Author: Lincoln Wang | Founder of MindsLeap | Global Partner at Founders Space | Founder of Founders AI Club
Lining up at the HR office door during lunch break just to ask one question — "How many vacation days do I have left?"
This isn't some mismanaged small company — it was the daily reality at Ulta Beauty, a beauty retail giant with 1,300+ stores and 65,000+ employees. Rachel Williamson, Ulta's VP of HR Strategy, described the employee information-finding experience this way: "Our employees were like embarking on a Lewis and Clark expedition, and even when they found content, it wasn't tailored to them."
Lewis and Clark — that famous transcontinental expedition in American history. Using this metaphor to describe a public company's employees finding internal policies is somewhat ironic. But behind the irony is a real business problem: when your workforce spans stores, warehouses, and headquarters, with each person's policy conditions varying by state, county, and city, a thousand-page handbook can't solve the problem. In fact, it IS the problem.
We're Not Building a Chatbot
On April 8, 2026, Ulta's HR service delivery platform went live. Many people's first reaction to "HR + AI" is: oh, feed the employee handbook to a large model and build a chatbot.
Rachel's answer was no. The problem isn't whether there's a dialog box — it's that information itself is fragmented, unsearchable, and means different things to different people. California employment regulations sometimes divide by county, sometimes by city. "If you put this in a paper handbook, it's a one-size-fits-all static document that quickly becomes outdated, extremely costly to maintain."
They chose ServiceNow, a system they call a "highly evolved Swiss Army knife." It's not the content owner — it's the content dispatcher. Raw data might be scattered across third-party systems, but ServiceNow presents it uniformly and gives personalized answers based on "who you are, where you are."
This isn't tech showing off. It's an answer to a simple question: how do you let 65,000 people in different situations get their own answers within the same system?
Fix the Data Foundation Before Talking AI
Ulta's AI deployment sequence is worth noting. They didn't start with HR — they started with IT service management, an internal project codenamed "Shimmer." Employees could submit requests: the store bathroom door handle is broken, the POS system has a故障. These seemingly trivial tickets are the lifeline of store operations.
"With this foundation, we integrated IT services into the HR portal too, so employees could not only ask HR questions but check the status of IT tickets they'd opened," explained Josh Siebert, Ulta's VP of AI & Data Platforms.
More importantly, Josh's mission at Ulta was "building the data foundation." He said: "We made a conscious decision: let ServiceNow drive the enterprise."
The weight of this statement is acknowledging a frequently ignored fact — AI isn't a plug-and-play add-on. You need data soil that can support AI first. Rachel's position itself is a signal: her CHRO wanted HR to shift from "day-to-day operations" to "more future-oriented, strategic thinking," so they created a new position specifically to drive this.
The technology gap was never the biggest obstacle. The organizational gap is.
Which Tasks Should Go to Machines, Which Shouldn't
When deciding automation scope, Ulta did something that looks笨 but is extremely important: line-by-line review of Rachel's team's ticket types, individually judging which could safely go to AI agents and which must retain human intervention.
Address changes, preferred name modifications, W-2 tax status inquiries — these were classified as "low-hanging fruit" and handed to AI. Rachel was frank: "My team didn't want to spend time on these things anyway — they'd rather invest energy in more valuable work requiring human-to-human conversation."
But "I was harassed at work" tickets must never be handled by AI agents. Josh's judgment was clear: "In HR, what's harder to judge is what should go to bots and what needs humans in the loop. That's why we had to bring legal and compliance teams into these conversations."
This distinction seems like common sense, but is frequently ignored in practice. Many enterprise AI projects fail not because technology isn't strong enough, but because boundaries are drawn too模糊 — either forcing unsuitable scenarios onto machines, or missing opportunities to free up human capacity through excessive caution.
The Hardest Part Was Never Choosing Technology
When asked "what was harder than expected," Josh's answer was unsurprising: "Like any large-scale technology launch, it's usually integration. The hardest part is working with third parties, understanding how they send data into ServiceNow so we can do something useful."
Translated into a business judgment: the real cost of every SaaS product you buy isn't the subscription fee — it's the hidden cost of making it work with other systems. Data migration, API integration, format conversion — these seemingly boring tasks often determine whether a project launches orrots.
Interestingly, Josh thinks future AI agents can help with this. They're about to replace the corporate intranet, migrating historical knowledge from legacy systems to ServiceNow — "with AI it should be relatively easy." This is more of a signal: AI's first scaled application may not be customer-facing flashy scenarios, but those dirty, tedious internal tasks nobody wants to touch.
Teaching AI to Speak Your Language
Ulta spent an expected amount of effort on an unexpected detail when tuning their AI agent: tone. They didn't want the AI answering vacation questions to suddenly spit out a platitude completely unrelated to the brand.
"It won't say 'everything is possible,'" Rachel laughed, "but we can make it say: 'You're sparkling, and you still have eight days of vacation.' We could absolutely do that if we wanted."
This detail sounds light, but points to a serious product proposition: when AI agents represent your organization in conversations with employees or customers, their tone IS your brand. You can accept an efficient machine, but you can't accept a machine speaking in someone else's voice.
Ulta did a lot of thinking about "tone" — does the AI agent's response sound like Ulta? This isn't a technical parameter-tuning problem — it's a brand asset management problem.
AI Isn't the Result — It's the Enabler
The most core sentence from the entire conversation comes from Josh's advice to latecomers: "AI isn't a result — it's an enabler."
This statement's value is negating the most common AI deployment posture today — top-down issuance of "we need AI" directives, then asking people below to find scenarios. Ulta's path was completely opposite: first there were problems (employees can't find information, HR teams exhausted by low-value tickets), then technology selection, and only then AI agent deployment.
Rachel's internal communication approach confirmed this. She didn't say "we're launching a new system" — she told HR colleagues: "You'll get a tool tailored for HR that will automate some work, taking those simple inquiries off your shoulders so you can spend time where you want to spend it."
The team's response: "Great, when can we use it?"
Not because people love new technology — but because they know too well how much life they waste daily "pushing paper."
For entrepreneurs considering similar paths, Ulta's experience can be condensed to one principle: start from the outcome, think clearly about what you ultimately want, then treat technology as the bridge to that outcome — not the outcome itself.
When a beauty retailer can use AI agents to stop 65,000 employees from lining up to ask about vacation days, the significance goes beyond HR. It reminds us: what AI truly needs to solve was never "can technology do it" — but "is your organization willing to卸 those low-value burdens from people."
After卸ing them? Return people's time to things that truly need humans.
This principle holds in any industry.
About MindsLeap
MindsLeap is an AI-native organization transformation accelerator.
In deep partnership with Silicon Valley innovation incubator Founders Space, we continuously connect cutting-edge global AI insights, the Silicon Valley tech entrepreneurship ecosystem, and real transformation scenarios for Chinese entrepreneurs.
This article was translated and adapted from the Chinese original with AI assistance.