Mina Elias, founder of Trivium Group, helps ecommerce sellers scale using Amazon pay-per-click ads. He has added 50 to 60 “AI employees” alongside 80 to 90 human employees, according to an interview for the 7 Figure Seller Summit, an online conference.

As Elias demonstrated, it costs him about $1,000 a month to run 42 recurring agents, with additional spending for technology development. These AI employees monitor listings, competitors, search volume and seasonality, account health, advertising, business waste, and keyword opportunities.

Elias is one of many e-commerce founders converging for the 7 Figure Seller Summit , a free online event running through Thursday, starting at 10 am EST each day. For the event, Gary Huang, founder of the annual event and owner of 80-20 Sourcing in Shanghai, is interviewing seven- and eight-figure e-commerce brand owners and operators to show exactly what types of AI workflows sellers are using right now to build thriving businesses—and to understand how they are mitigating risks such as AI “hallucinations.” Huang is also chair of the Supply Chain Committee for the American Chamber of Commerce in Shanghai.

Sellers have faced many challenges in recent years, as U.S. trade policy has changed. “The biggest negative trend affecting profitability over the past couple of years has been tariffs,” Huang said in an email interview. “The on-again, off-again, and back-on-again pendulum—and the uncertainty around it—has hurt confidence. I am hearing from many sellers that decision-making has been frozen across the board.”

However, AI seems to be giving sellers “an almost unfair advantage when it is implemented properly in narrow, supervised workflows,” he added. “Agentic technology has made a giant leap in what it can actually do in the past six months. Last year, AI was still in a hype bubble and was not there yet for many operating tasks. This year, we are definitely seeing the use cases.”

U.S. e-commerce sales exceeded $1 trillion in 2025 alone, according to an analysis of U.S. Department of Commerce data by Digital Commerce 360. That is triple what the figure was in 2015.

Here are some of Huang’s observations from the front lines about how AI is transforming ecommerce.

Using AI to build more efficient businesses. Huang informally surveyed 76 sellers in his teaching cohort and webinar audience. Among them, 34 said they were testing AI tools, 25 had some AI tools “partially working,” six were already running AI workflows, and 11 were not using AI yet.

Among those using AI, many e-commerce entrepreneurs are turning to tools such as Claude, Claude Cowork, Codex, and OpenClaw that allow them to use AI agents, under human guidance. Often, the results are as good or better than if a human performed the work, though not always, Huang notes.

However, merchants faced challenges, such as an overabundance of tools to choose from, confusion, and lack of knowledge of how to structure an AI workflow. Many told him they needed help with costly pay-per-click advertising. They are also looking for advice on optimizing listings and content, product research, operations and reporting.

Better listings: When it comes to improving revenue, said Huang, “Listing optimization is one of the biggest levers because it can improve click-through rate and conversion rate.”

AI can now gather merchant’s, reviews, competitor listings, keywords, and existing images, and then rapidly build a research bank and generate new copy and image directions, shrinking a process that might have taken two days to 10 minutes, Huang noted. Although hallucinations still happen, “a properly grounded agent can work from the seller’s actual product context and keep the text and images much more consistent than a generic chatbot,” he said.

Smarter pay-per-click advertising. Sellers are using agents to process search-term and campaign data at scale, identify wasted capital, find opportunities worth investing in, and prepare weekly action reports, he says. Huang has found that operators were most effective if they first selected the goal, such as profitability, ranking, or launch date. “The agent then prepared a negative-keyword, bid, and budget recommendations for a human to review before anything changed in the account,” he said. One vendor involved in the conference, a Japanese toy and electronics seller, reported increasing sales by 15x over six months, while ad spend rose by 15 to 20% in using this approach, he said.

Optimizing inventory. Poor forecasting can create steep financial losses for sellers. “Order too little and you go out of stock during peak season; order too much and storage fees and cash-flow pressure become killers,” Huang said.

To counter this, vendors are using a variety of tools, Huang said. One is Inventory Hero, which flags upcoming stockouts, suppressed listings, and dead listings; GoFlow, which connects marketplace and inventory data across multiple sales channels; and Nile, an agent-to-agent (A2A) commerce tool that creates an AI brand agent on the seller’s behalf that can chat directly with a customer’s agent on ChatGPT or another LLM.

“It acts like an AI salesperson,” Huang said of Nile. “When someone asks an AI engine for a product recommendation, the brand agent can provide the full catalog context: the product, photos, pricing, stock, benefits, proof, reviews, caveats, and why it matches that particular buyer. That makes the AI engine more likely to recommend the product and put it in a favorable light because it has richer, current information instead of generic information scraped from a page. This is a new approach to AEO, or answer-engine optimization. The goal is not only to rank in an AI answer, but also to have an agent that can represent and sell the product.”

Using AI with caution. AI often makes mistakes, and sellers recognize this, says Huang. “ The biggest risk is trusting the agent completely,” he said. “At this point, no successful seller I know is trusting an agent 100% to do everything. It does not replace human judgment in critical decisions, such as making the final call before placing a $100,000 purchase order or making an irreversible change to an advertising account.”

In one of the cohorts Huang teaches, a student followed an AI recommendation that deleted keywords from a winning campaign. “He told us it effectively wiped out about eight years of sales history,” Huang said. “Two months later, sales still had not returned to where they were.”

To prevent this, Huang recommends building guardrails into the process. “Let the agent monitor, analyze, draft, and surface options,” he said. “Let the operator approve changes involving money, customer communication, ad-account history, security, and brand reputation.” While AI can save time, it can also waste it, if it introduces mistakes into important e-commerce processes.