先别管双边理论：Lyft 与 Thumbtack 老兵讲流动性、供给和控制

Lauzier defines a marketplace as two or more distinct sides that provide value to each other, with an intermediary facilitating the exchange. How managed it is varies, from hands-off Craigslist to semi-managed Lyft. The biggest mistake he sees among early founders is obsessing over supply-and-demand ratios and economics papers before they have product-market fit.

His advice is to forget the marketplace dynamics if you do not yet have product-market fit and a growth strategy for at least one side. Instead, focus on the core exchange of value and go deep on one side. He suggests picking the hardest side, which at Thumbtack was demand: finding people with home projects was harder than finding plumbers.

To get started, use a crutch for the other side. Thumbtack and Airbnb both used Craigslist early on, and Lauzier describes Thumbtack posting jobs there behind the scenes to bring contractors to its platform. Teams usually know which side is hard, and Lauzier and Lenny Rachitsky agree it is supply roughly 80 to 90 percent of the time.

Rachitsky notes that Rover and TaskRabbit had the opposite problem, with plenty of people eager to walk dogs or do tasks. Lauzier lists common early tactics for the hard side: jumpstarting through existing channels like job boards, building value-added services such as OpenTable did for restaurants, and converting one side into the other.

At Lyft, where riders were on a waitlist because drivers could not be onboarded fast enough, the team tested a pop-up. Instead of saying no drivers were available, it told riders drivers were earning about $50 an hour and asked if they wanted to drive. It produced some conversions but was distracting and never a big source of supply, though Lauzier says it was meaningful at Uber.

For companies with scale, Lauzier says the most common trip-up is liquidity: the ability to match buyers and sellers efficiently. He pictures a Venn diagram of what supply wants to sell and what demand wants to buy, with liquidity as the overlap. At Lyft or Uber, it is the share of app opens with ride intent that turn into rides, and it also drives the engagement loop since more supply means more choice and more return visits.

Liquidity is usually measured as demand utilization, or fill rate of intentful demand, but exogenous factors like snowstorms or competitor bidding move it around. Lauzier prefers what he calls a market-health metric, the best predictor of liquidity. For Lyft it was pickup ETA: once the closest driver was around two minutes away, you hit a ceiling and further improvement barely changed whether people booked.

That proxy is more actionable. A supply team can ask whether adding 100 drivers actually reduces ETAs, without being thrown off by exogenous noise. Rachitsky adds that at Airbnb they used searches with dates as intentful demand, and that the biggest lever on fill rate is usually the amount of good supply.

Lauzier's take is that product-market fit is largely independent of marketplace dynamics, so you should measure it as any company would, such as asking what percentage of users would be very disappointed if the product went away. Rachitsky points to the Sean Ellis test and agrees that around 90% of marketplace challenges are the same as in any other business.

The difference is that you effectively have two product-market fits, one for each side. Often demand is compelling while suppliers are not satisfied, perhaps because margins are too high. Lauzier suggests measuring both in a traditional way.

On whether an idea suits a marketplace, he looks for three signs: fragmentation with a long tail of buyers and sellers, relatively uniform needs so supply can be somewhat commoditized, and a high barrier to matchmaking, meaning it is hard for people to find and vet each other. He notes that service marketplaces like Thumbtack are tricky because electricians want specific jobs, on specific days, and may cancel.

Lauzier says Sidecar, a Lyft and Uber competitor that later closed, tried to differentiate by giving riders full control with filters such as a minimum car year. In his view, that further fragmented the marketplace and hurt wait times, because a rider who slid the filter to 2020 might not realize they had just excluded a great driver in a 2018 Honda Civic and added ten minutes.

Rachitsky draws a general lesson about not over-listening to users and not giving them more options than they need, and notes Sidecar even let drivers set their own prices. Lauzier then offers the opposite advice for companies struggling with market health: open up your supply walls. At Thumbtack, many wedding customers checked a smoke machine for DJs, but only about 5% of DJs had one, so the filter eliminated 95% of supply.

Users told Thumbtack they did not actually care that much. The fix was to make the checkbox affect ranking rather than filtering, which respects the preference while reducing cognitive load. Rachitsky also adds that some marketplaces, like Cherry's on-demand car wash, fail simply because the economics do not work.

Asked for the biggest reasons marketplaces fail, Lauzier first names liquidity. Without a diagnostic framework you can run out of time and money before you build enough density for both sides to have a good experience. He felt this rush at Lyft and has seen it elsewhere.

The second is ignoring one side. Many large marketplaces focus only on the demand funnel, running ads and converting clicks, and hope that supply is good enough. Because marketplaces are laggy, when network effects fade there is panic that they forgot sellers are people too, and they must rebuild their value proposition for supply.

The third is quality. Being a marketplace implies curation, but many companies lack intentionality about their bar and keep lowering it to get more supply, until the whole site starts to look shady. Lauzier says you do not need the best quality, but you need to decide deliberately what quality you aim to provide.

Rachitsky asks about moving toward a managed marketplace, since quality is hard when you do not control supply. Lauzier warns that marketplace operators sit in an ivory tower thinking people will act rationally. At Thumbtack, the team replaced selling leads with selling direct bookings, expecting about 20% better ROI for pros, and pros hated it because they enjoyed the thrill of the sale and overestimated their closing ability.

