Ulta Credits AI for 95% of Sales. That Number's Been 95% Since 2023.
Ulta Beauty told Adweek on May 5 that AI personalization is supercharging the loyalty program that drives 95% of its sales. PYMNTS reported the same 95% loyalty share in 2023, two full years before Ulta rolled out the Virtual Beauty Advisor AI inside its mobile app. The number AI is actually moving is repurchase rate per recommendation, not the topline loyalty share that retailers love to quote on stage.
The 95% number predates the AI strategy by three years
The headline metric in the Adweek interview with CMO Kelly Mahoney is that 46.7 million loyalty members generate 95% of Ulta's sales. The implication, as the piece is structured, is that AI personalization is what got the company there. It isn't.
That 95% has been Ulta's loyalty share since at least 2023, when PYMNTS published the same figure with 43 million members and zero mention of generative AI. Motley Fool reported the same 95% in mid-2024. It is a structural feature of the program, not a recent lift. The loyalty members were already buying.
What changed between 2023 and now is the member count, which crept from roughly 38 million to 46.7 million. That is real growth. It is also growth that almost any well-run loyalty program with a decade of compounding can produce, and most of it likely came from store visits and the rebrand from Ultamate Rewards to Ulta Beauty Rewards in 2024, not from a chatbot that launched late.
I think the press release framing is doing the work here. Ulta has invested heavily in AI, the investments are real, and the company genuinely wants to talk about them. So the metric that's most flattering gets pushed to the front. The 95% is a fine number. It just doesn't belong in the same sentence as a 2025 AI rollout.
What the AI is actually changing
Once you set the topline aside, the more interesting work is happening underneath. Ulta has been stacking AI tools for years. Glossy's reporting covers the lineup: GLAMlab for virtual try-on, the Beauty Advisor Chat that launched in 2023, Iliad.ai for asset generation, and a portfolio play through Prisma Ventures into companies like Luum (lash robots) and MyAvana (hair analysis).
The Virtual Beauty Advisor inside the mobile app is the newest piece. It uses what Ulta describes as 45 million points of loyalty data to predict skincare needs before the customer searches. Agustina Sartori, the company's senior director of innovation, told Glossy the goal is to complement the in-store journey, not replace it.
None of this is small. GLAMlab alone has logged over 11.5 million visits with 82 million shades trialed since 2016. Those are real engagement numbers. But they are member experience numbers, which is a different category from the topline 95%.
Here is the actual lift, as best I can read it. AI personalization at Ulta seems to be improving frequency and basket size inside the loyalty cohort. The rewards members were already loyal. What's getting better is what each one buys per session and how quickly they come back. That's the margin story. It just isn't the story Ulta is telling, probably because "AI improves repurchase rate per email by some percentage we won't share" is a worse press headline than "AI drives 95% of sales."
The PYMNTS repurchase number is the one to copy
A separate PYMNTS piece in 2025 framed the same 95% differently: as a customer repurchase rate, not a sales share. That framing matters more for anyone trying to copy what Ulta is doing.
If you run a retention program at any retailer, the question to bring to your next QBR is not "what share of revenue comes from loyalty members" because, frankly, that number is table stakes once a program is mature. The number that tells you whether AI is doing anything is the conditional one: among customers who received an AI-served recommendation in the last 30 days, what's the repurchase rate compared to a holdout that received only the standard merchandised email?
That's an A/B you can actually run this quarter. Pick a category, segment the loyalty file into two cells, send the AI-personalized version to one and the rules-based version to the other, and measure repurchase at 14 and 30 days. From what I've seen at smaller retail ops, the lift on a well-built personalization layer is usually somewhere between 8% and 18% on repeat-purchase rate within a quarter, with the bigger numbers concentrated in replenishment categories like skincare and haircare. Variable rewards, sure, but a real number you can defend in a budget meeting.
What you absolutely should not do is copy the 95% framing. If a CMO walks into a board meeting saying "AI drives 95% of our sales because 95% of our sales already come from loyalty members and we put a chatbot on the app," the smartest person in the room will catch the attribution gap inside thirty seconds. And then the AI budget gets harder to defend, not easier.
The vendor signal nobody's calling out
One detail buried in PYMNTS coverage: Ulta's AI partner of record on the loyalty data side is SAS. Not OpenAI, not a flashy new agentic startup, not a hyperscaler reference build. SAS, which has been doing predictive analytics for decades and which most marketers think of as the platform their analyst team grumbles about.
That choice is a signal worth sitting with. The retailer with the data asset Mahoney called the kind of thing "marketers would die to have" did not pick the new shiny thing. They picked the platform with the longest track record on customer-level prediction, and they probably did so because their data team can actually operate it without a six-month onboarding from a consultancy.
For most marketing teams in 2026, the lesson isn't "buy SAS." It's "your AI personalization vendor doesn't have to be the one in the headlines." The model behind the recommendation matters less than the data engineering that gets the right signal to it. A loyalty file with ten years of clean transaction history will outperform a sloppy one with GPT-5 attached, every time. We've made a similar argument about DTC retention: the brands that fail aren't the ones with the wrong tooling, they're the ones with hollow first-party data underneath the tooling.
The number that should be on every retention dashboard
If you take one thing from the Ulta cycle, make it this: stop reporting loyalty share of revenue as a success metric for AI. It rewards the loyalty program, not the model. Replace it with two cleaner numbers.
The first is incremental repurchase rate from AI-served recommendations against a no-AI control. The second is contribution margin per loyalty member, tracked monthly, before and after each major personalization release. The first tells you the model is working. The second tells you whether the model is producing margin or just shifting purchases the customer would have made anyway.
Ulta probably has both numbers internally. They're just not the ones the company chose to put in front of a reporter, which is a tell on how much margin the lift is actually producing. If the AI were generating a clean 20-point lift on repurchase, that would be the headline. The fact that 95% is the headline is, in a quiet way, the most honest thing about the press cycle.
I'd guess Ulta's real AI lift on personalization is in the high single digits on repurchase, maybe low double digits in skincare specifically. Useful, defensible, worth funding. Just not "supercharging." We could be wrong about that, but the absence of any specific lift number in the coverage is, on its own, a number.
Notice Me Senpai Editorial
