What happens when a marketplace puts one price on demand and another on everything else.
Publishing two prices changes how sellers value the platform before it changes where they transact.
In late August, some Airbnb hosts in the United States got an email with the subject line "Share your link, get a lower fee." A new tool in their listing editor generates a trackable link. Share it on Instagram or in a newsletter, and any guest who books through it costs the host a much smaller service fee. Hosts report paying 6 percent or 10 percent instead of the 15.5 percent most now pay. Everything else is unchanged. The guest books on Airbnb, pays on Airbnb, and is covered by AirCover.
Platforms have done versions of this for more than twenty years, and the history moves in three steps. First the sellers start thinking about the platform as two products instead of one. Then they compare the second product, everything that is not demand, against what it would cost to buy elsewhere. Only after that does anyone actually move, and only where the answer to that comparison is unflattering. The first step happens almost immediately. The third step happens much less often.
Airbnb has published nothing about the program and describes it to hosts as a pilot it may change or stop, so treat the numbers as reported rather than confirmed. There is also a slider that lets a host pass part of the saving to the guest as a lower nightly rate, which turns out to matter more than it looks.
Valve takes about 30 percent when someone buys a game on Steam, less once a game has sold a lot. But a developer can generate Steam keys, sell them on their own website or through another store, and pay Valve nothing on those sales. Not a reduced rate. Zero, for roughly fifteen years.
If a low fee on seller-found customers pulled business off a platform, this is where it would have happened first. Developers kept the overwhelming majority of their sales on Steam anyway. When Valve reported in March that it paid out 76 percent of revenue to non-Valve games in 2025, it noted the figure does not even include what developers earned selling keys elsewhere.
What a developer would have to replace is not a store. It is the player's account, the library the game lives in, automatic updates, the friends list, the reviews, and a very large group of people who already have the software installed. Nobody sells a substitute for that.
There is a second reason that is easy to miss. Valve expects developers not to give Steam customers a worse deal than key buyers, so the developer keeps the whole saving and cannot turn it into a cheaper price for the player. The zero fee never becomes a reason for buyers to shop somewhere else. Anyone reading Steam as proof that good infrastructure wins on its own is reading it too simply.
OpenTable has priced by source almost since the beginning. A diner who finds a restaurant through OpenTable's own app costs the restaurant around a dollar, sometimes more. A diner who books through the widget on the restaurant's own website costs a quarter, and nothing at all on the higher-priced plans.
That split invited an obvious question. If the restaurant brings the diner, what is it paying for? A reservation book, a floor plan, guest notes, and confirmation emails. Useful software, and software plenty of companies can write.
The question got asked loudly. In 2017 Nick Kokonas, who ran fine dining restaurants in Chicago and then built Tock, argued publicly that many restaurants were not even tagging their own websites correctly and were paying the higher rate on diners who had come to them directly. His team released a tool so restaurants could check. He also made the arithmetic impossible to ignore, pointing out that a busy restaurant could spend four or five thousand dollars a month with OpenTable one dollar at a time, and pricing Tock as a flat monthly fee instead.
Competitors ran that line for a decade. Resy charges per cover on bookings from its own app and nothing on bookings from the restaurant's website, which is the same idea borrowed. SevenRooms built its pitch around direct bookings. Today there are reservation products with a flat fee and no cover charges at all, and their marketing runs the comparison in public.
This is where the story stops being a simple morality tale about fees.
OpenTable did not collapse. Its share of US restaurants slipped from around 51 percent to around 46 percent between 2022 and 2024, and it still led thirteen of the top fifteen American cities. Several well-known New York restaurants that had left came back. It is smaller than it was and it is still the biggest.
What happened to the challengers is more surprising. American Express bought Resy in 2019 and then bought Tock in 2024 for about $400 million. DoorDash bought SevenRooms in 2025 for $1.2 billion. OpenTable answered by partnering with Visa to give premium cardholders access to hard-to-get tables.
The market broke apart and then reassembled around companies that own demand. Restaurants did not stop paying someone to bring them diners. They changed who they paid, and several ended up paying a credit card company instead of a reservation site. The businesses selling just the software and none of the diners mostly ended up inside businesses with diners.
The first thing that changes is how the seller values the platform.
Across every case I looked at, the same thing happened straight away. As soon as a platform published two prices, its sellers began treating the platform as two products.
