Here’s how Revolve and Lulus are turning AI investment into commercial value with Zelig’s technology and disciplined execution.

A 2026 Deloitte study of 200 retail and consumer-products executives found that 82% plan to increase AI investment over the next 12 months, yet just 16.5% can quantify a return. Another 33.5% aren’t focused on measuring ROI at all. Blue Ridge Partners surveyed over 300 consumer industry executives. They found a similar disconnect: the average retail respondent spent roughly $19 million on commercial technology and AI tools last fiscal year, more than non-retail counterparts, yet expected substantially lower returns.

Like most new technology implementation, retail AI fails not from a lack of capital, but from structural execution gaps:

  • Data Silos: Information trapped in isolated e-commerce, POS, and inventory systems creates a fragmented stack that degrades AI accuracy.
  • Perpetual Pilot Phase: Deloitte also reported that ownership lies with tech leaders (54%) rather than P&L owners, so AI is treated as an IT experiment rather than a core business strategy.
  • KPI Clashes: Chasing top-line conversion without guarding profitability inflates high-cost return rates.
  • Hype Over Utility: Deploying trend-driven features without addressing operational bottlenecks leads to low adoption and tool abandonment.

Retail transformation isn’t just an IT project. Technology must work across the retail ecosystem—from merchandising and planning to e-commerce, operations, and finance—with support from top leadership. The winners in retail AI aren’t those who spend the most or adopt the fastest, but those who define the business problem, establish KPIs, test rigorously, and scale only when the economics support it.

An Operating Model Built for Commercial Impact

Revolve , the Next-Gen fashion retailer, offers a clear blueprint for overcoming these structural traps. Its success stems from its essential design: while traditional retailers attempt to patch AI onto legacy infrastructures, Revolve’s co-founder and co-CEO, Mike Karanikolas, is a software engineer who built the business as a data science platform from day one. Karanikolas treats the company as an integrated AI operation, ensuring that technology investments serve the greater business model rather than existing as isolated IT experiments.

When Revolve evaluated Zelig ’s Build A Look technology—a virtual, personalized styling platform founded by fashion veteran and CEO Sandy Sholl—it did so through the same lens. The initiative wasn’t treated as a marketing gimmick; instead, it was subjected to the same rigorous data science standards that govern Revolve’s entire operating model.

Grace Hong, Chief Product Officer at Revolve, emphasizes that the company avoids deploying AI simply because it is fashionable.

“We have one of the most rigorous testing programs. We don’t deploy AI features for press or marketing purposes. Everything that hits our site goes through the wringer. We typically test on web first because deployment and iteration cycles are quicker and easier,” Hong said.

To avoid the pilot purgatory trap, Revolve evaluated Zelig’s speed, recommendation quality, and visual output across its vast catalog, scaling from an initial 9,000 SKUs to over 59,000 SKUs. Rather than running a single static test, Revolve and Zelig continually reset test periods as changes were introduced, ensuring clean data and clear ownership of each test cycle.

“We give it a new start date so that we have clean results moving forward,” Hong said.

The results answered the ROI question directly, showing that AI can align conversion goals with bottom-line profitability and leading naturally into the next set of outcomes.

“Among the customers who interacted with the feature, our initial rollout of Zelig’s technology outperformed what we expect from any new feature on our platform — significantly increasing conversion rate and lifting average order value by a strong double-digit percentage. We expected increases in those metrics, but the pleasant surprise was a meaningful reduction in returns, which directly impacted our bottom line,” Hong said. “Expanding Zelig into our native app was a natural next step. This is the kind of technology that visibly changes how our customers shop, and we’re excited to keep building with a partner who moves as fast as we do.”

By focusing on utility over hype, Revolve used Zelig to address a classic retail KPI clash: driving higher sales engagement without inflating return rates. High engagement can encourage shoppers to order multiple items to try at home and send the rest back, increasing reverse logistics, labor, and markdown exposure.

“I was pleasantly surprised because both return rate and conversion rate were pointing in the right direction together,” Hong said. “When I saw return rate decrease, it just made me realize it’s not just driving purchases, but it’s improving their decision quality. I think being able to see it on the model, being able to see it in context creates a very real-time experience that helps them make a better decision.”

Hong also points to Sholl’s retail grounding as key to solving user adoption. “She completely came to us understanding the customer, understanding the problems in retail and the online experience,” Hong said.

Why Lulus Is Launching Zelig

At Lulus , AI isn’t new. Mark Vos, President and Chief Information Officer, is somewhat skeptical of the sudden tendency to label everything as AI, which frames the question of why the company’s next move matters.

