Marketplace moats fail in three characteristic ways. Learn them as category diagnoses, because you can usually predict the killer from the transaction's anatomy before the numbers confirm it.
What is disintermediation, and how do platforms defend against it?
Killer 1: disintermediation (leakage). The platform performs the expensive act, which is discovery and trust-building. Then the parties transact off-platform, free of the take rate. Leakage risk is highest in four situations. One is a high ticket, since a 15% take on a ₹5 lakh interior-design job funds a lot of off-platform coordination. Another is a relationship that repeats with the same counterparty: your maid, your physiotherapist, your firm's freelance designer. Discovery happens once and the value recurs forever. Leakage also rises when payment can move offline easily (UPI makes it frictionless in India), and when the platform adds nothing per transaction after the first match. Home services, freelancing, B2B sourcing, and real-estate brokerage are structurally leaky. Food delivery and ride-hailing are structurally tight, because every transaction needs fresh logistics that the platform performs.
Defenses come in rising order of strength. Punishment, meaning a ban on off-platform contact, is weak and adversarial. Payments/escrow and invoicing make the platform the safest way to pay. Insurance and guarantees, AirCover-style protection that's worth the toll, come next. Workflow lock-in puts scheduling and disputes on the platform, along with compliance. Strongest of all are heterogeneous, non-repeating matches. You get a different Airbnb host every trip, so leakage has nothing to attach to. To detect leakage, look for cohort transaction decay that's steeper for repeat pairs than for new pairs. That's the leakage signature. A growing mix of take-rate-free "lead generation" is another sign. Then do the scuttlebutt. Ask five Urban Company pros how many of their regulars book direct, and their answer is the leakage rate.
Advertisement
How does multi-homing hurt a marketplace?
Killer 2: multi-homing. Restaurants and riders are on Zomato and Swiggy. Sellers are on Amazon and Flipkart and Meesho. Drivers cycle between apps at a red light. When both sides use several platforms at once, the network effect stops being a moat and becomes a shared utility. So incentive spend never ends, because you must pay both sides not to drift. Take rates are capped, because sellers arbitrage the cheaper toll. And duopoly economics look like trench warfare until someone's capital runs out. What breaks multi-homing? Exclusive supply is one answer, through content exclusives or own-brand inventory. Single-homing one side through economics is another, with driver loyalty tiers or seller tools too good to duplicate. Liquidity so lopsided that the second app dies of emptiness also works.
India's food-delivery duopoly is the live case. Both platforms are rational, both are profitable post-2023, and both are structurally unable to raise commissions much, because every participant multi-homes. Compare the US, where DoorDash's share dominance (~two-thirds of the category) gives it pricing freedom its Indian analogs lack. For detection, run the share-stability test from the moat work. Share swinging >2–3%/yr means there's no lock-in. Also watch for incentive spend that refuses to fall with scale, and for seller/driver overlap surveys in broker or S-1 channel checks.
What is take-rate compression?
Killer 3: take-rate compression. The toll can fall for four reasons. Sellers gain power, through concentration into large sellers who negotiate or through a restaurant association boycott. Competition undercuts, with a rival at 12% commission. Regulation intervenes. App-store rulings force alternative payments, India's ONDC project explicitly attacks aggregator take rates by unbundling discovery from fulfilment, and consumer regulators scrutinize delivery charges and dark patterns. Or the channel shifts, as sellers mature into their own D2C sites, and Shopify exists to arm that very defection. Compression of 200 bp on an 18% take rate is an 11% revenue cut at 100% incremental margin, so it falls straight to contribution. The mirror image, take-rate expansion through ads and payments, drives the bull case, with the Gurley ceiling always overhead. To detect compression, track take rate by cohort and category over 8+ quarters, stripping out mix effects. Follow commission-cap news flow and large-seller revenue concentration. And track ONDC or regulatory volumes in the category.
What is subsidy-bought GMV?
And the meta-flag: subsidy-bought GMV. Negative contribution margin defended as "investment in growth" counts as investment only if cohorts mature into positive CM with the incentives off. You ran this test on ThaliGo's subsidy exit. If CM is negative and cohort activity collapses whenever discounts taper, the platform is just a pass-through for investor capital to consumers. India's quick-commerce land war (2024–26) is the sector's live laboratory. Watch which players' dark-store cohorts turn contribution-positive as they age, and which only grow while burning. Consolidated per the brief, the four marketplace red flags are (1) disintermediation/leakage, (2) subsidy-bought GMV (negative CM), (3) multi-homing/low switching costs and (4) take-rate compression. Add the disclosure flag too: liquidity or cohort metrics that get dropped without a word.