Ulta Uses AI to Personalize HR Support for 65,000 Workers

This was a ServiceNow-sponsored interview about Ulta Beauty’s deployment of ServiceNow’s HR service delivery platform and Now Assist. Ulta’s AI deployment in HR was framed by Josh Siebert as a response to a practical operating problem: employees could not reliably find the policies, guides, and internal information they needed to work at the company. The problem was not that Ulta lacked content. It had too much of it, spread across fragmented internal systems, with search that Rachel Williamson called “painfully ineffective.”

Ulta has more than 65,000 associates across stores, distribution centers, and corporate teams. That workforce creates a different design challenge than a headquarters-only deployment. Many associates are not sitting at a desk with time to browse an intranet. They may be on a store floor, on break, in a distribution center, or trying to resolve a personnel question quickly enough to get back to work.

Williamson said employees previously had to go on “Lewis and Clark level expeditions” to find internal content. The content itself covered the ordinary machinery of employment: policies, quick reference guides, how-to material for performance management, tools, resources, and “really everything about how to be an associate at the company.” When employees could not find what they needed, they called Ulta’s Associate Care and Support Team, asked a manager who often did not have the answer, or, where HR was physically available, waited in line during lunch or breaks.

That last example captured the cost of the old system. A question about PTO, an address change, a W-2, or a policy could consume both the associate’s time and the HR team’s attention. Williamson described lunch and break time as “sacred” for store employees. The internal information architecture was turning routine questions into delays.

The company went live on April 8 with an HR service delivery platform built on ServiceNow, including Now Assist. Siebert described the aim as improving employee experience while effectively “multiplying” Williamson’s team — scaling support without adding headcount.

The system was not, in Williamson’s view, merely an uploaded employee handbook with a chatbot attached. That distinction mattered. Ulta’s issue was not only retrieval; it was personalization. Employment rules and HR policies are not uniform across a national workforce. Williamson pointed to California employment law as an example, where rules can vary by county or city. A static handbook would either be too generic or too large to be useful.

“Because that book would be 10,000 pages,” she said when asked why Ulta could not simply distribute a rulebook.

The system, as described by Williamson, can return answers based on what applies to an associate in a particular city and state. In the broader design discussion, Siebert also described the need to serve different employee personas — corporate, stores, and distribution centers — by meeting them where they work and personalizing content so the journey is simple. If an associate asks about PTO, the answer should reflect the rules relevant to that individual, rather than a generic policy that still requires interpretation.

Ulta’s HR deployment sits inside what Josh Siebert described as a broader use of ServiceNow across the enterprise. He said the company chose ServiceNow in part because, from his prior experience, it integrates well with third-party systems. He called it a “highly evolved Swiss Army knife” and described it as a “system of action”: the original record or artifact might live somewhere else, but ServiceNow can surface it where employees can use it.

The first HR-facing product was an Ulta-branded HR portal. Rachel Williamson emphasized that the interface had to feel like Ulta — “pretty,” with orange, pink, and “sparkles” — but the more consequential change was functional. The portal includes both a Now Assist chatbot and a Now Assist-powered search field. Williamson said the search improvement was one of the biggest crowd pleasers for associates.

The HR launch followed an earlier ServiceNow deployment in IT service management, which Ulta calls Shimmer. Siebert said HR did not come first; IT did. The company then surfaced IT service capabilities within the HR experience so employees could find HR answers and also see open IT cases in another place.

The examples of “IT” work were deliberately mundane: a broken bathroom handle in a store, a technology glitch, or another operational issue that needs to be reported. Alex Kantrowitz pushed back on the instinct to minimize such tickets. A register, a store system, or a facilities issue can affect commerce directly. If enough systems are down, the store does not work or customers do not return.

That point shaped the larger discussion about automation. The question was not whether a bot should replace a human in every case. It was where automation lowers friction without introducing unacceptable risk.

Siebert’s answer was conditional: “It depends on the use case.”

In HR, Williamson said, that distinction becomes sharper than in many IT support workflows. Password resets are one category of request. Workplace harassment is another. The first can plausibly be automated. The second should not be handled as a routine chatbot exchange.

Ulta’s HR team therefore went through use cases with legal and compliance input. The question was not just what the bot could answer, but what the company should do when a serious or sensitive request appeared. Williamson described the work as deciding what the company was comfortable “completely offloading to a bot,” and what required a human in the loop.

Routine, repetitive requests were the first candidates: updating an address, updating a preferred name, checking W-2 status, and similar low-complexity contacts. These were not the interactions Williamson wanted her HR team spending its time on. The goal was to free the team for “more value add” contacts, especially associate concerns that need human judgment and conversation.

The HR automation program was presented by Rachel Williamson less as a labor-reduction story than as a response to a capacity constraint. She said her team had reached a pinch point. Ulta had grown, the associate base was large, and HR support processes could not absorb more work by simply adding people.

When asked why not “throw people at the problem,” Williamson gave the blunt answer: budgets would not allow it. Finance had said no more headcount.