Faire charges brands 15 percent on a retailer it introduced and nothing on a retailer the brand brought itself. Within a couple of years the standard advice circulating among brands was to use the marketplace for discovery at 15 percent and route every relationship you found yourself through the free path. Some now describe Faire as free wholesale ordering software, which is a remarkable thing for a marketplace to become in its customers' minds.
Etsy gives sellers 4 percent back when a buyer arrives through the seller's own link, taking the effective fee from 6.5 percent to 2.5 percent. Sellers read that 4 percent as Etsy's own price on Etsy's own traffic. Some pocketed it happily. A sellers' cooperative used it as an argument for building a platform with no fees at all.
None of that requires anyone to leave. Once the two prices are separated, the seller starts thinking about them separately, and that is true of the sellers who stay as much as the ones who go.
Publishing two prices causes suppliers to price the two parts separately.
The second step follows quickly, and it is the one platforms should worry about. Having separated demand from everything else, the seller starts asking what everything else would cost to buy somewhere else.
When a court stopped Apple from charging 27 percent on purchases completed on a developer's own website, the calculation developers described was not about principle. It was whether a few points of saving covered the cost of building, hosting and reconciling a checkout. For most small apps it plainly did not, and they stayed.
That is the real difference between Steam and OpenTable, and it is not that game developers are more grateful than restaurant owners. A developer who runs the comparison finds there is no real replacement to buy. A restaurant owner who runs the comparison finds six vendors selling the same thing at a flat rate. Movement follows the answer to that question rather than the size of the discount.
The size of the discount matters less than whether the remaining bundle can be replaced.
There is a cost here that platforms rarely price in. The lower rate is not only a message to your sellers. It is material for your competitors.
In every case, somebody used the platform's own split as sales copy. Reservation software companies advertise against the cover fee by name. Wholesale tools tell brands exactly how to route around the commission. Within days of the Airbnb emails appearing, at least one rival booking platform was advertising zero percent directly against Airbnb's reported 6 to 10 percent.
Once the platform publishes the benchmark, competitors get to use it too.
Source-based pricing creates a problem that did not exist before, which is deciding who gets the credit. That turns out to be a real product rather than a footnote.
Valve used to hand out keys freely and now gives 5,000 per game by default, with anything beyond that reviewed case by case. Faire tightened its definition of a self-brought retailer in 2024, spelling out which links, tools and time windows count. OpenTable shows the other failure mode, where the cheaper rate existed but many restaurants never claimed it because their own websites were not set up to prove where the diner came from.
Newer programs learned. Amazon's credit for sellers who drive outside traffic requires tagged links and counts purchases within fourteen days. Etsy uses a thirty day window and ignores links shared inside Etsy. If you charge differently depending on who found the buyer, deciding who found the buyer is now part of what you sell, and it needs building before launch rather than after people start gaming it.
None of this is certain, and Airbnb's terms are not fully public. But the history points fairly consistently in a few directions.
One feature deserves more attention than it has had. The slider that lets a host pass savings to the guest is precisely the move Valve does not allow. Steam's zero fee never reaches the buyer, so it never teaches anyone to look elsewhere for a better price. Airbnb's version can reach the guest, which makes it a stronger marketing tool for hosts and a riskier instrument for Airbnb.
What would the seller have to pay someone else to replace the part of your platform that remains?
That replacement cost is the real ceiling on the fee once demand has been priced separately.
The fee you can defend is limited by what your sellers would have to pay somebody else to replace the part of you they still need. So it is worth knowing, before you publish anything, which piece of your platform is genuinely hard to buy elsewhere. If the honest answer is that everything except your customers is available from four vendors at a flat monthly rate, a source-based discount will start a market for the rest of you rather than protecting you from one.
It is also worth being deliberate about who gets the discount. One rate for everyone hands most of the money to sellers who were never going to leave, and hands a clear benchmark to the few who could. And the version of this idea that has worked best is not defensive at all. Amazon pays sellers roughly 10 percent back on sales they bring in from outside, using the same mechanism to buy new demand rather than to protect demand it already had.
If you separate the price of finding the customer from the price of everything else, your sellers will start doing the same.
The Airbnb terms described here come from host messages reported in the trade press. Airbnb has not published the program, and the fee levels should be read as reported rather than confirmed.