AI That Personalizes Beauty for
37M Members at Scale
Agix built the AI Personalization Platform for Ulta Beauty, powering hyper-personalized recommendations, intelligent search, and predictive marketing across every touchpoint, driving a 25%+ lift in conversion and 30% improvement in customer retention across 37 million Ultamate Rewards members.
America's largest beauty retailer, where 37 million members discover their next obsession.
Ulta Beauty is the definitive destination for beauty enthusiasts across the United States, operating 1,400+ stores in all 50 states alongside a thriving e-commerce platform. With over 25,000 products from 600+ brands, spanning prestige and mass beauty, skincare, hair care, and salon services, Ulta serves an extraordinarily broad customer base united by a shared love of beauty discovery.
At the center of Ulta's loyalty engine is the Ultamate Rewards program, 37 million members strong and one of the most active loyalty ecosystems in retail. Despite this scale, Ulta was leaving significant personalization value on the table: generic product recommendations, batch-and-blast marketing campaigns, and reactive inventory management meant that each member's deeply individual beauty journey wasn't being reflected in their experience. Agix was brought in to change that, at scale, in real time, across every touchpoint.
37 million members. 25,000 products. Zero personalization connecting the two.
Ulta's loyalty program generated enormous behavioral data, purchase history, browsing patterns, salon visits, beauty quiz responses, yet recommendations remained category-level rather than individual. Campaigns went out as segments, not signals. Inventory was managed by historical averages, not predictive demand. The gap between what Ulta knew about each member and what it actually showed them was vast.
End-to-end AI. From customer discovery to continuous loyalty.
Personalized recommendations, intelligent search, predictive marketing, inventory optimization, and clienteling, all connected through a single AI platform that learns from every interaction.
AI Personalization Platform: six interconnected systems that learn from every member interaction.
Agix built six integrated AI systems, turning 37 million members' behavioral data into individualized discovery, hyper-targeted marketing, and predictive inventory intelligence across every Ulta touchpoint.
Personalized Recommendations Engine
AI-driven product suggestions built on each member's purchase history, browsing behavior, shade matches, category affinities, and repurchase timing signals. The engine serves personalized carousels across the app, website, and email, updating in real time as behavior changes. For new members, collaborative filtering from similar beauty profiles bootstraps personalization from the first session. Recommendations account for what a member has already bought, eliminating redundant suggestions and surfacing genuinely novel picks aligned to their taste.
Intelligent Search & Discovery
Natural language search that understands beauty intent, "moisturizer for dry sensitive skin under $40" or "long-wear lipstick similar to what I bought last fall", returning semantically relevant results ranked by individual preference signals, not just keyword match. Smart filters surface shade range, skin type compatibility, and member review sentiment as first-class filters. Search results blend catalog relevance with personal relevance, so two members searching the same query see results ordered differently based on their distinct preference profiles.
AI Beauty Consultant
A conversational AI layer that provides real-time beauty advice, shade matching guidance, routine recommendations, and product comparisons, available across the app and website. The consultant knows each member's skin tone, undertone, skin concerns, and past purchases, enabling advice that's genuinely personalized rather than generic. It handles the discovery conversations that would otherwise require a trained beauty associate, freeing in-store staff for high-touch clienteling while extending advisory capability to digital channels at scale.
Predictive Analytics & Demand Forecasting
SKU-level demand forecasting that incorporates purchase velocity, social listening signals, seasonal trends, and loyalty program redemption patterns, predicting demand spikes before they hit store shelves. The model flags trending products 2–3 weeks before they go viral, giving replenishment teams a window to build stock before stockouts occur. Forecasts are localized by store cluster, accounting for regional preference differences in shade ranges, brand preferences, and category mix, reducing both stockouts in high-demand markets and excess inventory in lower-demand ones.
Hyper-Personalized Marketing
Behaviorally-triggered campaigns that reach each member at the moment of highest purchase intent, after a browse session, after a salon visit, after a shade match, or when a replenishment window opens for a frequently purchased product. The system replaces batch scheduling with AI-driven send-time optimization, subject line personalization, and offer selection tuned to each member's price sensitivity and brand affinities. Open rates improved from 12% to over 28%; campaign conversion climbed from 3.2% to 8.7% within six months of deployment.
