The Failure Dividend: What Leaders from Airbnb, Google, Pets.com, Quibi, Duolingo, and Toast Learned from Their Worst Moments

Lenny Rachitsky introduces the episode as a compilation drawn from past conversations, motivated by a shift in his own research: he normally studies how the best companies and product leaders operate, but he argues that you can learn a lot—often more—from failure. The episode pulls the most interesting and insightful failure stories from the archive into one focused conversation. The framing turns failure from an embarrassment into a body of evidence.

Katie Dill, head of design at Stripe and former head of design at Airbnb and Lyft, calls her early Airbnb experience the biggest learning moment of her leadership career. She was hired to lead the experience design organization—about 10 people—which had previously reported directly to a founder and had low engagement scores. She came in 'ready to go' and 'swinging,' eager to apply what she had learned about improving collaboration across engineering, product, and design.

About a month in, a Thursday 8:30 a.m. meeting appeared on her calendar: an hour and a half with half the design team—five people—and an HR partner. In a room with white walls and a gray table, the five teammates took turns reading from a stack of papers about everything she was doing wrong. Dill went through the usual stages of grief but managed to listen instead of defending herself, and the common theme she heard was that she had not earned their trust.

She changed her approach immediately, starting with listening to what individuals cared about and what motivated them, while still moving the organization in the direction it needed to go. A couple of months later, the design team had the best engagement scores in the company. Her lesson is that you can inflict change on people, but if you want to do it with them, trust is the key element—so leaders should come in listening rather than swinging.

Paul Adams, chief product officer at Intercom, describes a keynote at Cannes, the world's biggest advertising festival, where Facebook had a standing slot. The year before, Mark Zuckerberg had been interviewed and faced a hard time over privacy, so Facebook asked Adams to take the stage. Three or four minutes into a talk he had given many times, he froze; because it was media-trained and rehearsed word for word, he could not remember what to say.

He had something like a panic attack, walked offstage while still mic'd, and cursed; he could hear people laughing. He walked back out, internally disarmed, and the rest of the talk went well. He became famous and infamous that night in Cannes, and ever since he still carries a tiny doubt before conference talks, though it has never happened again. The lesson he draws is that people have to adapt when life throws curveballs, because these things are not as big a deal as they feel in the moment.

Adams also worked at Google for four years on failed social projects including Google Buzz and later Google+, and he says much of the motivation came from fear rather than from a desire to make a great product for people or to understand how they struggled to communicate with family and friends. He learned how not to lead, even as Google was building extraordinary products such as Maps and Android. Google Buzz was a privacy disaster and Google+ had similar problems.

Halfway through his time there, Adams did formative research with dozens of people over roughly 18 months, mapping their social networks, the channels they used, and what worked and did not. The same pattern emerged every time: people needed much better ways to communicate with small groups of family and friends—obvious in hindsight with WhatsApp or iMessage, but not obvious then. Google tried to build around that insight with Google+, but Adams says the project came from the wrong place; when he left mid-project for Facebook, Google treated him as a possible spy, quarantined him, and forensically analyzed his laptop. At Intercom, co-founder Owen's pitch—'At Facebook you can design the product, but at Intercom you can design the company'—led Adams to help build a culture that embraces big bets and shipping fast, even though that constantly tensions with high craft standards.

Tom Conrad, chief product officer at Quibi and an engineering leader at Pets.com—and now CEO of Zero Longevity Science—has worked at two of the most famous product disasters in history. He says the biggest lesson is not about the specifics of either business: those experiences made him better and became dominoes that opened doors in his career. Even the high profile of Pets.com helped in later interviews, because everyone had heard of it.

At Pets.com, three over-funded pet e-commerce sites each raised more than $50 million and treated the market as zero-sum. They spent irrationally on national broadcast television advertising until it became an unwinnable arms race, a lesson that excess investment can be its own albatross. Timing mattered: Chewy later executed exceptionally well over a decade, grew brick by brick, and was acquired by PetSmart for $3 billion in what Conrad says was the biggest e-commerce acquisition of all time. The critique that shipping dog food could never work was wrong—it just could not work when 80% of the country's internet users were still on dial-up.

