Unbundling the Marketplace Stack

In a previous essay, I wrote about how the ‘light’ marketplace business model is structurally undermined. I took a first-principles, high level approach. Here, I try to map out the specifics: what the new paradigm looks like and where the gaps in the thesis are.

The table is the overview. Each layer the marketplace currently owns will have a better, cheaper alternative. This will allow innovation across each layer – better for the industry, buyers, and sellers. Below that, I go more in-depth about what each layer may look like. 

Summary

Customer acquisition and relationship

Marketplaces position customer acquisition as a service. The reality is more nuanced. Sellers use marketplaces because they have to, as that is where aggregated demand is.

Many of the established marketplaces spend around 15-25% of gross bookings on sales and marketing. Some of that is on supply, but a majority of that is probably on demand generation. Because spend is aggregated, marketplaces can run large campaigns that sellers may not be able to run themselves. But the KPI is always improving the marketplace brand awareness and driving bookings growth – rather than bookings for any individual seller. Sellers give up a meaningful chunk of revenue so the marketplace can run marketing campaigns for its own brand.

Sellers will shift away from marketplace customer acquisition, and pay more for outcomes and visibility for their business. We will see much more direct marketing, and services/tools that focus purely on customer acquisition for sellers. Because of this, we will see customer relationships shift towards direct channels, without marketplace lock-in.

Discovery and profile

You can argue that these marketplaces will become AI-concierge for their respective industries (e.g. Booking.com as the travel concierge, Thumbtack as the household work concierge). 

But the reality is… the frontier models are better positioned to be the concierge.It will understand you as a person — your preferences, schedule, and situation — in a way a vertical marketplace never can.

Whatever UX the marketplaces create, the frontier models can do, on-demand. The only thing missing is the supply base. But supply will follow wherever visibility and revenue opportunity is highest, which eventually leads to frontier models. 

Sellers will happily embrace AI discovery because it is cheaper discovery, and discovery directly to them. We are already seeing ChatGPT and Gemini work on agentic commerce protocols. The end game is all products and services, without the marketplace. Profiles will follow this pattern; instead of your listing belong to a marketplace, it will belong to a seller.

Scheduling, payments, and compliance

These layers will be owned by payment processors and scheduling platforms via API, without a marketplace intermediary. There are specific scheduling platforms. But one of the pros of the marketplace is getting paid bookings; rather than inquiries or spam like Craigslist. So it seems like the ones that will dominate are the payment processors that deal with what is core: compliance. Companies like Stripe can move up and down the value chain better than Calendly can. We will see their APIs dominate, as they continue refining their product for a world of agentic commerce.

Additionally, we will see payment processors gain adoption amongst individuals and bypass marketplaces. From payments, they can expand into accounting automation, tax compliance, and customer management.

Sellers will try to take control of their bookings, preferring for scheduling and payments to route directly to their processor rather than through an intermediary. 

Trust and escrow

This is where the most obvious gaps (and maybe opportunities) exist. Marketplaces currently have built-in trust and escrow via reviews and refunds. 

Reviews are valuable but depreciating. Frontier models were trained on review data from across the open web. Marketplaces that wall off their reviews to protect their data moat will find their SEO suffer. Furthermore, as discovery and customer acquisition diversify, there will be better ways to assess trust. Social media, for example, provides much richer data. As reviews become something that AIs summarize instead of humans reading one-by-one (we see Booking.com already doing this with review summaries), the most real-time, longitudinal data will win. Reviews will look more like a credit score; reputation that is transferable and comprehensive.

If payments become individualized, refunds will not be handled by the marketplace anymore. In most cases, the payment processor handles it — chargebacks, disputes, refunds are already core to what Stripe does. Additionally, Stripe can add a resolution center where disputes can be handled directly first, then upgraded as a ticket if it needs further review. But in some cases, you may want a real guarantee – whether that is making sure the house, freelancer, or other service is as advertised. Similar to reviews, we will see a company that treats this as core competency, and tries to be the escrow layer for various services. 

These changes require other dominoes to fall first, so nothing seems to have been built yet. But we will see:

• A company that specializes in the ‘trust graph’ for services, by using AI to provide the most accurate, detailed, and verifiable source of trust. They will understand how to pull and score data for different services faster and more accurately than others. Frontier models and other platforms will pay via API to provide customers with upfront, clear, third-party trust. 

• A company that specializes in the ‘escrow layer’ for services, providing buyers with an optional, transaction-level guarantee. For a small fee (e.g. 2-4% of transaction), a buyer gets a refund if the service fails to deliver. The seller opts in to offer it; the buyer opts in to purchase it. Neither party is forced to pay a blended marketplace take rate. A more honest model — proportional to actual risk, priced transparently, and owned by a company whose entire incentive is to resolve disputes fairly rather than protect platform reputation.

Conclusion

Counter-arguments are real but transitional. Marketplaces have distribution and brand awareness. But frontier models have all the distribution, and demand for many of these industries are already growing organically. Marketplaces have also built up trust, and consumer behavior could be difficult to change. But the trust is mostly perception. Airbnb doesn’t check the homes in rural Japan to see if they are real listings, nor do they deal with compliance country-by-country. The UI and brand is important, but doesn’t justify the take rate they currently charge. 

It seems more likely that there will be a better model of connecting buyers and selling, which means killing the margins that these companies have successfully built themselves on. It seems that we will see the marketplace stack unbundled into core pieces, orchestrated and coordinated through AI rather than a marketplace.