← Blog

When a Rival Platform Enters, One Big Sale Will Not Save You

The complementors most likely to defect are the ones your broad loyalty programs never reach. Here is how to spend your retention budget across food delivery, e-commerce, and home-sharing.

Based on the research ofJohannes Loh and Ambre Elsas-Nicolle, "Platform competition and strategic trade-offs for complementors: Heterogeneous reactions to the entry of a new platform," Strategic Management Journal 47(6), 2026

Platform Governance · Complementor strategy · August 2026

When Epic Games opened a store against Steam in 2018, Johannes Loh and Ambre Elsas-Nicolle tracked which developers left and which stayed. Their study, published this year in the Strategic Management Journal, found complementors do not defect as a bloc. The developers who multihomed to the entrant were resource-weak independents with the least to gain from Steam's large shared audience and the most to gain from escaping its crowded competition. The studios that stayed were the ones whose games depended on network effects, the multiplayer titles that could not afford to split their player base. The twist that matters for every platform: those flight-risk independents became less responsive to the incumbent's platform-wide sales, not more. The broad loyalty lever reached the people who were never going to go.

The rule underneath the finding

Strip out the gaming specifics and a general law remains. When a rival opens, the complementors most likely to defect are the ones with the fewest resources and the least dependence on your network, and they are precisely the segment your everyone-is-invited programs fail to move. Two questions place any seller, host, driver, or restaurant on the map. How much do they depend on the demand only you can supply, and how badly is your own internal competition hurting them. The answers name your flight risk, and one question follows for the retention budget: are you spending it on the people who will leave, or the people who will stay regardless (see the exhibit)?

Food delivery: the independent defects, the chain rides your loyalty program

The economics force the split. DoorDash charges restaurants marketplace commissions of 15%, 25%, and 30%, with the higher tiers bundling DashPass, and Uber Eats runs 20% to 30%, adding a surcharge on orders from its Uber One loyalty members. For a single-location restaurant running a 10% to 15% net margin, a 30% cut is not a fee, it is the difference between open and closed, which makes that operator the textbook flight risk. Locally owned delivery co-ops have gone straight for them: 937 Delivers in Dayton charges a flat structure that works out to roughly 8% to 14% of sales, the Lexington co-op returns effectively all of the order value to the restaurant, and franchised co-ops land near 17%, against the majors' 30%. Meanwhile the citywide free-delivery week, the broad orchestration move, lands on the chains and heavy users who lean on your order density and were staying anyway.

So do not answer a co-op with a marketing blast. Three targeted moves actually hold the independent. First, offer segmented commission relief in any market where a co-op has launched: quietly match or beat the Basic 15% tier for the independents at risk, funded by the lifetime value of not losing them, and accept the tradeoff that chains may later demand parity. Second, if you cannot cut the take rate, spend on visibility instead of cash: guaranteed placement, a local-favorites carousel, waived menu and photography fees, which cost less and lean on the one thing a co-op cannot match, your demand. Third, remove the reason to leave entirely by bundling commission-free direct ordering, the way DoorDash Storefront gives a restaurant its own ordering site: you cannibalize a slice of marketplace orders, but you keep the operator on your logistics and off the co-op.

E-commerce: small sellers list everywhere, brands anchor to Prime

Amazon shows the same split at scale. Its all-in take can exceed half of a third-party seller's revenue once the roughly 15% referral fee, fulfillment, and advertising stack, third-party sellers are about 60% of units sold, and around 61% of Amazon sellers already list on at least one other platform, with Walmart and Shopify each near 36%. The lower-fee rivals are explicit: Temu's seller take is reported in the low single digits, and Walmart Marketplace, at 6% to 15%, crossed 200,000 sellers in 2025 on its fastest-ever growth. The tell is Amazon's own defensive move. In January 2024 it cut referral fees on apparel priced under 20 dollars, a surgical response to Temu and Shein, not a broad Prime Day, and under-20-dollar clothing selection then rose 27%.

The lesson is to read Prime Day for what it is: a rally for the brands anchored to Prime traffic and fulfillment, who were not leaving. Spend your actual retention budget on the contested segment. The first option is Amazon's own: category-targeted fee cuts on exactly the price bands a rival is undercutting, precise and reversible, at the cost of margin in that band. The second is to raise the operational cost of multihoming, bundling fulfillment credits or low-cost logistics so a seller's best inventory stays in your network, though this does nothing for sellers who already resent the dependence. The third is advertising and placement credits that offset the pay-to-play crowding pushing small sellers out, cheaper than fee cuts but weak medicine if the category is genuinely saturated. Match the lever to why the seller is leaving, price or visibility or lock-in, rather than firing all three at everyone.

Home-sharing: commodity listings cross-list, unique stays stay

Short-term rentals split along the same line. Vrbo's all-in host cost sits near 8%, undercutting Airbnb's roughly 15.5% host fee and Booking's 15%, yet AirDNA's data shows the defection is not uniform. In New York, 69% of listings are Airbnb-only and only 23% appear on both platforms, and small hosts with one to four listings draw more than half their bookings from Airbnb alone. The hosts who cross-list are the professional managers running standardized, interchangeable portfolios; the unique, hard-to-replace listing tied to Airbnb's demand network and its Superhost status stays put.

That tells you exactly where a blanket host bonus is wasted, on the loyal single-homers, and where to aim instead. One option is targeted fee relief for the commodity, price-sensitive segment in markets where Vrbo and Booking are active, accepting margin pressure and the risk that Superhosts ask for the same. A second is to lean on status rather than cash: Superhost requirements, a 4.8 rating, 90% response, under 1% cancellation, and ten stays a year, make the badge a switching cost that money cannot quickly rebuild, which holds differentiated hosts far more cheaply than a discount, though it does little for a commodity apartment. A third is to serve the professional managers who will multihome no matter what, with channel-manager integrations and pricing tools that make you the easiest platform to run at scale, so even a host who lists everywhere routes their volume and their best inventory through you. You cannot stop the commodity host from shopping on price, but you can decide whether they still show up on your app first.

Spend where the defection is

The mistake this pattern exposes is treating the complementor base as one bloc to defend with one tool. When a rival enters, segment first: sort your complementors by how dependent they are on the demand only you supply, and how badly your own competition is hurting them. The anchored, network-reliant segment gets your collective programs, the DashPass promotion, the Prime Day, the citywide host bonus, which are cheap precisely because those complementors were loyal anyway and will amplify them. The flight-risk segment gets targeted economics and visibility that neutralize the rival's specific pitch, sized to whether they are leaving over price, crowding, or lock-in. Amazon's surgical apparel fee cut is the template, and the broad mega-sale is the trap. When the turnout at your sale looks like a win, check who showed up. If it was the segment that was never leaving, you spent your defense budget on applause while the people you needed to keep quietly listed somewhere cheaper.