English NewsSunday, 15 November 2026
Latest
Business

Why AI is not producing results for most sellers, and what actually works

2026-04-02

Why AI is not producing results for most sellers, and what actually works
Photo: (CC BY 3.0)

Plenty of sellers have put real hours into AI. They have tried the chatbots and image tools, wired up automations, bought prompt packs, and used AI to draft emails, spin up ads, and analyze competitors. Yet many cannot point to one clear, measurable business result: revenue has not jumped, margins have not improved, and they are not getting meaningful time back. These are not beginners. So what is going wrong?

The pattern looks like this: a new tool drops, you test it, it produces output, and then it is unclear whether it moved the needle, so you move on to the next thing. Over time the business ends up with a patchwork - an automation here, an agent there, a project for proposals, a scenario for content. Lots of pieces, none connected, none tied to what actually drives the business forward. One agency owner doing almost $900K a year had AI touchpoints across his whole operation - five platforms and 40 to 60 hours invested per month - and when asked what was working, he could not draw a line from any of it to a real result.

The tools are not the problem; they largely do what they promise. The missing piece is strategy: no architecture and no system behind the individual experiments. Many sellers approach AI the way they approached Facebook ads a decade ago - find a tactic, try it, see what happens. That works for a single channel. AI is not a channel; it is infrastructure that touches everything. Without a plan, you end up with disconnected experiments scattered across the operation.

Meanwhile, a smaller group of businesses is pulling ahead, and the gap compounds daily. Top e-commerce brands run full agent loops: one agent scrapes competitor ads daily, another analyzes which hooks are gaining traction, a production agent generates new variations in the brand's voice, and an analysis agent watches performance in real time. These systems test 50 ad variations while competitors test two a week. Winning agencies have AI handling 60 to 70 percent of delivery - one cut client hours from 50 to 30 and raised prices because quality and speed both went up. Software companies have agents resolving 70 to 80 percent of tier-one support, onboarding new users automatically, and flagging churn risks before customers consider leaving.

Timing matters because the gap does not stay flat. A human team improves linearly at best; a well-built agent system improves compounding, with every cycle feeding the next. A competitor who started six months ago is not six months of work ahead - they are six months of compounding improvement ahead. Businesses that build real AI systems in the next 12 to 24 months will be hard to catch for years after; you cannot hustle your way out of a compounding gap.

The encouraging part: most businesses have not figured this out either, so the window is still open. The fix is to stop hunting for the right tool and start building the right system that tools plug into. "Using AI better" is not a goal - getting 10 to 15 hours back per week, scaling without doubling headcount, and improving margins 20 to 30 percent are goals, and AI can drive each of them when set up correctly.

Separately, a practical advertising tip making the rounds: one audited account had the word "repair" scattered across 23 different search terms, none with a single sale. Each term had only 8 or 9 clicks, so none looked worth blocking - but combined they added up to over 200 clicks, zero sales, and $847 of pure waste. One negative phrase keyword fixed all 23 at once. The lesson is to stop reading search term reports line by line and start looking for repeated word patterns (sometimes called N-gram analysis): break losing search terms into one-, two-, and three-word chunks and see what keeps appearing. The average unoptimized account is estimated to waste 28 to 40 percent of ad spend on searches that will never convert.

A useful mental model for Amazon discovery right now is two flywheels sharing one center of gravity. The first is traditional keyword search: a shopper types a query, the A9 algorithm matches keywords, the sale boosts keyword ranking, and the wheel spins faster. The second is conversational AI search through Rufus: a shopper asks something like "what is the best water bottle for hiking that keeps drinks cold all day?" and Rufus reads your listing, reviews, Q&A, and brand story to decide whether you are semantically relevant. A sale from either engine feeds both. The practical takeaway: listings still need strong keyword architecture, but they also need natural, conversational language, and reviews and Q&A now directly influence AI-driven discovery. The old playbook is not dead - it is just half the picture.

One cautionary tale circulating in the seller community: a seller we'll call a long-time pet products merchant sold sturdy, well-reviewed dog leashes for years and had just crossed seven figures when a customer emailed to say a leash clip had snapped and whipped back into their eye. The seller refunded the order and assumed that was the end of it. Days later a lawyer's letter arrived: the injured customer was a surgeon claiming lost income, medical costs, and long-term career impact. Amazon then took the listing down pending review and delayed payouts while it asked for documentation. The seller's lawyer's first question was simple: "Do you have product liability insurance?" He did not - he had always assumed a leash was too harmless to need it. Even though the business survived, the legal fees, frozen listing, and cash-flow pressure cost far more than coverage ever would have. The takeaway: you do not buy insurance because you expect something bad to happen; you buy it so one incident cannot become a business-ending event. In e-commerce you are not just selling products - you are carrying the risk that comes with them.

One more stat worth sitting with: Amazon had 2.4 million active sellers in 2021. By the end of 2025 that figure had fallen to 1.65 million - a smaller field, but a more competitive one.

More Amazon business news