For the platform seller watching TikTok and AI shopping agents arrive at once, wondering whether fourteen years of Amazon actually prepared them for either.
The short answer. Amazon taught me demand capture, not demand creation. Fourteen years of Amazon selling mean I can tell you a SKU's real margin, its stock cover and its conversion rate before breakfast, but they never taught me how to make someone want a product they did not already want. TikTok does exactly that, and the AI shopping agents arriving now want the same discipline Amazon already forced on me, which means the old skill is not obsolete so much as it is half of what is coming.
Fourteen years selling on Amazon, and only recently have I started asking what it actually taught me, not about Amazon itself, not about keywords or advertising or how a listing gets built, but about the kind of operator it turned me into. Platform selling, above everything else, taught me to be efficient.
The genius of a marketplace like Amazon or eBay is simple. A seller does not have to build an audience first, because the audience is already there. Economists call this a two sided platform, one that brings together two groups, buyers and sellers, whose value to each other grows as the other side grows. More products make the platform more interesting to buyers, and more buyers make it more interesting to sellers, which is the network effect, and it is why a big platform only gets harder to compete with as it grows.
Once a platform is big enough, though, the whole arrangement changes shape. Third party sellers now generate 69% of Amazon's total GMV, up from 60% in 2019, and Amazon and its sellers together moved roughly $830 billion of goods in 2025. That is an enormous demand machine, and Amazon owns every bit of it.
On Amazon, not one euro goes toward making someone want a frying pan, because someone has already typed "12 inch induction frying pan" into the search box before you ever see them. The work starts at who wins that customer: whether the product is right, whether the price is right, whether the image beats the next one, whether the listing actually converts, whether the item is in stock and will arrive tomorrow, whether the rating sits at 4.6 or 4.2, what the click costs, and what is left once every cost has been paid. It makes a platform seller a little neurotic as an operator, and I mean that almost as a compliment.
Customers grow, but so do competitors, fulfilment costs and advertising prices, so revenue alone will never carry you, and every year you have to be a little better than you were the year before. In Marketplace Pulse's 2026 Seller Index, 49% of Amazon sellers named platform fees one of their biggest margin pressures, 46% named advertising, and 47% reported that their margin had fallen year on year.
What surprised me more than the fee numbers themselves is what sellers actually do about it. Among those most frustrated by fees, only 24% are actively reducing their Amazon dependence, while 42% are growing their Amazon revenue share instead, which means people complain about the toll and then pay it faster, because the demand it buys is hard to replace anywhere else. Call it commerce's own Red Queen effect: you have to keep running just to stay where you are.
And it teaches you real things: SKU profitability, stock turns, pricing, conversion, cash, logistics, forecasting and advertising incrementality, in a way a business that can hide weak efficiency under growth for years never has to learn, because the platform does not care about your explanations. If stock runs out, your ranking drops. If the price is wrong, the customer picks someone else. If advertising does not convert, the money is simply gone, and if you bought too much stock, that cash sits uselessly in a warehouse. The feedback loop is fast, and it is genuinely merciless.
When a platform brings you the customer, it also sits between you and that customer permanently. It sets the rules, it sets the search results, and it decides how much organic visibility survives around the advertising, and as your dependence on it grows, it can quietly change how the economics split between you.
This is not incidental to platform economics. It is close to the point, because a platform is not a neutral marketplace so much as an active designer of the market, and current research even looks at this through how much a platform controls a customer's consideration set, meaning which options a customer even gets to see.
Amazon is an extreme example of it. Marketplace Pulse estimated back in 2023 that a typical private label seller's combined referral, fulfilment, storage and advertising costs could reach roughly half of revenue: fifteen percent in referral fees, twenty to thirty five percent in fulfilment, and up to another fifteen percent in advertising. The platform brings you the market, and in return it can take a growing share of the value that market creates.
Amazon is fantastic at teaching demand capture, but it does not teach demand creation at all. On Amazon, the customer usually already has intent, whereas on TikTok they often had none to begin with. Nobody opens TikTok because they need a new lipstick, a blender or a pair of shoes. Someone watches a video, an interesting person shows the product, the product becomes desirable, and only then does interest turn into want and, eventually, into a purchase.