That constraint changed the internal case for automation. Williamson said she was hearing in skip-level meetings that her team was at a breaking point. HR professionals were on calls while watching email queues grow. They were trying to give full attention to sensitive associate concerns while seeing the backlog accumulate. That was bad for the HR team and for the employees seeking help.

The message to the team, Williamson said, was that the new tool would be purpose-built for HR, automate some of the easy contacts, and offload routine work so people could spend time where they wanted to spend it. Her characterization of the team’s response was not fear but impatience: “Yes. Please. When?”

She acknowledged that some people were still adapting to the new system and learning its workflows. But overall, she said, sentiment was positive because the team understood the work it would now be able to prioritize: associate experience, associate concerns, and matters where human support mattered.

Williamson said she had seen automation used this way before at Amazon. When she joined Ulta and took over the operations part of the HR team, she saw manual work bogging people down and started looking for IT partners who could help automate processes.

That is where Williamson and Josh Siebert’s work converged. Siebert said he was looking for another business partner after the IT rollout and saw HR as a willing function looking for a “step jump.” Williamson brought a technology-forward operating mindset; Siebert brought prior ServiceNow experience and the enterprise platform remit.

Siebert referred to a four-phase HR approach with ServiceNow, but did not enumerate the phases. The work he described included the HR service delivery launch already live, a planned corporate intranet migration, and a later move toward employee lifecycle workflows such as onboarding, promotions, and role changes.

When asked what was harder than expected, Josh Siebert did not point to model capability. He gave the answer he said he has seen in every large-scale technology rollout: integrations.

The difficult work was getting third parties to land data in ServiceNow in a way Ulta could use. A portal that answers questions depends on underlying content, records, and workflows being available, current, and structured enough to support action. In Siebert’s framing, the platform’s value comes from connecting systems so employees can do something, not merely retrieve text.

He also saw AI as potentially useful for the integration and migration work itself. Ulta is preparing to replace its corporate intranet and move historical knowledge from a legacy solution into ServiceNow. Siebert said moving that content “should be relatively easy with AI,” especially compared with past migration work. He also said AI could support content reviews against business rules.

The less visible work was making AI answers usable inside a company: reviewing tone, consistency, and fit. Siebert said Ulta spent time thinking about how Now Assist answered questions and whether it sounded like Ulta. The team did some tuning, though he joked that the system would not necessarily say “the possibilities are beautiful” in every response.

The humor pointed to a real implementation question. In employee-facing AI systems, the answer is not only whether the bot retrieves the right document. It is whether it responds in a way that fits the company’s voice and handles ambiguity appropriately.

Kantrowitz suggested AI could also help detect inconsistencies in messaging or policies by running across a corpus of internal material. Siebert agreed. The same technology used to answer employee questions could, in principle, identify places where the underlying content conflicts or drifts from business rules.

Expense handling came up as another possible area for automation. Kantrowitz described the familiar problem of an expense submission sitting unresolved because it does not fit a category. Siebert said large language models can view images and understand what to do with them; the remaining requirement is building the workflow to take the next action.

That distinction recurred throughout the discussion. The model may be capable of classifying a receipt, reading a document, or drafting an answer. But the enterprise value, as Siebert described it, depends on the workflow around it: what system receives the data, what action follows, what rules apply, and where a person must intervene.

Ulta’s next major platform move is the corporate intranet migration, involving HR and store operations. After that, Josh Siebert said the company plans to move into employee lifecycle workflows. The examples he gave were onboarding new employees, supporting promotions, and helping employees who change roles. The ambition is to have ServiceNow and AI-backed agents support those transitions.

Onboarding was discussed as a particularly obvious pain point because both Siebert and Rachel Williamson joined Ulta within the last two to three years. The experience was still fresh enough for Siebert to describe “the pain of the onboarding process.” Kantrowitz suggested that automation could condense some of the administrative work that often dominates the first few days of a job, when a new employee is filling out forms before getting deeper context on the role.

Williamson’s role itself was created as part of a top-down HR transformation. She said Ulta’s chief human resources officer at the time wanted the HR function to become more future-focused and strategic. Historically, she said, HR had been operating day to day. Her role was designed to help the team think about where it was going, how to optimize, and what initiatives would support that direction.

Siebert said he was hired to create the data foundation that would allow for systems like ServiceNow and AI in the future. Ulta made a conscious decision that it wanted ServiceNow to power the enterprise. In his telling, the timing worked because Williamson brought a technology-forward HR agenda and he brought prior ServiceNow experience.

The final advice from both was pointedly anti-hype. Siebert said companies should think about outcomes upfront. In a new technology journey, he starts with an assessment and works with the business partner to understand what they want. As an IT leader, he described himself as the enabler of the business outcome.

Williamson’s advice was similar: start with the end in mind, identify key stakeholders, secure buy-in early, and bring them through the journey. Siebert added that strong change management buy-in matters.

Kantrowitz summarized the implication: the wrong way is to issue a top-down edict that the company needs “AI in action” and tell people to figure it out.

Williamson put the principle more simply: “You need to have a real problem to solve.”

Siebert’s version was the clearest statement of the operating philosophy behind Ulta’s deployment.