Winning with AI: A Conversation with Paychex’s Dave Wilson

Paychex processes nearly 1 million payrolls every month. It’s a complex, high-stakes process, where getting it right matters. As Dave explains, rather than piloting AI on something safe, the company went straight to payroll, the heart of its business. 

That “shoot for the fences” approach helped Paychex become the first HR, benefits, and payroll provider to embed an agentic AI system directly into live payroll processing. The key to their success? A solution that sustained, if not grew, customer trust and a seamless user experience.

Greg asks Dave about how the organization used “shadow mode” to build confidence in the AI’s accuracy. The team learned early on that it couldn’t simply hand the work to the AI agents; human experts still needed to validate the output. Before touching a single live payroll, the AI agent ran in parallel with payroll specialists. 

Dave explains how operating in the shadows helped the team reach its high bar for accuracy. When human intervention was needed, the development team used that insight to refine the AI models and improve performance. And even when specialists needed to step in, the AI had already completed some of the work. That saved them time and freed them to focus on engaging with clients. 

Dave reflects on Paychex’s journey from proof of concept to production and what it takes to build at scale as AI continues to evolve rapidly. He explains how Paychex spent equal time tuning the AI and building the pipeline required to get agents out—all while the technology changed rapidly.

His team’s time in the trenches brought its share of frustration and setbacks. His advice for leaders navigating a similar path: Embrace vulnerability, be open to failure, figure out how to improve, and keep learning through the inevitable roadblocks while remaining hyperfocused on the solution. 

Dave and Greg elaborate on the power of shadow mode and how the period of iteration allowed Paychex to train agents, incrementally improve performance, and ultimately learn some key lessons. 

Greg shares an example of one such lightbulb moment. When the AI agent couldn’t find an employee listed as “Father” at a church client, the solution wasn’t to go into the code and tell it that “Father” was the pastor, not a first name. It was to instruct AI to understand the context of the client and connect the dots. The agent got it right from then on. 

Since 1971, Paychex has built its business on customer trust. Dave says that today, that means providing the expertise and personal engagement customers expect from the company’s payroll specialists alongside the speed of automation. By taking the mundane work off specialists’ plates, agents free them to focus on what customers value most, without customers noticing the difference. 

Embedding agents helped Paychex achieve an 80% reduction in wait time and 30% reduction in effort when a human needed to get involved. But the real win is what specialists did with the time they got back—advising clients and strengthening the client relationships Paychex is known for. 

Today, 80% of Paychex’s customers connect with the company online, while others still send payroll information by email. Dave sees AI as an opportunity to bridge those experiences, bringing greater personalization to every interaction, regardless of the channel.

As he looks ahead, Dave is focused on using AI to serve customers wherever and however they choose to engage. As the company continues to innovate and reinvent the payroll cycle, speed is just one goal. The companies that win will be the ones that understand their differentiation and use AI to amplify it.