Stitch Fix

By its 2021 fiscal year, Stitch Fix had crossed $2 billion in annual revenue for the first time and was serving almost 4.2 million active clients (Stitch Fix, financial results, September 2021). What followed was a sharp reversal: sales and active clients fell for seven consecutive quarters, the company laid off 15% of its salaried staff in mid-2022 and a further 20% in January 2023, alongside a CEO change, as it closed its UK operation and consolidated its distribution network (CNN, January 2023; Modern Retail, September 2023). At the same time, generative AI was making it more plausible than ever to remove the human stylist from the loop altogether: by mid-2023 large language models could already interpret a client’s style feedback, draft outfit combinations and generate marketing copy at a scale no human team could match (Stitch Fix Newsroom, June 2023). For a company under real cost pressure with thousands of stylists on its books, the dilemma was concrete: let the now-much-better algorithm make the final pick too, or keep paying people to do it.

Two paths were available. Path A: full prediction, no human judgment in the loop — is best illustrated by an idea from outside fashion retail entirely. In 2012 Amazon filed, and in December 2013 was granted, a patent for “anticipatory shipping”: a system that would use purchase-prediction algorithms to ship products toward a customer’s area, or even to their door, before an order was placed (Practical Ecommerce, 2014). It was a bet that prediction accuracy could eventually substitute for the moment of human choice entirely, flipping “shop then ship” into “ship then shop.” A decade on, independent reviewers found no public evidence that Amazon had ever deployed the system at meaningful scale (Logistics Viewpoints, September 2023); the patent’s own text acknowledged that badly predicted shipments would have to be returned, redirected or given away, and those costs are exactly what stood between a clever algorithm and a workable business (Jafari Law Group, patent analysis). Path B: the route Stitch Fix chose, including through its leanest years — kept a deliberate split between prediction and judgment: hyper-digitise everything that is genuinely a forecasting problem (demand, logistics, an initial shortlist of items by size, fit and style), and keep a human stylist responsible for the final, judgment-based pick that actually goes in the box. Leadership stated this explicitly even at the company’s commercial peak: announcing its first $2 billion revenue year, then-CEO Elizabeth Spaulding credited the results to “our unique combination of data science, and creative human judgement” (Stitch Fix, financial results, September 2021).

Stitch Fix’s recommendation engine has always done the first pass on its own: when a client requests a Fix, algorithms interpret that client’s fit and style data to surface a shortlist of items from hundreds of thousands in inventory, which a stylist then curates from rather than searching manually (Stitch Fix Newsroom, June 2023). From 2023, the company layered generative AI on top of that backbone in three concrete ways. It used OpenAI’s embeddings inside its own recommendation models to interpret the freeform text feedback clients had built up over years, and used GPT-4 to summarise that history so stylists could grasp a longtime client’s preferences in seconds rather than rereading dozens of notes (Stitch Fix Newsroom, June 2023). It used GPT-3 to draft advertising copy and product descriptions for review by human copywriters, cutting the time to plan and write an ad campaign from roughly two weeks to under a minute of review per asset, with a 77% pass rate, and generating up to 10,000 product descriptions every 30 minutes for human merchandisers to check (Stitch Fix Newsroom, June 2023). And it built its own Outfit Creation Model, trained on outfits its stylists had created, to generate roughly 13 million new outfit combinations a day and surface around 43 million to clients across their shopping feeds, emails and ads (Stitch Fix Newsroom, June 2023).

In August 2025, CEO Matt Baer announced a further suite of generative-AI features explicitly designed to free up stylists’ time for client relationships rather than replace them: a conversational AI Style Assistant to help clients articulate style preferences, Stylist Connect, a chat platform letting clients message their stylist between Fixes (client satisfaction with the feature was reported at nearly 100%), Family Accounts for styling an entire household, and an early version of AI-generated style visualization (Stitch Fix Newsroom, August 2025). That visualization feature launched publicly in October 2025 as Stitch Fix Vision: clients upload a selfie and a full-length photo, and the platform generates personalized images of themselves wearing shoppable outfit recommendations, delivered weekly into a personal “Vision gallery” (Stitch Fix, press release, October 2025). Stitch Fix’s Chief Product and Technology Officer Tony Bacos described it as “an entirely new approach to style discovery, unlike any of the existing ‘virtual try on’ experiences that require shoppers to do all the work” (CX Dive, October 2025).

None of this replaced the roughly 1,600 human stylists who still curate every Fix, though their working arrangements changed substantially over the downturn: the company moved its styling workforce to a fully part-time, flexible-hours model in March 2024 after most stylists had already shifted away from full-time roles, and stylists who chose to leave received severance, healthcare and career transition support (Retail Dive, January 2024). What stayed constant was the division of labour: Stitch Fix’s Vice President of Styling and Client Support, Sarah Funderburk, described it as “a fairly intuitive separation of who does what best,” with AI narrowing down options so that “the human can step in and make all of the connections that are nuanced to the client” (US Chamber of Commerce CO—, July 2025).

This is an upskilling case, not a reskilling one: Stitch Fix’s stylists are still stylists. The role and its core function — picking the right items for a specific client — never changed; what changed is that AI now does the prediction-heavy groundwork so each stylist can spend more of their time on the relationship and judgment work that clients actually pay for, which is exactly the “existing role made more capable” pattern the upskilling definition describes.

The Amazon comparison sharpens why this mattered. Anticipatory shipping bet everything on prediction accuracy eventually being good enough to remove the human decision entirely, and a decade later there is no evidence it was ever deployed at scale: the cost of guessing wrong, acknowledged in the patent itself, undercut the economics of a judgment-free model (Logistics Viewpoints, September 2023). Stitch Fix never made that bet. By keeping a human stylist as the explicit, permanent owner of the final decision, it could push AI aggressively into every part of the business where prediction genuinely is the right tool — forecasting, logistics, ad copy, product descriptions, outfit generation, and now photorealistic style visualization — without ever having to ask an algorithm to take responsibility for a judgment call it could get expensively wrong.

What makes the case durable rather than just well-timed is that the architecture survived the company’s worst years intact. Between 2021 and 2024, revenue fell by roughly a third, active clients nearly halved, and the company cut staff twice and changed CEOs while closing a national market and consolidating its warehouse footprint (CNN, January 2023; Modern Retail, September 2023). Even under that pressure, with generative AI more capable than ever of producing a passable outfit on its own, Stitch Fix’s response was to renegotiate how stylists worked, not to remove them from the loop. That decision is now central to the company’s own account of its turnaround: heading into its third year of recovery, with revenue growth accelerating quarter over quarter, the company’s leadership credits the combination of AI and “the expertise of our human Stylists” by name, not AI alone (Stitch Fix, financial results, December 2025).