“AI is not new,” he said, noting that Lulus has worked with AI systems for five years or longer. Lulus has, for example, built its own machine-learning and predictive algorithms for inventory placement across its warehouse network and is fully AI-empowered in software development.

That background makes Lulus’ decision to launch an external platform notable. For Vos, the value lies in solving discovery friction rather than adding tech complexity. Traditional fashion e-commerce forces shoppers to view static products one at a time. Zelig enables customers to mix and match items and instantly see full outfits visualized on a model. “What Zelig does—it kind of levels it up,” Vos said, because a customer “can combine different pieces and actually visually see that rendered live.”

“We firmly believe that our customers are actually the chief merchants,” Vos said. “The customer can essentially merchandise their own closet and determine how our assortment is going to evolve.”

Rather than depending solely on merchant instinct, Lulus uses customer data and predictive algorithms to inform merchandising and inventory decisions. Vos sees the interaction data generated through Zelig as another potential signal that fine-tunes those decisions without creating data silos.

From A Shoppable Closet to Personal Agents

The technology is moving past individual sessions into features that unify fragmented data. Today, Zelig announced the official launch of its new “Shoppable Closet” experience along with an expansion of its real-time mix-and-match feature directly into Revolve’s native mobile app, with the rollout governed by the same operating standards across channels.

The Shoppable Closet integrates customers’ purchase history from the prior two years with saved, “hearted,” and items already stored under the Build A Look feature. Instead of starting from scratch each visit, a shopper can pair a jacket today with a dress purchased six months ago.

Hong described the new experience as “a deeper integration for that closet feature,” embedded directly into Revolve’s site.

“We are really excited about the closet feature because it’s something that we’ve talked about for decades, or as long as Revolve has been alive,” Hong said. “This is what personalization is really about: one-to-one.”

Part of Zelig’s operational appeal is its ability to deliver value quickly, without accumulating technical debt. Instead of integrating third-party APIs—tools that enable different software systems to communicate—Zelig invested five years building its own proprietary recommendation and vision models from the ground up. This foundation enables real-time, stylist-level fit and on-body simulations, with new features launching in under six weeks and minimal strain on internal tech resources.

Such persistence fundamentally changes the value of retail data. Traditional POS systems show what was bought and return logs show what was sent back. But those reports don’t explain why a product failed. Did the customer dislike the skirt? Was the price wrong? Was the fit wrong? Or could there be nothing wrong with the skirt at all, but she couldn’t find the top, shoe, or jacket that made the outfit work?

By functioning as an advanced customer management tool (CRM 2.0), this interactive digital closet tracks customer activity across marketing, purchases, and follow-up. It reveals what shoppers want to create, which items they combine most often, and where combinations don’t work—giving merchants deeper insights than traditional sales data alone can provide.

“If they do engage at a certain level but ultimately decide in favor of something else, that might be additional data points that we could use to fine-tune basically our merchandising or inventory decisions,” Vos said.

Looking ahead, Vos foresees static e-commerce sites increasingly becoming dynamic experiences powered by personal AI agents.

“I believe that the concept of a static website is going to go,” Vos said. “I believe that five years from now, you, Angela, will have your own shopping agent.”

“You control your data. You control your agent,” Vos said. “It’s the brands that need to tailor their information and their website or their purchasing to you and not the other way around.”

“What we need to do is… make our data consumable and interpretable in the most efficient way for the LLMs,” Vos said, describing a future in which “their bot needs to be able to talk to our bot.”

This future creates new challenges for retailers. Deloitte reports that about 58% worry they could lose direct relationships with customers as AI agents become more common. As shoppers rely more on AI to make purchases, brands must ensure their products and data work with these systems—while still maintaining valuable customer connections.

As Vos notes, the time saved by basic automation has a ceiling. “There’s only so much time…you can take out or become more efficient,” Vos said. “If you can combine AI to support your merchandising, your planning, your inventory, your pricing and the personalization all together, that’s where I think you have unlimited potential.”

Vos warns retailers not to get distracted by AI hype: “What kind of business are you? What is your reason that you’re out there? Focus on how you can use AI to make that much stronger and continue to build over time.”

The real ROI gap in retail AI isn’t a technology problem; it’s an execution problem. As Revolve and Lulus demonstrate, winning in the next era of retail belongs to brands with the operational understanding to turn complex data into better customer decisions, not those who deploy the most tools.