He saw similar perception problems at Lyft when trying to make driver earnings less volatile. So he is a believer in market forces and empowerment: provide guardrails for a good experience, set a clear quality bar, give supply coaching and tools, then step back and invest in hands-on tactics only where gaps remain. He also notes the legal risk that controlling supply can lead to employee classification in the U.S.

As an example of raising quality without becoming managed, he points to Toptal, which vets heavily and advertises accepting around 3% of applicants. Allegedly the real rate is even lower, but the company thought the true number would sound fake. Rachitsky notes this only works with plenty of supply wanting in.

Lauzier was leading driver-side product at Lyft when General Motors invested half a billion dollars in its funding round. On Christmas Eve he got a call from Lyft's CEO and GM's CTO, and they decided to build a rental company. GM had vehicles coming off lease that it had to sell at auction, while Lyft had a supply crunch and could not onboard drivers fast enough.

Lyft had noticed that 50% of job seekers and welfare recipients in the U.S. do not have a car, a large pool it could not tap. Renting cars let it manufacture supply surgically by market, volume and price, and even offer to pay for the car if the driver drove 30 hours a week. Within about 18 months, Lauzier says, it was the fourth-largest rental fleet in the U.S.

It also gave Lyft control over vehicle quality, since it could bring in newer cars in markets where quality was low. The drivers were very loyal, and Lyft could require, for example, not driving for Uber and driving a minimum number of hours. Rachitsky points out that this sits further along the managed-marketplace spectrum and gave Lyft a lever Uber did not have.

Around 2014 and 2015, Uber was roughly 30 times Lyft's size, so everything Lyft did had to be ten times more efficient per person just to survive. Uber ran DMV-style onboarding offices with large lead times, and Lyft did the same in its first three or four markets. Lyft instead leaned on its brand and community of fierce advocates and began paying its best drivers $35 per mentor session to inspect a new driver's car and documents and ride along.

The results were unexpected. Mentors shared personal tips and their contact information, giving new drivers social proof that beat the company's best marketing copy. The program was scalable, since a small team could vet 10 or 20 top drivers who then onboarded others in other markets, and it also improved retention of the best drivers, who saw becoming a mentor as a promotion worth up to $70 an hour. Lauzier says this let Lyft match most of Uber's footprint with a tenth or twentieth of the resources.

The next iteration targeted people who dropped out of the funnel, such as those who had not entered their SSN. After building a team of hundreds of account executives to phone them, Lyft let drivers act as recruiters, claiming leads on a mini sales dashboard for $20 per driver converted to activation. They outperformed trained salespeople and also helped smooth supply and demand by giving drivers something to do during low-utilization times.

Rachitsky notes Uber is worth around $150 billion against Lyft's roughly $5 billion and asks why Uber won. Lauzier says he has been away from Lyft for years, so this is a naive view. He thinks the biggest blow was inherent in the vision, which was anchored in transportation, changing how cities and roads are designed, and dynamic shuttles.

That meant Lyft never invested in food delivery or goods and parcels, which in his view crushed it during COVID. Uber, with a logistics platform for moving people and things, rebounded faster because it was diversified. Lauzier also notes Lyft had leaned heavily into shared rides, the opposite of what people wanted during the pandemic, and says he is sad that a new COO killed shared rides.

Lauzier, who has spent his career in the U.S. and now lives in France, sees product management growing fast in Europe but under different market dynamics. In the U.S., a liquid job market leads to greater ownership and accountability, with PMs given a meaningful piece of the business. In France, jobs are harder to change and firing is expensive, which he thinks leads to less autonomy, more micromanagement and founders who struggle to let go.

He also sees startups hiring product people later, outsourcing to product studios such as Mozza, and a dominance of business over tech, with top schools like HEC being business schools. Equity is also less meaningful: product leaders often consider it nearly worthless, and for executives it is often under 10% of compensation versus often more than 50% in the U.S. He adds that the startup scene is vibrant, especially in AI with companies like Mistral AI and Hugging Face, and that the French government has committed around $2.5 billion to AI by 2030.

For European companies wanting to operate more like U.S. ones, he admits he has not cracked it but offers pointers: educate employees about the value of equity and the business trajectory, and build teams around ownership and accountability for a slice of the business rather than feature factories. Rachitsky wonders if this is just founder mode, and Lauzier adds that venture capital scarcity forces a business-case-first mindset.

Lauzier says founding his first company has been humbling, since zero to one is harder than scaling within a company where the problem space is already validated. His wife developed health issues about three years ago, and navigating U.S. care meant waiting three months for a neurologist who saw her for about eight minutes, then repeating the cycle with another specialist. The experience was isolating, with conflicting advice and no one fighting for her.

He says nearly half of Americans have a chronic condition, and he frames the problem as systemic rather than about practitioners. After talking to hundreds of patients and doctors, he built Nurra, which connects people with complex conditions to a personal health advocate available 24/7. Advocates find appointments, help prepare for them, explain diagnoses and test results, research treatments and chase referrals, with the mental model of acting as if the user were your partner or parent.

Rachitsky asks how this relates to marketplaces, and Lauzier says he is following his own advice: ignoring marketplace dynamics, jumpstarting the advocate side for now and focusing on the harder demand side. A marketplace component might come later. He also mentions his Reforge marketplace course, which he recommends only to those who already have product-market fit.