How A Beauty Company Built An AI Agent — With Rachel Williamson and Josh Siebert

TL;DR

• 1

  Ulta used AI because HR search was broken, not because it wanted an AI project: Rachel Williamson said associates were forced into "Lewis and Clark level expeditions" across fragmented intranet content, often ending in calls, emails, or lunch-break lines outside HR.

• 2

  The first wins were low-risk HR tasks, not emotionally loaded cases: Ulta automated things like address changes, preferred name updates, W-2 status, and PTO-related questions, while keeping humans involved for issues like harassment or associate concerns.

• 3

  Personalization was essential because a handbook could not handle the complexity: Williamson explained that employment rules vary by state, county, and city, so ServiceNow could tailor answers based on where an employee works instead of serving one static policy to everyone.

• 4

  This was framed as capacity relief for a stretched team, not headcount replacement: With more than 65,000 associates and budget limits preventing more hiring, the goal was to "multiply Rachel's team" and let HR spend time on higher-value conversations instead of repetitive admin work.

• 5

  The hardest part was classic enterprise plumbing: Josh Siebert said integrations with third-party systems were tougher than expected, while noting AI should help with upcoming migrations like moving Ulta's legacy intranet content into ServiceNow.

• 6

  Tone and change management mattered as much as the model: Ulta tuned Now Assist so responses felt like Ulta, launched an orange-and-pink branded HR portal, and stressed that successful rollout required clear outcomes, stakeholder buy-in, and strong change management from the start.