AI is generating both excitement and concern in the workplace, and especially in HR. Tools like Wisq’s AI HR generalist, Harper, promise to automate tasks and slash workloads by more than 80%, yet many organizations find their AI experiments deliver only incremental results instead of the transformative change they hoped for, says Clayton Holz, the head of product at the agentic HR platform Wisq.

AI can transform HR departments if deployed pragmatically, Holz shared during a From Day One webinar about “The Agentic HR Operating Model: Moving Beyond Chatbots to AI-First HR.” Success requires proactively navigating around fundamental limitations, redesigning processes, and rethinking organizational structure, he says. 

“You are going to have to do work,” Holz said. “But I can clearly see a world in which, in the HR domain, AI is doing a lot of the tier zero, tier one, maybe some of the tier two work, and then augmenting your work further up the pyramid.”

The Reality Check on AI Agents

Transforming HR departments with AI starts with a clear understanding of the technology’s current capabilities and limitations. Holz outlines several fundamental constraints HR leaders must work around. 

First, AI models are probabilistic, not deterministic. “A model with no supporting safety algorithm is going to be consistent anywhere from 60 to 90% of the time,” he said, adding that out-of-the-box AI tools like Copilot often perform worse on HR-specific tasks. 

Second, artificial intelligence struggles with complex, multi-turn instructions, sometimes leading to “hallucinations.” Models are trained to be sycophantic, which conflicts with HR’s need to be consistently fair and empathetic. “AI is more than willing to go ahead and make a decision on behalf of your organization that might not be present in the source material,” Holz said. 

Third, AI doesn’t magically drive behavior change. For example, AI can’t do much if employees aren’t motivated to create individual development plans. “AI is not suddenly going to change that motivational gap and change people's behavior,” Holz said. “It might make the experience of doing it different and better, but it's not necessarily going to change their behavior.”

Holz advises that the best use cases for artificial intelligence are scenarios where “people just keep doing what they were doing before, and AI is there so they don’t have to change any behavior at all.”

Building the Bridge With IT

A major hurdle many HR teams face is getting their IT departments on board with the decision to give AI a larger role. Holz says HR teams should be proactive and data-driven when communicating with IT departments. 

“Vague asks are going to be difficult for them to prioritize,” he said. Holz advises HR leaders to articulate their needs to IT teams using three key dimensions: care (employee experience), compliance (risk reduction), and cost (time savings).

He recommends building bridges by helping IT teams to understand the unique nuances of HR work, such as the rules governing a leave of absence. “The best examples that I’ve seen, honestly, are when someone in IT has recently gone on a leave of absence, because then they have empathy for some of the things that you’re dealing with,” Holz said.

Redesigning HR Processes for AI

The core of Holz’s advice for HR leaders centers on the methodologies used to redesign processes for the implementation of AI. He suggests a hands-on workshop approach, starting with a subject-matter expert interview to map out multi-step processes like handling a request for promotion in detail, including the back-and-forth and waiting periods. 

Once that’s out of the way, the next step is to codify each step of the process. This systematic breakdown makes the workflow shape visible. “This is a great opportunity for redesign,” Holz said. “We’re looking for places where we can remove, combine, reorder, and standardize steps.”

Finally, each step of each process should be evaluated and placed on a two-by-two grid. One axis measures the level of risk involved if the step is incorrectly done, while the other measures the amount of human judgement required. Doing so creates a clear opportunity map for applying AI to processes:

• Low-risk, low-judgment tasks: Examples include tasks like sending a reminder email or retrieving a standard policy document. These are prime candidates for full automation.
• High-risk, high-judgment tasks: These are important tasks like granting final approval on a sensitive employee relations case. Such tasks should be primarily handled by human experts, while AI serves as an assistant that helps to curate information or generate drafts. 
• Tasks in the middle: These include tasks such as the initial assessment of a promotion request against set criteria. Most of these tasks can be handled by AI, but a human review step should be built in for quality assurance. For example, in the case of a promotion request, the approval routine might be automated, while a human communicates the final decision to the employee after evaluating the AI’s recommendation. 

Preparing Policies for an AI Teammate

Holz says company policies must be AI-ready to operate the technology effectively. Ambiguous policies that are confusing to humans will be even more problematic for artificial intelligence. He highlights some of the most common issues organizations face, including outdated handbooks, over-reliance on jargon, and vast amounts of unwritten “tribal knowledge” governing discretionary decisions. 

Holz recommends archiving old policy documents, using plain language, and running policies by focus groups consisting of new employees to test for clarity. “If they don’t get it, it’s likely that AI is also not going to get it consistently,” he said. He also recommends codifying the unwritten rules that govern discretionary decisions, like what counts as a “close family member” for bereavement leave. This codification is essential to offload repetitive work to an AI agent. 

One of the most profound shifts in attitudes Holz proposed regarding AI is to view it as a team member, instead of a tool. “They’re going to need a manager, they’re going to need onboarding, they’re going to need supervision, they’re going to need performance feedback,” he said.

Holz predicts the slow, consensus-driven policy management model will hinder the effectiveness of AI systems adopted and sees forward-thinking companies shifting toward a hub-and-spoke model with clear, centralized policy owners. “This is going to be a big [shift], allowing you to move much more quickly and get more out of AI going forward,” he added.

The Future of HR Service Delivery

Holz envisions a not-too-distant future where AI handles the majority of HR service delivery, freeing humans up for tasks that require more human skills. “I think all of [those transactional requests] could be done in part or entirely by AI,” he said. 

The shift toward AI-driven HR may also encourage organizations to standardize policies that were once open to broad interpretation. “I could see policies becoming more black and white, candidly, or black and white for large shares of the population, so that decisions can be made in a more programmatic and consistent way,” he said.

Holz’s message to HR leaders is that AI has the potential to transform processes, but that requires a proactive, process-oriented, and collaborative approach. The teammate of the future is waiting to be onboarded. 

Editor’s note: From Day One thanks our partner, Wisq, for sponsoring this webinar. 

Ade Akin covers artificial intelligence, workplace wellness, HR trends, and digital health solutions.

(Photo by CL Stock/Shutterstock)