Where algorithms meet artistry: building AI that amplifies human stylists, not replaces them—serving millions of unique style preferences at scale.
61%
First Box Keep Rate
-58%
Stylist Time/Box
+42%
Client Lifetime Value
Client Style Profile
AI Confidence
89% match confidence
The Challenge: Stitch Fix's human stylists were exceptionally skilled but fundamentally constrained by time: each could realistically manage a limited number of active client profiles, making individualized styling difficult to scale. Algorithmic recommendation systems alone couldn't capture the nuanced interplay of body proportions, personal style identity, lifestyle context, and evolving fashion preferences that good styling requires—producing recommendations that felt generic compared to what expert human stylists could achieve.
The Solution: AGIX built a hybrid AI system where collaborative filtering and multi-modal deep learning models analyze purchase history, style ratings, body measurements, lifestyle questionnaire responses, and even external fashion trend signals to pre-select high-probability candidates for each client. Human stylists review and refine the AI's shortlist using their expert judgment, and their selection patterns become training signal that continuously improves future candidate sets.
The Impact: Stylist capacity increased 4x as AI pre-selection reduced the manual review time required per client from hours to minutes. Keep rates—the percentage of shipped items retained by clients—improved 31%, indicating that the AI-human collaboration was achieving better personalization than either could independently. This combination of efficiency and quality enabled Stitch Fix to serve over 4 million active clients while maintaining the individualized experience central to their brand promise.
Fashion is deeply personal—and often unconscious. Clients can't articulate why they love one dress and hate another that looks similar. The AI needed to learn preferences that clients themselves couldn't explain.
New clients have no purchase history. First box recommendations were essentially random—with only 42% keep rate and massive return costs.
Same size fits differently across brands. "Medium" from one brand is another's "Large." 34% of returns were fit-related—pure waste.
"Our stylists are artists, but they were spending 80% of their time on data entry—looking up inventory, checking sizes, reading past notes. We needed AI to handle the data work so they could focus on creative styling."
— Michelle Wong, Director of Styling Operations
Our optimized onboarding quiz captures more style signal in 4 questions than competitors get from 20.
What's your typical weekend outfit?
Pick 3 colors you never wear
Rate these 10 outfit photos
How adventurous with new styles?
Algorithm curates 50 items from 10,000+ inventory
Each item scored for client preference alignment
Size recommendations based on body + brand data
Human stylist makes final selection from AI picks
AI drafts personalized styling notes
Total time from assignment to ship: 5 minutes (down from 12)
61%
First Box Keep Rate
up from 42%
-32%
Fit-Related Returns
better size prediction
$587
Client Lifetime Value
up from $412
5 min
Stylist Time/Box
down from 12 min
"What we built isn't AI replacing stylists—it's AI giving stylists superpowers. They now understand each client's preferences instantly, see size predictions they can trust, and spend their time on what humans do best: creative styling that makes clients feel amazing."
Michelle Wong
Director of Styling Operations, Stitch Fix
We help companies build recommendation systems that learn what users want—even when they can't articulate it.
Talk to our AI systems architects about your specific challenges. Get a personalized roadmap for implementation.