Pets.com went from nothing to a public company to out of business in about 19 months, but Conrad notes that it did not go bankrupt: the leadership team shut it down and returned the remaining balance to investors, which no public company had done before. They concluded that market conditions made it impossible to raise more capital for a company that still needed investment to reach profitability, so winding down early was better than spending every last penny on a fruitless salvage attempt. That decision reframes the failure as a disciplined stop rather than a chaotic collapse.

Quibi relit Conrad's enthusiasm for the industry after he had left Snapchat in December 2018 and thought he might be done making software; he had considered becoming a pastry chef, a landscape photographer, or making bad music for SoundCloud. Meg Whitman and Jeffrey Katzenberg recruited him to what was then called New TV, promising to weld the best of mobile and Silicon Valley consumer tech to Hollywood-style content made for the phone. After declining for seven months, he agreed to help a couple of days a week and fell deeply into building the team and product from scratch.

The deeper lesson is that companies are also a math problem: investment goes into an equation and returns come out on some time horizon, and product execution only influences some variables. If the equation is fundamentally broken, no amount of iteration can rescue the output. Quibi bet that a couple of billion dollars could build a bespoke content library scaled enough to drive subscriptions and retention; it made 70 shows in 18 months, more than all major broadcast networks combined in a year, and planned for about a third of content spend to be daily content produced at network-news level.

That daily content was designed to be made the day of or day before it aired, with no back catalog, in professional studios—and Quibi launched two weeks into Covid, when it could only be made in hosts' garages, making it look like YouTube content rather than something set apart. Conrad argues the foundational equation was wrong: the premium bespoke library would have needed to be much larger, and the company needed time to iterate on the format. The real cost was not $2 billion but $6 billion to $10 billion, a risk-reward profile no one could stomach.

Sri Batchu, former head of growth at Ramp, says growth experiments in his experience have roughly a 30% success rate, so most things do not work. He wants a culture where people are not afraid to take risks or fail, but he defines failure as not learning rather than as not driving revenue. Teams should celebrate failure only when the test was designed correctly and failed conclusively; otherwise, a new executive with the same intuition will repeat the same ineffective idea for years.

Batchu argues there are only two ways to make an experiment successful: a very large sample size or a very significant treatment. B2B companies rarely have the large sample size that consumer companies like Facebook can get in two hours; a B2B company might take two years to reach the same number of touchpoints. To compensate, he recommends maximizing the treatment effect—throwing all the tactics and resources that might move the needle into the test, because costs can be rationalized later if it works. If the best, most expensive version still fails, the team can say it tried everything and should not spend more time on it.

He points to account-based marketing as a common enterprise-software example: teams often test it halfway, see no lift, and then a new go-to-market executive repeats the same half-hearted test. At Ramp, Batchu's team designed the experiment to maximize both the number of target customers and the number and types of ways they touched those customers, so the strategy itself could be judged. He extends the logic to a micro email test: instead of only tweaking copy, a team can change the trigger, the content, the personalization, and the design together. The trade-off is that a failed all-levers test does not reveal which lever was wrong, but it can still tell you whether the overall hypothesis is wrong.

JZ, head of product at Webflow when the episode was recorded and a former Airbnb colleague of Lenny's, says one of the hardest habits to untrain in new product managers is jumping to solutions. People become attached to a specific implementation they can see in their heads and start with 'I want to build X' instead of understanding real users and their problems. Her course forces students to set aside the thing they want to build and first understand the problem and whether there is an opportunity.

Her own biggest product mistake was Airbnb Plus, an effort to address trust in Airbnb's quality at a time when managed marketplaces were creating competitive anxiety. The team went solution-first and competitor-afraid, deciding to inspect and more directly manage inventory. JZ says they should have stepped back to the real problem: people want to know what they are getting into, and Airbnb needed to represent homes better. They also ignored Airbnb's strategic weakness in operations and the unit economics of sending inspectors and managing inventory.