TikTok calls this discovery commerce, and the scale behind it stopped being marginal a while ago. More than 100,000 European businesses had joined TikTok Shop by June 2026, across France, Germany, Italy, Spain and Ireland, and TikTok itself reports triple digit daily GMV growth between August 2025 and February 2026. In the UK specifically, TikTok Shop now counts more than 300,000 small business sellers, and LIVE Shopping sales grew 55% year on year.
That is a completely different muscle from the one Amazon builds. An Amazon seller optimises demand that already exists, while a social seller has to build interest, culture, community, entertainment and want from nothing. And here, at least in my own case, there is a genuine skills gap: we got very good at serving demand, and we never had to get equally good at creating it.
Just as ecommerce looks like it is drifting toward TikTok, a second shift is arriving from a completely different direction, as AI starts shopping alongside people and, increasingly, partly on their behalf. McKinsey estimates AI agents could be orchestrating between $3 trillion and $5 trillion of global consumer commerce by 2030, a trillion of that in the US alone, and in Europe 38% of consumers already say they use generative AI to research products or decide what to buy, with 63% using it to compare brands, prices and reviews, according to McKinsey's own survey of French, German and British consumers from December 2025.
Google launched its Universal Commerce Protocol in January 2026, an open standard that lets AI agents pull product data, pricing and inventory and complete checkout without the customer ever leaving the chat. Shopify has already made merchant catalogues available to ChatGPT, Microsoft Copilot and Gemini by default, live for millions of merchants, and in the first quarter of 2026 alone AI driven traffic to Shopify stores grew eightfold year on year, while orders coming from AI powered search grew roughly thirteenfold.
In this world, the interface can be a single sentence: "I need a good induction frying pan, there are five of us, no teflon, and my budget is under 80 euros." The agent then does the job a person used to do scrolling through Amazon's search results, and something genuinely interesting happens right there.
An AI agent does not fall for a beautiful website. It needs information: what the product is, what it is for, what it costs, whether it is in stock, when it will arrive, what other customers say about it, whether there are returns, and whether the seller can be trusted. OpenAI already says, in its own words, that it ranks merchants in ChatGPT shopping by availability, price, quality, and whether they are the maker or primary seller of the item, which should sound extremely familiar to anyone who has spent years on Amazon.
It is Amazon, roughly, to the power of ten: structured product data, availability, price competitiveness, reviews, delivery performance, conversion, returns, trust and unit economics. Amazon sellers have spent years training for a world where the buyer never browses a whole page, because an algorithm decides which handful of products even make it into consideration in the first place. The gatekeeper is simply changing hands, as Amazon's algorithm gives way, possibly, to the customer's own.
| What | Figure | As of |
|---|---|---|
| Amazon third party share of GMV | 69%, up from 60% in 2019 | 2025 |
| Amazon and sellers, total GMV | $830 billion | 2025 |
| Sellers naming Amazon fees a top margin pressure | 49% | Marketplace Pulse 2026 Seller Index |
| Fee-frustrated sellers growing their Amazon share anyway | 42% | same survey |
| European businesses on TikTok Shop | 100,000+ | June 2026 |
| UK small businesses on TikTok Shop, LIVE Shopping growth | 300,000 sellers, up 55% year on year | 2026 |
| Global agentic commerce, McKinsey forecast | $3 trillion to $5 trillion | by 2030 |
| European consumers using generative AI to research or decide purchases | 38% | McKinsey, Dec 2025 |
| AI driven traffic growth to Shopify stores | 8x year on year | Q1 2026 |
This is, I think, the most interesting split of the next era of ecommerce. On one side sits social commerce, whose job is creating want, and on the other sits agentic commerce, whose job is helping someone choose. TikTok is asking whether it can make you want something, while an AI agent is asking whether that thing is actually the best product for you, and a serious business now needs a good answer to both questions.
Which is also why I do not think the future of ecommerce is your own site against the marketplaces. Distribution is fragmenting far beyond that binary: TikTok can create the want, ChatGPT can do the research, Google or Gemini can compare the options, Amazon can still handle the transaction and the fulfilment, and a manufacturer's own systems can finally deliver the product. The customer's journey no longer belongs to one company.
Three things, and none of them need new software to start.
