ADP’s AI leader balances innovation with control

ADP has developed an approach to artificial intelligence that prioritizes compliance while still encouraging new ideas. The company’s chief AI officer, Roberto Masiero, describes this as balancing trust and speed, ensuring AI performs only tasks where accuracy is guaranteed.

“We cannot accept being mostly right,” Masiero stated. “Payroll compliance demands absolute precision, guiding AI toward highly predictable outcomes that avoid errors or deviations.”

An innovation lab with guardrails

Masiero, who was recently named chief AI officer, previously oversaw ADP’s Innovation Lab. This internal team operates like a startup, testing new products and services under strict success criteria—such as revenue potential and client reach—before moving them into full development.

Two major products, ADP Mobile and the ADP Marketplace, began in the lab. The platform now serves over 20 million users, while the marketplace facilitates HR data sharing and API integration. The lab also works with ADP Ventures, the company’s investment arm, to identify startups and form partnerships. These collaborations help shape secure AI practices within the organization.

Academic ties further strengthen ADP’s strategy. The lab partners with the ADP Research Institute, which maintains connections to Stanford and other universities. While these relationships influence AI development, Masiero stressed that the technology remains tightly controlled.

For instance, ADP integrated conversational AI into its small-business payroll platform but imposed strict limits on generative AI to prevent compliance risks. “A key part of our work is setting guardrails to keep the model grounded,” he explained. “Accuracy is non-negotiable.”

AI as a productivity tool, not a replacement

The lab’s efforts extend beyond external products. ADP uses AI internally to improve efficiency, particularly for service teams. The technology helps employees learn from shared experiences and content, leading to better decisions. IT staff also benefit, with engineers using AI to speed up development.

“Some don’t write a single line of code,” Masiero noted. “They describe what they need, and the AI handles the programming or syntax for the feature.”

Masiero’s strategy reflects a common challenge in enterprise AI: leveraging its potential without losing control. Many regulated industries face similar pressures, where even small mistakes can have serious consequences. ADP’s method—restricting AI to specific, repeatable tasks—aligns with practices in healthcare and finance, where predictable results are essential.

The lab’s investment in AI spans six years, though early efforts focused on machine learning. Initial uses included intent recognition and translation. The rise of generative AI introduced new hurdles, particularly around reliability and instruction-following.

ADP’s measured approach isn’t due to limited resources. The company has the financial and technical capacity to deploy AI more aggressively. However, in an industry where payroll errors can lead to legal and financial penalties, precision remains the priority.

“Initially, we relied on machine learning and trained models for specific tasks,” Masiero recalled. “Large language models changed the setting, bringing new complexities.”