How a Borrowed Camera and One Week of Data Saved Airbnb

In 2009, Airbnb had a problem they could see clearly in their data: New York listings were not converting. People were visiting the site, looking at listings, and not booking. The team had a theory about why.

The theory was that the photographs were bad.

Guests were trying to book accommodation from blurry, dimly lit photos taken on phones. The photos communicated nothing useful about the space. If you were deciding between a hotel, where you knew exactly what you were getting, and an Airbnb listing you could barely make out from the images, you booked the hotel.

Brian Chesky and Joe Gebbia did something that would not scale and was not meant to: they flew to New York, borrowed a camera, and spent a weekend photographing host apartments themselves.

The Test and What It Measured

This was not a branding initiative or a product roadmap item. It was a test with a very specific hypothesis and a metric they could read immediately.

The hypothesis: if the photos are better, listings convert better. The metric: revenue in New York listings over the following week, compared to the week before.

The result was unambiguous. Revenue in the photographed New York listings doubled in a week.

The reason Chesky and Gebbia went themselves rather than hiring photographers immediately is important. At that stage, they could not be certain the photography was the variable. Doing it themselves was the fastest way to get a result with the fewest confounds, they were not simultaneously changing prices, copy, or anything else. If revenue moved, it was the photography.

Revenue moved.

What the Data Told Them to Do Next

Having confirmed the photography hypothesis, the question became: how do you scale a founder flying to cities with a borrowed camera?

The answer Airbnb built was a professional photography program. They hired and contracted photographers in cities, offered the service to hosts, and tracked the outcomes, conversion rates, booking frequency, and revenue per listing, on photographed versus un-photographed listings at scale.

The data from the scaled program confirmed what the initial test suggested. Professionally photographed listings outperformed unimproved ones consistently. By 2012, Airbnb had a network of over 2,000 contracted photographers across 13 countries.

The key thing about this progression is that each step was data-led. The initial test gave them a hypothesis-confirmed result in one market. That result justified investing in a more expensive test, a small contractor network, which generated the data to justify the global rollout.

Nobody approved a global photography program on intuition. They approved it because the data from each prior stage said the investment would return.

The Reason This Works as a Strategy

The Airbnb photography story is often told as a scrappy startup move, founders doing manual work that doesn’t scale. That’s accurate, but it misses why the approach was intelligent rather than just resourceful.

Testing in a single, bounded market before scaling is a way of getting a data signal cheaply. New York in 2009 was small enough that Chesky could do it himself, observable enough that the revenue metric was clear, and isolated enough that a week’s data was meaningful. The test cost a plane ticket and a few days. The information it produced was worth the global photography program that followed.

This is the underlying discipline: matching the cost of a test to the confidence you need to justify the next step. A cheap, fast test that answers one question precisely is more valuable than a larger initiative that changes multiple variables at once and tells you nothing clearly.

Airbnb went from $200,000 in weekly revenue in 2009 to a $113 billion IPO in 2020. The photography test is not the reason for that trajectory. But it is a useful example of the decision-making pattern that produced it: observe a problem in the data, isolate the variable, run the cheapest test that will tell you whether you’re right, then scale what works.

The Transferable Pattern

Most marketing functions do not run enough controlled tests. They launch campaigns across multiple channels simultaneously, change creative and targeting and budget at the same time, and then try to attribute the result. That attribution is nearly impossible when multiple variables moved.

The Airbnb approach is the opposite: change one thing in one market, measure a clear metric over a short window, and read the result. It requires discipline about what you’re testing and patience with small-scale initial tests, both of which run counter to the pressure to move quickly at scale.

The businesses that compound on data advantage are usually the ones that have a formal habit of running small tests before committing large budgets, not because they’re more cautious, but because the cheap test makes the expensive rollout more confident. Chesky’s borrowed camera was not a budget decision. It was an epistemological one.