Customer Loyalty & Retention AI
Churn prediction models that identify at-risk members weeks before they lapse, based on declining purchase frequency, reduced app engagement, and shifting browsing patterns, triggering personalized win-back offers before the relationship goes cold. Lifetime value scoring segments members into investment tiers, ensuring high-LTV members receive premium service levels and early access to launches. Loyalty point optimization nudges guide members toward redemption moments that convert to additional purchases, increasing redemption rates from 12% to 34%.
Every touchpoint personalized. Every recommendation earned by data.
The platform unifies member data across channels, app, web, store, and salon, creating a single living profile that powers personalization across recommendations, search, marketing, and in-store clienteling simultaneously.
Higher conversion. Deeper retention. Inventory that moves with demand.
Measured against pre-deployment baselines across Ulta Beauty's member base post-launch.
We had 37 million members telling us exactly what they loved through every purchase, every browse, every review, and we were sending them the same email everyone else got. The AI finally lets us have an individual conversation with each member at a scale no human team could manage. The redemption numbers don't lie.
Unified data across channels, the single change that unlocked every result downstream.
The core insight was that Ulta already had extraordinary data, 37 million members generating signals across app, web, store, and salon. The problem wasn't data poverty; it was data fragmentation. Member behavior across channels wasn't being connected into a single profile that could power cross-channel personalization. Unifying that data was the prerequisite for everything else.
The second key was timing. Beauty is an intent-rich category, members browse specific products at specific moments for specific reasons. Batch marketing that doesn't capture those moments delivers a fraction of the value of communications timed to actual purchase intent. The behavioral trigger system replaced the calendar-driven campaign rhythm with a member-driven one.
Channel unification created the single member profile
App, web, store POS, salon visit, and loyalty redemption data were siloed before the platform. Unifying them into a single member graph was the foundational step, every personalization capability downstream depended on seeing the whole member, not fragments of them.
Behavioral triggers replaced campaign calendars
Sending marketing at the moment of highest intent, not at a scheduled time, changed conversion math fundamentally. A member who just browsed a foundation and then received a personalized shade recommendation with a time-limited offer converted at dramatically higher rates than the same member reached on Tuesday morning with a generic 20%-off blast.
Social listening accelerated inventory lead time
Trending products in beauty often go viral within 48–72 hours of a high-reach post. By monitoring social signals and surfacing demand predictions early, the inventory system gave buyers a 2–3 week window to respond, converting stockout risk into first-mover advantage for members seeking trending products.
Churn prediction intervened before lapse, not after
Identifying at-risk members weeks before they fully disengage, based on subtle behavioral shifts, not binary absence, allowed proactive intervention with high-value win-back offers while the relationship was still recoverable. Retention AI operating at lapse was far more effective than re-acquisition marketing after it.
What this system doesn't solve.
Personalization AI compounds with behavioral history. It has real constraints for new members and in categories driven by impulse rather than pattern.
New Members Have Thin Personalization Signal
The system performs best for members with 3+ months of behavioral history. New members onboarding with a fresh loyalty account receive cohort-based recommendations, meaningful, but not individually calibrated. Cold-start accuracy improves significantly after the first 2–3 purchases. The onboarding quiz helps bridge this gap, but quiz responses don't substitute for revealed behavioral preferences at scale.
Gift and Gifting Behavior Corrupts Profiles
Beauty is a significant gift category. When members purchase for others, a birthday gift, a holiday set for a family member, those purchases signal preferences that don't belong to the buyer. The system doesn't currently have a reliable mechanism to flag gift purchases at the transaction level, which means profile noise from gifting can temporarily skew recommendations toward categories the member doesn't personally use.
Viral Demand Can Outpace Inventory Response
The social listening model identifies trending products 2–3 weeks early in most cases, but for products that go viral overnight via a single high-reach creator post, the prediction window narrows to days. Replenishment lead times for some international or indie brands can exceed the prediction window, meaning stockouts still occur for the most sudden viral events even with AI-assisted demand forecasting in place.
