An interview with Sameer Singh, CTO at Envive and former Professor of Computer Science at UC Irvine.

How do you see AI transforming e-commerce?

There are two main areas where AI is going to make a major impact.

First is the consumer side. Shopping today can be frustrating — customers either have a very specific idea of what they want or only a vague sense of it. AI can act as a personal shopping assistant, helping people communicate their needs (even the implicit ones) and match them to the right products.

For example, if someone says they want to spend $100, AI might be able to suggest a product for $110 that offers significantly more value — and help them see why it’s worth it. It’s not just about recommending products; it’s about understanding human preferences, constraints, and trust.

This experience won’t just live inside a chatbot. Every part of a website — what images are shown, which descriptions are highlighted, how navigation feels — can adapt dynamically to the user. AI-driven personalization can make every digital storefront feel like it was designed for that specific shopper.

Then there’s the merchant side. AI gives brands access to incredibly rich behavioral data — how users browse, what they add to cart and remove, when they convert, what triggers hesitation. Historically, these insights have been analyzed at a surface level using aggregate data. AI enables a much deeper, more nuanced understanding of why people shop the way they do.

This not only feeds back into improving customer experience, but also helps merchants make smarter decisions about everything from product design and pricing to marketing language and website structure. It’s about using data to simulate and predict outcomes before taking them live — applying AI not just for automation, but for better creative and strategic thinking.

You were a professor at UC Irvine before joining Envive. What did you focus on in your research and teaching?

My academic focus was primarily on AI, machine learning, and natural language processing — essentially, how computers understand and generate language.

Teaching was one part of the role, but a lot of my time was spent running a research lab with PhD students, exploring questions at the frontier of how machines interpret meaning. We published papers, built models, and tried to push the boundaries of what AI could understand and explain.

How do you think AI models are going to evolve in the coming years?

So far, most AI models have focused on publicly available data — text, images, videos, music. That’s what’s helped them become so powerful. But there’s a ceiling there: models are only as good as the data they’re trained on.

The next frontier is real-world and proprietary data — data that comes from how people actually use AI systems in context. Think about robotics, navigation, or e-commerce. When users interact with AI in these settings, they generate valuable behavioral data that helps models learn what works and what doesn’t.

In e-commerce specifically, every click, every conversation, every abandoned cart becomes part of that feedback loop. AI will increasingly learn from the real world, not just about it — adapting dynamically to user behavior instead of relying on static, pre-trained datasets.

What advice would you give companies trying to stay ahead of these changes?

Have a robust framework for evaluating AI.

Too often, companies get excited about impressive demos — those “magical” moments when an AI system does something remarkable — and assume that performance will hold in production. It rarely does.

Organizations need to rigorously test for all possible edge cases, iterate continuously, and use AI to help evaluate itself. Building this kind of systematic evaluation early on is how you reduce risk and make AI deployment sustainable at scale.

How do you see machine learning specifically changing?

A few years ago, “machine learning” meant that you trained your own models on your own data. You had control, and your model improved as your data improved.

Now, we’ve shifted toward relying on massive, centralized AI models — the ones built by OpenAI, Google, Anthropic — which don’t learn continuously from your specific data. They update when those companies decide to retrain them, not when your customers interact with them.

In that sense, we’ve lost some of the “learning” in machine learning. These algorithms were designed to evolve, but today’s implementations are largely static. One of the reasons we started Envive was to bring that continuous learning back — where models improve through every real interaction, not just periodic updates.

Envive emphasizes brand safety as a key differentiator. Why is that so important?

For most merchants, what makes them unique isn’t just the product — it’s their brand identity. The tone, the aesthetic, the trust they’ve built with their audience.

When brands use generic, off-the-shelf AI tools, they often risk losing that identity. The model might describe a product incorrectly, make comparisons that don’t fit, or even “hallucinate” information — and suddenly, the brand looks inconsistent or untrustworthy.

That’s why brand-safe AI matters. It ensures that every interaction — whether it’s a chatbot conversation or a product recommendation — aligns with how the brand wants to represent itself. In many cases, we’ve seen brands lose conversions not because of poor product quality, but because their AI made them look unreliable.

Envive focuses deeply on protecting that relationship between brand and customer, making sure the model always represents the brand’s voice, values, and credibility.

Takeaways

AI is no longer just a backend optimization tool — it’s becoming the core interface between customers and brands.

From personalized shopping experiences to intelligent data feedback loops, AI has the power to transform how people discover, evaluate, and connect with products. But as Sameer Singh reminds us, the real breakthroughs will come from learning in context, evaluating responsibly, and building AI systems that protect the brands behind them.