Maisa Benatti, CEO of AIUTA – Interview Series

Maísa Benatti CEO at AIUTA is a visionary leader in fashion-tech and generative AI, currently spearheading the company’s mission to revolutionize digital fashion experiences from London since December 2024. In her role as Chief Executive Officer she’s focused on building scalable AI solutions that help brands produce premium visual content, enhance customer journeys, and unlock growth opportunities, drawing on her deep expertise in product design, UX research, and AI from her prior stint as Chief Product Officer at AIUTA and senior product leadership roles at companies like FARFETCH and Amazon. Her background spans generative AI for personalization and discovery, large language models, and e-commerce innovation, consistently driving forward AI applications that make fashion shopping more intuitive and engaging.

AIUTA is a modular AI innovation platform transforming how fashion brands and retailers engage shoppers online by integrating advanced technologies like virtual try-on, personalized styling, and AI-driven content creation directly into digital storefronts. The company’s suite of solutions enables immersive, interactive shopping experiences that boost conversion and reduce returns by letting consumers visualize apparel on themselves in real time, while also empowering brands to scale studio-grade visuals and outfit recommendations with proprietary AI models that handle lifelike rendering, body-preserving virtual fits, and seamless catalogue integration.

Your career began in fashion design and trend research before moving into customer-facing and product leadership roles at Farfetch and Amazon Fashion. How did those early experiences in fashion shape your perspective on where AI can create the most meaningful impact in commerce?

I didn’t come into fashion because I wanted to design beautiful things — I came into it because I was curious about how the industry actually works. Even early on, I was much more interested in questions like: How does this scale? Why is this expensive? Why does it break once it reaches the customer?

When I moved into customer and product roles at Farfetch and later Amazon Fashion, that curiosity turned into something very concrete. I could see how much of the online shopping experience depends on visuals, and how little those visuals actually tell you about how a garment will fit or behave on your body.

That’s where AI started to feel meaningful to me, not as a creative trick, but as a way to close the gap between what brands show and what customers actually receive. If AI could help represent clothing more honestly and at scale, it could improve trust, reduce waste, and make commerce work better for everyone involved.

AI in fashion is often surrounded by hype. What are the biggest misconceptions brands have about applying computer vision and generative AI to retail, and where do most initiatives fall short?

The biggest misconception is that if something looks impressive in a demo, it’s ready for production. In reality, those are two very different things.

Fashion is full of edge cases: different body types, fabrics, constructions, and brand standards. Many AI tools aren’t built to handle that complexity consistently. They might generate something that looks good once, but falls apart when you try to apply it across an entire catalog.

Another issue is that fashion is often treated like a generic visual problem. It isn’t. Clothing is a physical object, and if your system doesn’t understand how garments behave in the real world, the output may look appealing but won’t be trustworthy. The impact of that gap is significant — around 40% of online fashion orders are returned, which hurts both retailers’ P&L and the environment. When visuals misrepresent fit or appearance, uncertainty increases, and that’s usually where AI initiatives stall.

Many retailers cite fit and appearance uncertainty as a major driver of returns. From your experience working on size and fit at both Farfetch and Amazon, what does it actually take for AI to reduce returns in a measurable way?

Reducing returns isn’t about adding a feature, it’s about increasing confidence at the moment of purchase.

Size and fit also mean different things across regions. Fit expectations are cultural. A “perfect fit” in one market may feel too tight or too loose in another. AI can map those patterns and personalize recommendations based on regional behavior and individual preferences, not just measurements.

There’s also a structural issue in fashion imagery. Most products are photographed on one sample size, often clipped in studio to look more flattering. That creates unrealistic expectations. Shooting every garment on multiple body types would be ideal, but for most businesses, it’s operationally complex and prohibitively expensive.

That’s where AI becomes transformative. It allows brands to show how garments behave across diverse body types and skin tones at scale, with realism.

Realism is key. If what shoppers see feels truthful and consistent across the catalog, confidence goes up. When confidence goes up, returns go down.

Virtual try-on demos can look impressive in isolation, but scaling them across thousands of SKUs is a different challenge. What technical or operational hurdles emerge at catalog scale, and how has AIUTA addressed them?

At scale, you quickly realize that a good demo isn’t enough. From my experience running virtual try-on and fit initiatives at Farfetch and Amazon, the biggest challenges weren’t just accuracy-related. They were cost, speed, and operational complexity. Systems that worked in pilots became too expensive, too slow, or too manual once you tried to roll them out across a real catalog.

Latency is a big part of that. If a virtual try-on takes too long to load, customers simply won’t use it, no matter how accurate it is. That’s why performance was a core design constraint for us from the beginning. Today, AIUTA’s virtual try-on loads in about 4 to 7 seconds in production, which is significantly faster than most solutions on the market.

