Why Airbnb Cut $1B in Performance Marketing — and Got Almost All the Traffic Back

In Q2 2020, with travel demand evaporating and Airbnb's IPO on hold, Brian Chesky walked into a room of marketing leaders and told them to turn off paid marketing. Not pause it for a quarter. Turn it off. The room didn't love this. Some of them had spent years building the performance machine — bidding on "vacation rental" and "Airbnb New York" across every channel that would take a credit card. They were the ones who'd grown the marketing budget from a rounding error to $1.62 billion in 2019. And now the CEO was telling them their machine wasn't really doing what they thought it was doing.

What happened next is one of the cleanest natural experiments in modern marketing. Airbnb dropped total marketing spend by roughly two-thirds — from $1.62B in 2019 to $545M in 2020 — and the traffic stayed. By Q4 2020, 91% of Airbnb's traffic was direct or unpaid, and the company posted a profitable quarter going into its IPO. The performance team's worst fear — that turning off the paid funnel would collapse demand — never showed up. The brand carried the load.

For most marketers reading this, the reflex is to treat Airbnb as a special case. Two-sided marketplace. Beloved consumer brand. Pandemic-era exception. Don't try this at home. That instinct is half right — most companies are not Airbnb. But the framework Chesky used to make the call is portable, and once you see it, you'll start spotting the same trap in your own paid budgets.

What Chesky was actually testing

The thing being tested wasn't "does brand work better than performance." That framing is a marketing-conference cliché. The real question — the one any CFO would ask if they'd seen the dashboard — was simpler.

How much of the traffic that performance marketing claims to drive would have come anyway?

Performance attribution platforms are designed to take credit. When a user searches "airbnb new york," clicks the brand-name paid ad at the top of the SERP, and books, that conversion gets logged as paid-search-driven. But the user typed your brand into the search bar. They were already coming. The paid ad was a tax you paid to your acquisition channel for traffic that brand demand had generated. This is brand-search cannibalization, and at companies with strong brand equity it can absorb 60–80% of the paid-search budget without anyone noticing — because every channel report says "look at this ROAS."

Chesky's bet was that Airbnb's brand was so strong, and direct demand so dominant, that the performance funnel was mostly billing him for traffic he already owned.

The numbers that made the call obvious

Three numbers, public from Airbnb's S-1 and post-IPO disclosures:

• Direct + unpaid as a share of total traffic in 2019: ~90%. Even before they cut paid, only ~10% of inbound was performance-attributable. That's the floor.

• 2020 marketing spend: $545M, down ~$1.08B. Roughly $1B in cuts.

• Q4 2020 revenue: $859M, up 22% YoY in a pandemic. Bookings recovered fast as travel reopened, but the recovery happened with a fraction of the paid budget.

If your brand traffic share is already at 90%, the math on incrementality gets brutal. To justify the marginal performance dollar, you have to prove it drove a booking that wouldn't have happened otherwise. At a 90% direct-traffic floor, the ceiling on incremental performance lift is 10% of total — and most of that is brand-bidding cannibalization on the user's own search of your brand name. The actual incremental contribution of performance was a single-digit percentage of bookings, against a $1.6B price tag. That's the room Chesky was sitting in.

The framework, in three questions

You don't need to be Airbnb to use this. The decision logic is the same across companies:

• What's your direct + branded-traffic share? Pull GA4 or your equivalent. Sum direct + organic-brand-search + email + app-direct. If it's above 70%, you have a brand-cannibalization problem hiding inside your paid budget. If it's below 40%, performance is doing real work and you should leave it alone.

• What does your performance team's "incremental ROAS" actually measure? If the answer is "last-touch attribution," you don't have an incrementality number, you have a vanity number. Ask for a holdout test — geo holdout, audience holdout, anything where a real control group exists. If they can't run one, that's the answer.

• Can you survive a six-week paid-off test? Not "should you," but can you. If turning off branded paid search for six weeks would crater demand, your brand isn't carrying load and performance is doing the work. If demand barely moves, congratulations — you found your $1B.

What you can do this week

Three moves, in increasing order of organizational courage:

1. Pull a brand-vs-non-brand split on your paid-search spend. Most teams have never explicitly broken this out because the blended ROAS looks better. Brand search is almost always 4–8x ROAS on a last-touch basis; non-brand is usually 1.5–2.5x. The brand line is the one to interrogate. 2. Run a one-week branded-search holdout. Pick a region. Turn off brand bidding. Watch what happens to total branded sessions and bookings. If the loss is small relative to the spend cut, you've found a leak. 3. Bring the result to your CFO before your CMO. This is a finance conversation dressed up as a marketing conversation. The CFO already suspects performance marketing is overpaying for owned demand; they just don't have the numbers. Hand them the numbers.

What to expect

The Airbnb result won't generalize cleanly to a Series-B SaaS company with a 30% direct-traffic share. The lesson isn't "cut performance." The lesson is that brand strength sets a ceiling on how much incremental work performance marketing can possibly do, and most attribution systems are designed to obscure that ceiling. The CFO at your next budget review is going to ask harder questions about marketing efficiency in 2026 than they did in 2024 — agentic AI tools have made the incrementality test cheap and fast, and the holdco earnings reports show what happens when those questions go unanswered. Be the marketer with the holdout result, not the one with the ROAS dashboard.