- 09 Dec, 2025 *
We’re seeing more and more pitches for companies that are ‘unbundling’ the marketplaces of old.
The thesis roughly goes like this.
In the pre-AI era, marketplaces had a classic two-way network effect. Being the place for the best supply meant you were the place for the demand, and vice versa. Buyers and sellers ‘met’ on your site, and value was created and therefore extracted by the marketplace platforms. It led to massive monopolies, particularly in the traditional ‘classifieds’ businesses - Rightmove (property) and Autotrader (cars) being two prime examples in a UK context.
Marketplaces had a role because a centralised place to aggregate inventory was a crucial part of the online purchase jour…
- 09 Dec, 2025 *
We’re seeing more and more pitches for companies that are ‘unbundling’ the marketplaces of old.
The thesis roughly goes like this.
In the pre-AI era, marketplaces had a classic two-way network effect. Being the place for the best supply meant you were the place for the demand, and vice versa. Buyers and sellers ‘met’ on your site, and value was created and therefore extracted by the marketplace platforms. It led to massive monopolies, particularly in the traditional ‘classifieds’ businesses - Rightmove (property) and Autotrader (cars) being two prime examples in a UK context.
Marketplaces had a role because a centralised place to aggregate inventory was a crucial part of the online purchase journey. Google Search was primarily an index of places to find your potential answer(s) (facilitated by the famous 10 blue links) but it didn’t give you the answer. That bit still relied on marketplace platforms.
Put simply, the user flow went: Search => Marketplace => Specific Search => Transact.
In the post-AI era, things change.
Rather than being a directory to point you towards the place your answer might be, LLM-based Search wants to be the answer. The aim is to bundle the end-to-end consumer journey into the same UI. Google Shopping was in some ways a first step in that direction, but worked only for a relatively tight set of product queries ("black t-shirt"), and still offered a list of options rather than a concrete recommendation.
If the LLMs are doing the aggregation of the supply side and bundling it tightly into the Search response, the role of the marketplace changes.
No longer does it serve as a way finder, helping the user navigate the internet to find potentially relevant answers within a dedicated platform housing all/most available options. In other words, it no longer channels demand. Instead, at its core, it’s an aggregator of digitised data about the available inventory. The data decouples from the network effect. If the same data is available elsewhere on the internet (a competing marketplace, direct from a seller, etc.) - particularly if in a ‘better format’, e.g. higher context, lower price - the LLMs can simply re-direct the demand to the same piece of inventory elsewhere.
The paradigm moves to something along the lines of:
~Search => Marketplace => Specific Search => Transact.
That creates a conundrum for dominant marketplaces today, many of whom exist in monopoly or oligopoly structures that look highly susceptible to disruption outside-in. The appropriate response strategy depends on their perspectives on the ‘end game’ in agentic commerce.
Option 1 is to do (close to) nothing. Focus on having the most dominant internet space for a given set of inventory, and push to be the LLMs source of truth when it answers a user’s query (also known as Agentic Engine Optimisation, and other related terms). Even though the marketplace is a step further down the food chain than it used to be, the supply side still benefits from the visibility it gets in LLM results as a proxy of being listed on the marketplace - in other words, the marketplace is still an indirect channel of demand - and therefore the network effects (and associated economic rent extraction) may still hold, at least partly.
Option 2 is to push to maintain a role as a standalone branded destination, and effectively compete with the generalist AI models with a highly verticalised end-to-end consumer experience. That probably requires a much better on-site search experience comparable with the assisted-sale UX paradigm that AI offers, a much richer feature set going beyond ‘just’ search and discovery, and a lot of marketing budget to continue pulling people towards your site in the face of increasing adoption of the major branded AI platforms (ChatGPT, Gemini, Claude, etc.).
Option 3 is somewhere in the middle: building connectors, apps, MCP servers, etc. to offer the best set of ‘tools’ for the large scale AI platforms to answer a specific query. This gets closer to a white-labelling strategy, or akin to the Shopify app ecosystem. It "skates towards where the puck is going" (h/t Wayne Gretsky), but it’s arguably a less attractive business model. The network effects are diminished given the marketplace is no longer the demand channel. There’s platform risk that the generalist AI players switch to a better/shinier tool, or build one themselves (á la Amazon Private Label). And it requires much deeper embedding into the supply side to ensure that ‘listings’ - which commoditise quickly with the exponential growth of crawlers - is not the ‘only’ asset you sit on.
As per my general principles, the truth is probably somewhere in the middle - or some combination of the above combined with the other things I haven’t thought of. The clarity on the exact direction things will take is lacking, but the overarching message is clear: marketplaces won’t be what they used to be going forwards.