AI brand agents represent a company. They automate work that customers ask them to do, making them crucial frontline elements of the customer experience.
Stitch Fix, a clothing service that helps consumers find -- and update -- their wardrobe style, is known for its human stylists who craft their customers' looks. After a tumultuous post-pandemic period that precipitated leadership changes, Stitch Fix currently claims more than 2 million customers as it works to right the ship. The company managed to post revenue gains late last year despite tariff turmoil that created unsteady import pricing.
Enter agentic AI, which helps stylists craft the customer experience. We talked with Tony Bacos, chief product and technology officer at Stitch Fix, to discuss the method of melding human and synthetic fashion curation into its continuously improving conversational AI Style Assistant, launched in 2025 on a foundation of 15 years of customer data.
Editor's note: The following interview has been edited for clarity and length.
There have been ups and downs, as well as leadership changes, at Stitch Fix over the last few years. Where are you now?
Tony Bacos
Tony Bacos: I've been with Stitch Fix for a little over two years. I followed Matt Baer, the CEO, by four or five months, something like that. In those two years and change, we've been on a transformation journey. We've been working to stabilize and turn around a business that had been in a pretty consistent decline for a couple of years. We've been pretty successful in executing that transformation, including stabilizing the business, introducing some retail best practices, [best practices] in how the core business is operated, and leaning into building a better client experience.
Now we're at the point where we're thinking about that next wave of our transformation.
AI was part of Stitch Fix long before generative and agentic came along. Explain that evolution up to now.
Bacos: We just celebrated our 15th birthday. We are kind of OGs of AI-forward companies, you could say, at least in retail. The business model begins with a team of stylists who shop on behalf of clients. They do that with a rich amount of information that clients share with us about their preferences -- everything from the price points that they like to shop in to style and fashion preferences, fit, brands, you name it.
From the beginning, AI ingested all the preference information clients shared with us through an initial onboarding quiz -- as well as updated information we collected over time by watching their behavior. What do they keep? What do they return? What kind of feedback do they provide on all the items? All that information is used to improve our recommendations algorithm.
That algorithm takes a broad assortment of products and narrows it down to a shoppable, digestible range for the stylist to shop from. AI is a way to take a vast amount of information about both products and client preferences and use it to curate a storefront for stylists. That's been the formula for us since day one.
Then came generative and agentic AI.
Bacos: Over the last couple of years, we've gone quite a bit further with AI than recommendation algorithms. Now we're using AI to build client experiences.
We've recently beta-launched Vision -- a virtual try-on experience with the added benefit of our algorithms and us picking the outfits for clients, so they don't have to browse the site, find something they're interested in, and click a bunch of buttons. For clients who have opted in, we deliver a new set of images every week, dropped into their Inspiration Gallery. Over time, it begins to look like an Instagram reel. These photos are not of you in a cold, antiseptic dressing room. They're photos of you in the real world, outside, in a business setting.
We do thematic image drops. For the Super Bowl, we did tailgate parties. For the holidays, we did holiday parties -- things that people look at and really connect with, because they look like them. It feels like them. It's a version of their best self.
Generative AI has a reputation for understanding searcher intent. It can widen the understanding of queries to natural language. That has to be a big benefit for Stitch Fix, because consumers aren't necessarily fashion insiders with the lingo.
Bacos: There's another capability we launched recently that helps the client articulate what they're looking for in ways that a lot of us people who aren't in the fashion industry sometimes have a hard time describing -- such as, "What do I want?" "What's it look like?" We just don't have the vocabulary. We don't know what that particular pattern or whatever is called.
We've had a feature called Request Notes for some time. Clients can write their stylist a note when they have a fix [clothing shipment] coming up. They can tell the stylist that maybe they've got a wedding they're going to, a trip, or maybe they've just recently lost some weight and want a closet refresh.
Those have been effective since their introduction -- but only as effective as the client's ability to articulate what they're looking for. A lot of times, the stylist will read a note and go, "Man, I wish I knew what they meant." It's very ambiguous.
So we introduced an AI assistant into that flow, where, as a client starts writing the note, the AI will interpret. It will generate some photos and say, kind of, "Is this what you mean? Is this what you have in mind?" And it will act like an editor to help a client express what they're looking for much, much more clearly.
We're very transparent with clients about this. We don't pretend it's a human. We don't pretend it's a stylist. We say upfront, "This is AI. We're trying to help make sure that your stylist understands what you're looking for." That translation layer and the additional context have been phenomenal for our clients because it improves the quality of the fixes that we send them. We see the metrics around what items a client keeps and whether they're happy with that fix, and whether they want the same styles again; all those things go up when they use this AI tool. That's another example of using generative AI in a fairly novel way that's fit to our business process, and not just a cheap party trick.
So what's under the hood, IT-wise?
Bacos: We have a fairly sophisticated mix of handoffs through some of these processes that include a combination of large language models and in-house tools that we have built. We string them together into workflows that are essentially agentic but also use the best of what each of these components does. It's not built on a single foundation model or even a single vendor. We're trying to use the best of all of these to ultimately improve our client experience.
Fashion is constantly moving forward; it's six months ahead of what we see on the street. How can generative AI keep up with that? How can it spot the next great thing when it is incapable of coming up with its own ideas -- it just parrots back the information you fed into it?
Bacos: AI in its current form -- as exciting and mind-blowing as it is -- is a sophisticated auto-complete tool. It takes all the data that you have, tries to anticipate what you're trying to accomplish and auto-completes. It does it one word at a time.
In most cases, that's not how fashion works. Fashion is not a predictable sequence of "We have this, and therefore, naturally, we're going to have that next." Fashion is subjective. It's influenced by the whim of humans in design houses and at brands who have their own vision of what's going to come next. There's a little bit of a Thunderdome of ideas around those trends. The winners emerge and people resonate with them.
We believe the role of human stylists will remain critical. Humans are not just looking in the rear-view mirror, trying to keep the same thing going.
Tony BacosChief product and technology officer, Stitch Fix
We believe the role of human stylists will remain critical. Humans aren't just looking in the rear-view mirror, trying to keep the same thing going. They're staying on trend and they realize our clients want to as well.
AI is going to continue to struggle with the ability to look forward and to have credibility determining the next season, the next color, the next silhouette or whatever. It's what makes this industry fun -- it is always changing.
It's easy to picture an AI agent putting us all in mom jeans, hoodies and other comfortable garments that everyone upvotes because they love to wear.
Bacos: We're familiar with the problem of narrowing choices and getting into a rut over time. Even our best algorithms, if we let them, if we're never pushing them, challenging them and telling them, "Hey, here's the new trend, here's the new thing," they are just going to narrow it down gradually. They're going to say, "Oh, Tony loves puffy vests," and "This shirt he kept last time, send it to him again because that's probably the safest bet."
Our teams have to continually nudge our algorithms, stylists and clients to make sure we're not just optimizing for what you used to like. But we're also in partnership with each client. We gently nudge them forward and say, "Hey, this is what's in this year. What do you think?"
Don Fluckinger is a senior news writer for Informa TechTarget. He covers customer experience, digital experience management and end-user computing. Got a tip? Email him.
Tony Bacos