One. Get your product data machine readable now, wherever you sell, with full attributes, GTIN codes, and accurate stock and pricing behind structured markup, because agents cannot buy what they cannot parse, and this is genuinely cheap to fix compared with everything else on this list.
Two. Run one demand creation experiment with actual humans watching, whether that is a short video or a live session, as long as the point is making people want something rather than converting them on the spot. It will feel uncomfortable if platform selling is all you have ever done, so do it anyway, and start small.
Three. Do not drop the operating discipline Amazon built into you: availability, price competitiveness, reviews, delivery and returns are now the same inputs both a human and a machine buyer will use to choose you, and building that discipline further is itself the kind of shovel work AI is good at helping with.
No. Amazon still owns an enormous share of existing demand, 69% of it through third party sellers on roughly $830 billion of GMV in 2025, so the real question is not Amazon or not Amazon but whether Amazon is the only channel you actually know how to run.
If demand creation is not yet a skill in your business, it is worth learning somewhere, and TikTok is currently the cheapest place to learn it: over 300,000 UK small businesses are already selling there, and UK LIVE Shopping sales grew 55% year on year, so doing it small and deliberately beats skipping it altogether.
Not replace, more likely sit alongside it for years to come, but the ranking factors OpenAI has published for ChatGPT shopping, availability, price, quality, and whether a seller is the maker or primary seller, are close cousins of what already decides an Amazon ranking, so the muscle transfers more directly than most people expect.
Product data. It needs to be clean, complete, structured and accurate on stock and price, because every agentic commerce protocol launched so far, both Google's Universal Commerce Protocol and OpenAI's Agentic Commerce Protocol, depends on that before anything else can work.
More than before, not less. An agent optimising for availability, price and quality is rewarding exactly the discipline platform selling forces on an operator, and the skill nobody trained for is demand creation, which is exactly the gap worth closing.
Anyway. Fourteen years in, and I still look at a business mostly through SKU margin, stock cover, cash, conversion and operational efficiency, and I did not become disciplined because I set out to be. The environment made me that way, and when a platform owns the demand and slowly raises the toll, the only way to stay profitable is to keep getting better, which turned out to be useful training for whatever comes next.
One new skill still has to be learned, though. We already know how to take existing demand and win as much of it as we can, and now we have to learn to create it too.
The next generation of commerce companies will need to be good at both: creating demand for humans and being selected by machines.
Amazon 2025 GMV and third party share (69%, up from 60% in 2019, $830bn total): Marketplace Pulse, "Amazon GMV Surpassed $800 Billion in 2025". Amazon seller margin pressure (49% fees, 46% advertising, 47% margin decline, 42% of fee-frustrated sellers growing Amazon share versus 24% reducing dependence): Marketplace Pulse 2026 Seller Index, cited in "The Paradoxical Dependence of Amazon and Its Sellers", published 16 April 2026. Typical private label seller cost share (approximately 50% of revenue: 15% referral, 20 to 35% fulfilment, up to 15% advertising): Marketplace Pulse, "Amazon Takes a 50% Cut of Sellers' Revenue", 13 February 2023. TikTok Shop Europe (100,000+ businesses across France, Germany, Italy, Spain and Ireland by June 2026, triple digit daily GMV growth August 2025 to February 2026, TikTok's own reporting): TikTok Newsroom and Marketplace Pulse coverage. TikTok Shop UK (300,000 small business sellers, LIVE Shopping up 55% year on year): TikTok UK reporting, July 2026. McKinsey agentic commerce forecast ($3 trillion to $5 trillion globally, $1 trillion in the US, by 2030): McKinsey, "The agentic commerce opportunity", October 2025. European consumer AI usage (38% using generative AI to research or decide purchases, 63% to compare brands, prices and reviews; survey of French, German and British consumers, December 2025, n=749): McKinsey, "Europe's agentic commerce moment". Google Universal Commerce Protocol (launched January 2026): Google, product documentation. OpenAI ChatGPT shopping ranking factors (availability, price, quality, maker or primary seller status): OpenAI Help Center, "Shopping with ChatGPT Search". Shopify AI channel adoption (catalogues live for millions of merchants in ChatGPT, Copilot and Gemini; AI driven traffic up 8x and AI search orders up roughly 13x year on year, Q1 2026): Shopify, "How agentic commerce works", 18 June 2026. All retrieved 1 September 2026.