ADP just handed Roberto Masiero a new title: chief AI officer. He was already running the company's Innovation Lab, an internal team built to test new products before they reach ADP's customers. Now he also owns the harder job of deciding where AI is allowed to make decisions and where it is not.
ADP is not a small player experimenting on the side. The company processes payroll for a huge share of American workers, and it has said its data covers more than 26 million private-sector employees just in the United States. When a company operates at that scale, a small AI error is not small anymore. It shows up in paychecks, tax filings, and benefits for a lot of real people at once.
That is why Masiero set the bar so high. In his words, the goal is for AI to be right all of the time on the tasks it handles, not mostly right. Being mostly right is fine for a chatbot that recommends a restaurant. It is not fine for software deciding how much money lands in someone's bank account.
This caution is not universal in the payroll industry. Rivals have been racing to launch AI agents that act on their own, making decisions and taking action without a person approving every step first. ADP is deliberately not doing that yet for its core payroll and compliance work. It is choosing to let AI handle narrower, lower-risk jobs, like helping customer service staff find answers faster or helping engineers write code from a plain description, while keeping people in charge of anything that touches compliance.
That split is worth paying attention to. ADP is not avoiding AI. It has been using machine learning for about six years and has AI tools built into products used by small businesses. It is simply refusing to let AI run unsupervised in the one area where mistakes are most expensive: money that belongs to someone else.
There is a broader shift happening here too. Two years ago, almost no large company had a chief AI officer. Recent surveys of large companies now show most of them either have one or are actively hiring for the role, similar to how chief information security officer became a standard job once cyberattacks became a board-level risk. AI is being treated the same way now: not just a tool for one department, but a risk that needs its own owner at the top.
The practical takeaway for any business handling sensitive data, money, or legal obligations is not to copy ADP's exact tools. It is to copy the thinking. Decide which tasks in your business must be correct every time, and keep AI away from those until it can genuinely meet that bar. Everywhere else, let it move fast. The companies that figure out that line first will be the ones customers trust with AI, while the ones that blur it will be the ones explaining a mistake to a regulator.