Lenny notes that from the outside Airbnb Plus looked clever—establish a minimum quality bar for a platform whose inventory was unpredictable—but JZ doubts there was ever a clear path. She warns against magical thinking about unit economics, where teams assume everything will work at some future scale; the unit economics need to work at the beginning. For Airbnb, guest reviews were a free quality signal, and other targeted fixes could have been cheaper than inspections—sending a lockbox, for example, or partnering with local cleaners. The lesson is to ask what problem a specific listing has and choose the most tailored solution rather than one blunt instrument.

Gina Gotthilf, COO of Latitud and an early growth leader at Duolingo, says people are encouraged to present their A side because it impresses, opens doors, and keeps them growing. Podcasts and stage talks therefore become the Instagrammable version of a person or company—just the highlights. Her A side includes meeting President Obama, working on Mike Bloomberg's presidential campaign, helping Duolingo scale from 3 million to 200 million users, working with Tumblr in Latin America, and having Andreessen Horowitz invest in her company. But between those highlights are many B moments that get shoved under the rug.

Her B side includes not knowing what she wanted to do—she considered being an actress or the person at SeaWorld who works with a dolphin—and not getting into any Ivy League or top-choice school. She dropped out of Reed College, the same school Steve Jobs dropped out of, because she was depressed and could not get out of bed, then eventually returned and graduated; a college counselor looked at her transcript and asked what she had even done with her life. The learning she took was that it is not enough to do things that matter and learn; you also need to be able to tell the story and understand what other people perceive as valuable.

She applied to a hundred companies and heard back from few, finally landing an internship at a tier B/C digital marketing agency in New York because she wanted to live there—then the agency forgot to apply for her visa on time, so she lost it and returned to Brazil. She left for another company, was laid off, lost her visa again, found another opportunity, and was fired. At Tumblr, which she thought was 'it,' the company could not figure out how to wire money to Brazil, so she was unpaid for six months; she and colleagues borrowed money to pay contractors, and then Tumblr laid her off when it decided to sell to Yahoo.

She started an agency to help US-based tech companies grow in Latin America, and a connection through Tumblr's growth in Brazil led Duolingo to her when she was 26. She helped Duolingo in Brazil as a consultant, then said yes to Chile, Argentina, and Mexico despite not knowing those markets, then went full time across Japan, China, Korea, Turkey, Spain, France, and elsewhere, eventually owning growth across communications, social media, government partnerships, and an A/B-testing growth engine with engineers, PMs, and designers. After leaving Duolingo five years later, she still did not have everything figured out: she tried nonprofits, worked on Bloomberg's unsuccessful campaign, and later joined Latitud. Lenny adds that careers are long—he spent nine years at a random San Diego startup before things took off—and Gotthilf agrees that at 26 she thought her career was over, when in fact there was much more time and opportunity ahead.

Maggie Crowley, VP of product at Toast, asks every product candidate: 'What's the worst product you've ever shipped?' She does not think someone is a good PM if they have not shipped something really bad, because that means they have not had enough reps. It is not only that they have shipped a bad product but that they can admit it and name which one.

Her own example was a rewrite of an existing core product—a red flag from the start. An engineer she had worked with many times said it would take six months; it took two and a half years and still was not done, and feature parity took forever. People rotated in and out, everyone thought it was dumb, and sunk cost fallacy kept it alive. The causes were arrogance that they could do it, skipping discovery, not writing a one-pager, and not doing enough technical and design research into requirements.

The project was not a success, though Crowley notes she did not get fired. Her advice is blunt: do not rewrite, and do not do a side-by-side rewrite. Lenny adds that once teams get too far down a redesign or rewrite, everyone is building in the new world, the launch experiment comes back negative, and the team ends up saying it just has to launch and get back to neutral someday; Crowley's response is: don't do that.

Lenny closes by thanking the guests for being vulnerable and sharing career failures, and he hopes listeners leave with a new perspective on how setbacks and challenges can be exactly what they need to reach the next step. He invites listeners to share their own failure stories through YouTube, the newsletter, Twitter, or his website. The compilation's through-line is that failure is not the opposite of a good career; it is part of the evidence that a career is being built.