY uses synthetic data to power AI supporting analysts with insight generation and variance analysis. In our Client Zero deployment, the copilot cut time spent on manual tasks by 60% while enhancing auditability through a multi-agent setup. The value wasn’t just in the AI’s capabilities, it was in how much more confident our people were using it.

To succeed, human oversight is essential. Experts must guide data generation and validation to ensure it is realistic and ethical. When done right, synthetic data delivers models that are effective and aligned with human values, reinforcing trust at every step.

Secure simulation for people-centered decisions

Synthetic data powers simulations where AI and people can safely learn, test, and improve. As AI takes on more autonomous roles, human-centered simulations become essential.

Consider self-driving cars. Engineers use synthetic environments to model dangerous events like fog or sudden pedestrian crossings. These tests ensure the AI makes the right decisions, earns human trust, and minimizes risk. Waymo Driver AI, for example, has been trained on 20 million miles of real-world driving, but 20 billion miles of simulated driving.

Physical AI systems—like robots trained in simulated environments via platforms such as NVIDIA Isaac Sim—use synthetic data to accelerate learning, slashing real-world training costs while prioritizing human safety and reliability.

Healthcare organizations use synthetic patient records to train on rare conditions or improve medical imaging in underrepresented clinical scenarios, improving preparedness and patient safety while complying with privacy regulations. Business leaders also benefit from simulation. One company used synthetic data to generate over 200,000 go-to-market scenarios, allowing executives to test strategies before investing in new markets.

Simulations reduce risk, strengthen trust, enhance human understanding, and allow iterative learning. They create safe spaces for people and AI to prepare for complex decisions, reinforcing human judgment and confidence.

Confidence is the new currency

No matter how advanced the technology, AI will not scale without trust. Customers, regulators, and employees must believe AI systems are safe, fair, and explainable. Synthetic data helps create this confidence.

It enables organizations to develop and test AI using data that mirrors real-world behavior, without using personal records. This reduces risk, ensures compliance, and reinforces accountability.

Synthetic data also supports bias mitigation. By generating inclusive datasets, teams build AI models that perform across demographics and scenarios. At EY, this is integral to our AI assurance approach.

Transparency is key to earning buy-in. In one case, a client’s legal team approved an AI rollout only after understanding that training was based on synthetic customer portfolios. This clarity accelerated adoption and minimized pushback.

In short, trust isn’t an outcome—it’s an input. And synthetic data helps deliver it by design.

FINAL THOUGHTS

Poor data undermines even the best AI, and protecting data to maintain human trust is non-negotiable.

Synthetic data addresses this. It works as a strategic enabler providing secure, scalable, and smart training grounds for AI to perform under real-world conditions. It helps prioritize human safety and confidence. It allows organizations to innovate responsibly, simulate confidently, and build trust from the start.

To lead in the agentic era, executives must champion synthetic data as a foundation for innovation, while pairing it with governance that keeps humans in the loop. Those who move early will set the standards others follow. Organizations that center trust and human oversight will define the future of intelligent business.

Joe Depa is global chief innovation officer at EY.