E-Commerce Lost Its Sales Associates – This Brings Them Back

01 /The Concept

Physical retail stores have sales associates – people who answer shopper questions or proactively ask "What are you looking for?" or "Can I help you find something?" Annoying as that can sometimes be, stores hire them for good reason: in aggregate, they sell more.

Remark brings that same dynamic online, connecting e-commerce visitors with real human experts who can help them find the right product.

The platform opens by asking a visitor a few questions to understand what they're looking for and where they're stuck, then routes them to an available expert on that topic – who can answer questions, give personalized recommendations, and suggest specific products from the store's catalog, all via live chat.

The platform's AI records every conversation, extracts the substantive insights, and automatically generates a landing page for that product category – one that addresses the real questions and criteria real shoppers bring to it. These pages are well-suited for search optimization because they're built around the actual language people use when searching.

Experts on the platform are professionals who work in fields where they use the products in question. Sporting goods stores, for instance, can connect shoppers with a climbing instructor, a yoga coach, a figure skater, or a member of a national ski team.

Connecting buyers with genuine experts measurably improves purchase conversion. And good recommendations – ones actually grounded in what the shopper needs – significantly reduce returns, since people are buying with understanding rather than guessing. That's not a small thing: return rates after physical retail purchases run around 3%, versus 25% for online purchases.

Experts also tend to recommend better products, which lifts average order value (AOV), and they often flag accessories or add-ons a less experienced buyer wouldn't have thought to consider.

The numbers back it up: Remark clients have seen visitor-to-buyer conversion climb to 20% and total revenue rise by 9%. The average time to connect a visitor with a relevant, available expert: 12 seconds – well within the tolerance of an engaged shopper.

Remark launched its platform two years ago. Since then, it has onboarded 45 e-commerce stores and 50,000 experts, who have collectively completed 100,000 customer conversations.

The company just raised its first round of funding: $10.3 million.

02 /Why It Matters

"People who talk to salespeople buy more and more often" – Remark's claim, and it generally holds true.

It's also a useful sales funnel insight: the willingness to enter a conversation is itself a qualifying signal. Visitors who won't engage probably aren't worth pursuing further down the funnel.

This observation gave rise to the "conversational commerce" category, which triggered a wave of chatbot development – bots that automatically answer visitor questions, increasingly powered by AI.

But chatbot conversations are too transactional. Bots can't read emotional nuance, pick up on hesitation, or mirror a human's mood. And what shoppers often want from these conversations isn't just facts – it's personal experience: what the expert actually used, what they concluded from it, and what they'd genuinely recommend based on having been through it.

Result: most chatbots frustrate people and move the conversion needle only modestly. Live experts perform significantly better – but live experts are expensive.

Remark's answer is to take the best of both worlds. It's building AI agents trained on the successful conversations of its live experts – capturing not just arguments and examples, but communication style.

The goal is to create digital twins of real people, not a generic unified chatbot. Conversations through Remark should eventually be a seamless flow between AI agents and human experts – with live people stepping in specifically when the AI senses the conversation isn't moving toward purchase effectively enough.

This raises an unresolved question about compensation. When there's no AI involved, an expert earns a commission on sales they close. But when the AI was trained on conversations from multiple experts – should those experts keep earning commissions on sales the AI agent initiates? If so, in what amount, and how do you split it among everyone whose data was used? As the pool of training experts grows, each individual's share of any commission would approach zero.

Alternatively, maybe each AI agent should be a precise digital twin of one specific expert – one the store can hire or fire. That's actually a compelling model.

Either way, this is an open question the industry hasn't settled yet.

Several other startups are building in the same direction of connecting third-party human experts to online sales and support funnels.

Experify ([covered here](/review/poshhupat-i-poslushat)) built a platform that connects e-commerce shoppers with people who already own the product they're considering, so they can get a real first-person opinion. Since that review, Experify has added conversation-based review generation that gets published on review sites – similar to how Remark auto-generates landing pages from its chats. Experify raised $4.03 million in its first round.

Limitless ([covered here](/review/tehpodderzhka-na-frilanse)) built a technical support platform that matches experienced product users with newcomers, with smart routing that directs questions to the right expert. Limitless has raised $20.5 million.

03 /Opportunities

These platforms are essentially running an Uber model – but for customer support and advisory services.

And just like original Uber, AI is coming for part of the job. In ride-hailing, AI may eventually replace drivers; in these human-powered support platforms, AI agents trained on real expert conversations may eventually substitute for or augment the live experts. In both cases, you have to start with humans – because humans still do the job better.

So the direction worth pursuing is building human-AI hybrid platforms for buyer and user support.

The general architecture is now understood, and the necessary technology exists. The remaining open question is how to compensate the human experts whose conversations train the AI, and who may still be called in for complex situations – essentially functioning as ongoing fine-tuning contributors.

These platforms likely have applications well beyond what's been described here. A coding assistant that handles most queries via AI but can instantly escalate to a live senior developer for tricky problems is an obvious adjacent use case.

And that pattern probably applies beyond programming. Where else do you see genuine demand for human-AI hybrid support platforms? Or do you think AI will simply get good enough to handle it all on its own?