Operational complexity is just as important. Many solutions require heavy preparation, detailed inputs, or ongoing manual work from brand teams. AIUTA is designed to work from very simple inputs, with minimal effort required from retailers, while still preserving garment accuracy. As a result, we can move from generating hundreds of images per week to thousands of images per day, allowing brands to scale virtual try-on across large catalogs without adding operational overhead.

Finally, there’s consistency. Many systems start distorting fabrics or proportions once you scale. By controlling the full pipeline from garment capture and annotation to model training and deployment, we’re able to maintain garment identity and realism at the scale enterprise retailers actually operate at.

AIUTA combines computer vision pipelines with generative models. How do you ensure realism and garment accuracy while still delivering outputs quickly enough for enterprise workflows?

We’re very intentional about what we optimize for. Speed matters, but accuracy comes first.

Our systems are trained specifically on fashion data, and that distinction is important. We don’t rely on standard e-commerce datasets that tend to replicate the same narrow body types over and over again — we’re talking about typically tall, thin models. Instead, we own and have synthetically created a highly diverse dataset built around realistic human representations, with different body shapes, proportions, and garment constructions.

Because the models are trained on that diversity, they understand things like fabric texture, drape, and construction in a much more realistic way. As a result, we don’t have to “fix” outputs after the fact. Realism is built into the system from the start.

On the infrastructure side, we’ve also invested heavily in performance. The result is that brands can generate high-quality outputs in seconds, not minutes, which makes the technology usable in real production environments rather than just experimentation.

What are the biggest challenges enterprises face when integrating AI tools into existing creative, merchandising, and content production workflows?

Most enterprises already have complex systems and processes in place. The biggest challenge is that many AI tools are built as standalone products, not as part of a broader workflow.

For AI to actually be adopted, it has to integrate cleanly, technically and operationally. That means APIs, security compliance, predictable quality, and clear ownership when something goes wrong.

There’s also a trust element. Brands are understandably cautious about letting generative systems touch their core visual assets. That’s why reliability and quality control matter just as much as innovation.

You have led AI and personalization initiatives inside large marketplaces. How does building AI within a major platform differ from building a deep-tech company focused solely on fashion AI infrastructure?

Inside a large platform, AI is one of many moving parts. You’re often optimizing within existing constraints and balancing competing priorities.

Building AIUTA is very different because the entire company is focused on one problem: fashion visuals. That focus lets us go deeper, technically and creatively, and move faster when we see something break in production.

It also allows us to build long-term infrastructure rather than short-term features. We’re not just solving today’s use cases, but designing systems that can evolve as the technology does.

How do you balance personalization with privacy when working with sizing data, body representations, and shopper behavior signals?

Privacy has to be built into the system from the start. You can’t treat it as an add-on later.

At AIUTA, we focus on improving representation rather than collecting more personal data. By better understanding garments and body diversity in aggregate, we can deliver more relevant experiences without relying on sensitive individual information.

That balance is critical — especially in fashion, where trust plays such a big role in purchasing decisions.

AI-generated visuals raise concerns about beautified bodies or misrepresented garments. How do you approach authenticity and brand integrity when deploying generative systems at scale?

This is something we think about constantly, and for me it also became very personal.

During my pregnancy, I became much more aware of how limited most AI systems are when it comes to representing bodies that fall outside narrow norms. Pregnancy is temporary, but it’s a very real physical state, and yet it’s often completely missing from training data. I could see firsthand how easily AI fails when bodies change, even slightly, from what the system considers “standard.”

Many generative systems are designed to make things look “better,” which, for many, can wrongfully mean smoother, slimmer, or more idealized. But in fashion, that approach quickly breaks trust. If the body is beautified or the garment is subtly altered, customers end up with a product that doesn’t match what they were shown.

At AIUTA, we deliberately design for realism. Our goal is to show clothing on real bodies, including bodies that don’t fit narrow industry expectations, whether that’s different sizes, proportions, or more complex anatomies like maternity. We also pair AI with human quality control to catch edge cases and ensure every output aligns with brand standards.

Authenticity isn’t just a value for us, it’s essential for long-term adoption. If customers and brands don’t trust what they’re seeing, the technology simply won’t work.

Looking ahead, how do you see generative AI reshaping the broader fashion ecosystem, including digital twins, monetization models, and the future of creative production?

I think we’re moving toward a world where visuals are no longer static assets, but living systems.

Digital twins, for garments and for models, will become much more common. That opens up new monetization models and allows creative work to scale in ways that weren’t possible before.

More broadly, creative production will become faster, more flexible, and more responsive. Ultimately, the brands that succeed will be the ones that use AI thoughtfully and deploy it as infrastructure that supports accuracy, creativity, and trust.

Thank you for the great interview, readers who wish to learn more should visit AIUTA.