AI Product ScoutAI commercialization archive
← Back to archiveRemark cover
Remark

Remark: Why E-Commerce AI Should Sell Expert Judgment, Not Chat

Remark shows how e-commerce AI can commercialize by turning expert product judgment, buyer questions, recommendation conversations, traffic-based pricing, and conversion evidence into a measurable on-site sales layer.

AI ProductsCommerceMarketingSales TechAI Agents

A shopper opens an outdoor brand’s website to buy cycling shorts. He knows he will ride long distances and expects hot weather, but he does not know which padding, cut, or fabric will work. The product page lists parameters. Reviews say the shorts are comfortable, except for the people who say they rub. No one on the page can absorb the actual question.

In a physical store, this is a normal sales moment. A person who has ridden long distances can ask two or three questions and narrow the choice to a few products. E-commerce moved the inventory online, but it often removed the most valuable part of the store: human judgment at the moment of hesitation.

Remark is trying to put that judgment back into the website.

It is not best understood as a polite customer-service bot. It is an AI shopping advisor that compares products, asks follow-up questions, understands use cases, and recommends what to buy. The more important product detail is that Remark did not begin by letting a model invent that advice. It first put real product experts into conversations, then used their knowledge and selling behavior to train AI advisors.

The company raised a $16 million Series A in 2025. TechCrunch reported at the time that Remark had moved from taking a cut of transactions to a SaaS model priced by website traffic. That change is more interesting than the financing. Remark is no longer only betting on its share of a single order. It is selling a continuously running on-site sales layer.

Treat The Question As The Expensive Traffic

The common waste in e-commerce is not only a lack of visitors. It is that a visitor has already expressed uncertainty, and the page still cannot answer well.

“Can this jacket handle ten degrees below freezing?” “My dog pulls hard. Will this harness hold?” “I need a carry-on bag that fits a laptop.” These are not ordinary support tickets. They are sales moments. A good answer can move the shopper toward checkout. A generic answer sends the shopper elsewhere.

Remark’s product puts brand product knowledge, real expert experience, and the live catalog into the same conversation. Its company materials say expert sources can include chefs, stylists, skin-care specialists, athletes, and other domain people. The point is not to add a human-looking avatar. The point is to teach the system which question to ask first, what information changes the recommendation, and when it should advise against a product.

That differs from ordinary site search. Search matches products against conditions the shopper has already provided. A good advisor helps reveal the conditions. “For commuting” is not enough. Laptop or no laptop? Subway or bike? Waterproofing or weight? Store associates once handled those distinctions through experience. Remark turns them into a repeatable conversation.

It Captures How People Sell, Not Only What Products Are

Many brands already have product pages, knowledge bases, and thousands of reviews. But those assets mostly answer “what is this product?” They rarely explain “why should someone like you buy this one?”

The judgment that closes a purchase often lives in sales associates, support people, returns teams, and heavy users. They know how buyers describe problems. They know which follow-up question changes the path. They know when a cheaper item is actually the better recommendation. That is exactly the data that a generic chatbot does not have.

Remark’s early use of human experts was not just an attempt to make AI feel more human. It was a way to collect scarce commercial data: how buyers state needs, how experts clarify those needs, which answers convert, and which recommendations reduce returns.

This also explains why Remark can talk about outcomes rather than chat volume. The company says its AI advisors cover more than 60 brands, that revenue grew fourfold over the past year, and that customer retention is 100 percent with roughly 130 percent net revenue retention. These are company-disclosed figures, not independently audited benchmarks. Still, they show that Remark is trying to prove value through sales results and renewals, not merely through the number of conversations.

Its customer cases make the same argument. PEARL iZUMi reports that shoppers who spoke with the AI converted at 30 percent and that return on ad spend was about seven times higher. DECKED’s case study is framed around millions of dollars in revenue. These results come from Remark and customer case materials, and they do not disclose the full attribution method or control group. But for a brand, they create a budget-level claim: which visitors were saved, and which orders appeared after expert advice entered the page.

From Commission To Traffic-Based SaaS

Remark’s earlier model took a share of transactions. That made it resemble an outsourced salesperson: little revenue if it did not sell, more revenue if it did. The model can win early customers, but it has trade-offs. Attribution becomes contentious. Revenue depends on seasonality and promotions. The product team can be pulled toward last-click behavior rather than long-term trust.

Moving to traffic-based SaaS changes the business meaning. The brand is no longer buying “your share of this order.” It is buying advisory coverage across high-intent website visits. Remark can then price against coverage, conversation volume, catalog scope, and continuous optimization rather than waiting for one order to settle.

That pricing matches the product. Users do not visit a brand site because they want to chat with AI. The conversation happens while they are already choosing, comparing, or hesitating. Pricing by traffic acknowledges that Remark is infrastructure before the transaction, not a decorative marketing widget.

It also gives the product a clearer expansion path. If the advisor proves that it can capture high-intent questions on product pages, it can extend into paid landing pages, bundles, returns reduction, and support. The same knowledge that recommends the right product can help explain why the wrong product should not be bought.

The Risk Is Turning Advice Into Pushing

The obvious failure mode is that the AI recommends the most expensive item every time in the name of conversion. That may make short-term metrics look good. It can also create returns, bad reviews, and long-term damage to brand trust.

Remark’s product logic points to a harder standard. A good advisor is not merely fast. It narrows the choice when the shopper is a fit, and it backs away when the fit is wrong. The brand has to give the system inventory, pricing, promotions, and real product limits. The system has to turn those details into specific, honest advice.

Public information is not enough to know whether Remark can do this equally well in every category. The cost of sourcing human experts, the process for keeping product knowledge fresh, and the ability to reproduce conversion data across many brands remain open questions. E-commerce attribution is also inherently complex. A conversation before purchase does not mean the conversation alone created the purchase.

Even with those limits, the opportunity is concrete. Online stores do not primarily lack more product information. They lack someone who can turn information into a choice.

If AI can convert expensive, scarce, experience-based judgment into a deployable, measurable, and billable layer, then e-commerce budgets do not need to go only toward more traffic. Some of the next budget can go toward making the traffic already on the site less likely to leave confused.

Remark’s lesson for AI builders is precise: the valuable unit is not a chat response. The valuable unit is expert judgment at the moment when a buyer is ready enough to ask.

Reference Sources