Shade Match Accuracy Requires Quality Input Data
The AI Beauty Consultant's shade matching capability is only as accurate as the member's self-reported skin tone data and the brand's shade range metadata. Inconsistent shade naming across brands, one brand's "light medium" is another's "medium", creates mapping challenges that the system resolves probabilistically but not perfectly. Members who don't complete shade profile onboarding receive lower-confidence shade recommendations as a result.
Is an AI Personalization Platform the right build for your retail business?
What powers this system.
AI Personalization Engine
Member behavior modelling, preference inference, and real-time recommendation ranking, the data and ML layer that powers individualized product discovery across every channel.
Predictive Marketing Automation
Behaviorally-triggered campaigns, send-time optimization, and individualized offer selection, replacing batch scheduling with member-driven marketing that reaches each customer at peak intent.
AI Beauty Consultant
Conversational AI that provides shade matching, routine building, and product guidance at scale, extending the expertise of trained beauty associates to every digital touchpoint without staffing cost.
Inventory Intelligence
Social-signal-driven demand forecasting, SKU-level trend prediction, and localized replenishment recommendations, reducing stockouts on viral products while preventing overstock in low-demand markets.
Loyalty & Churn Prevention AI
Predictive churn scoring, at-risk member identification, proactive win-back offers, and LTV-based investment tiering, intervening before members lapse rather than re-acquiring them after they leave.
Omnichannel Retail AI Systems
End-to-end AI purpose-built for omnichannel retail, unifying member data across store, digital, and loyalty channels into a single profile that powers consistent personalization everywhere a member shops.
Common questions about AI-powered personalization for loyalty-driven retail.
The platform unifies member behavior across channels through the Ultamate Rewards ID, every in-store purchase, digital browse, app session, and salon visit is attributed to the same member profile regardless of where it happens. A member who browses foundation online and then buys skincare in-store generates a unified behavioral signal that informs recommendations across both channels. This cross-channel stitching is what distinguishes the platform from session-based personalization that loses signal when a member switches between digital and physical touchpoints.
Send timing is driven by behavioral triggers, specific member actions that indicate elevated purchase intent, combined with a learned model of each member's historical purchase timing patterns. When a member browses a product for 90+ seconds without purchasing, or views a shade match result, or when their repurchase window for a frequently bought product opens, the system evaluates whether a personalized communication is warranted. It balances intent signals against communication frequency limits to avoid overreach, a member will never receive more than a configured maximum number of messages per week regardless of trigger volume.
The model monitors multiple leading indicators simultaneously: social media mention velocity for specific products and ingredients across TikTok and Instagram, Ulta's own browse and wishlist add rates (which often spike days before purchase velocity does), search query volume for specific product names, and cross-brand purchase patterns (when members who bought a trending brand start buying related categories). By weighting these signals in combination rather than relying on any single indicator, the model identifies demand inflection points earlier than sell-through data alone would reveal.
Yes, though with configuration adjustments for scale. The core personalization architecture, behavioral signal collection, preference modeling, triggered marketing, churn prediction, works at much smaller member bases. At lower scale, collaborative filtering relies more heavily on category-level cohorts rather than individual behavioral history, and cold-start performance is a larger fraction of total interactions. The inventory forecasting component becomes less valuable at smaller SKU counts. We typically recommend the full platform for retailers with 50K+ loyalty members and 300+ SKUs; smaller retailers benefit from a focused deployment of 2–3 components rather than the full six-system build.
A standard chatbot operates on scripted decision trees and keyword matching without memory or member context. The AI Beauty Consultant is built on a large language model fine-tuned on beauty domain knowledge and grounded in each member's complete profile, skin tone, undertone, skin concerns, purchase history, stated preferences, and past recommendations. It can answer "what foundation would work for me?" with a specific, personalized answer rather than a generic response, because it knows the asking member's profile. It can also cross-reference the catalog to recommend specific in-stock products that match the member's shade range and price sensitivity, not just general product categories.
Ready to turn your loyalty data into individualized experiences that convert?
Most projects go from kickoff to deployed AI system in 8–16 weeks. Let's talk about what a personalization platform, predictive marketing engine, and inventory intelligence system could do for your member retention and conversion rates.