Matteo Franceschetti on Eight Sleep, AI Teams and the Cost of Ambition

Eight Sleep co-founder and CEO Matteo Franceschetti joins Harry Stebbings to discuss hiring, marketing, hardware economics and the company's use of AI. They also talk about working with a spouse, the costs of ambition, and the founder's expectations for health technology and the future of work.

Franceschetti says hiring has become as competitive as it was around 2021, with high compensation offered not just to engineers but also to product managers. He says Eight Sleep has looked increasingly to Europe and other markets because retaining people in the United States is difficult when they receive repeated offers.

He says the company has lost a few people to prominent AI firms and other fast-growing businesses. Sometimes the gap in compensation is so large that a startup cannot reasonably match it. The observation is based on his hiring experience, not a comprehensive survey of the labor market.

Angel investing has also changed how he operates. Updates from non-competing portfolio companies give him visibility into fundraising, valuations, hiring and marketing channels. He treats those reports as a way to understand what other founders are seeing rather than relying only on information from his own business.

Asked what investors misunderstand about hardware, Franceschetti cautions that founders may not initially know the real cost of production or product returns. In his experience, early estimates of the bill of materials and cost of goods can be materially wrong. Investors should test those assumptions rather than treating a prototype as a finished manufacturing process.

He says word of mouth accounts for around 40% of Eight Sleep's revenue even at its current scale. Paid channels include Meta, Google and TikTok. The figure is the founder's account of company attribution; the interview does not independently audit the percentage.

TikTok started working for Eight Sleep in the preceding six months, he says, after earlier attempts failed. He credits iteration and producing sufficient content with creators. His team begins a channel with a small weekly budget and a customer-acquisition-cost cap, then increases spending when results justify it.

Franceschetti recommends incrementality tests about every six months: pause a channel in one geography and compare performance with a similar area where it remains active. As channels are added, their contributions can change. He says Eight Sleep uses its own models instead of relying solely on a platform's reported attribution.

As an example, he says Meta's reported acquisition cost can be lower than the company's estimate of the true cost. A geographic test can help estimate lost revenue after advertising is paused, though the areas need to be comparable. His specific estimate is company experience, not proof that every advertiser faces the same gap.

He advises launching growth channels one at a time. Start with a manageable budget, set an acceptable acquisition cost and scale only while the channel meets it. Spending aggressively can create impressive short-term growth, he says, but a later pullback may leave a company with weaker year-over-year comparisons.

Eight Sleep looks for immediate payback and positive contribution margin on the initial transaction, Franceschetti says. Subscription revenue then adds to that base. He credits an early investor with pushing the company to focus on day-one economics and says the business could grow faster if it accepted weaker margins.

If economics were no constraint, he says he would prefer spending more on marketing rather than cutting prices. Brand awareness remains a bottleneck in his view. He also describes subscription churn as low single digits annually; that is a company-reported metric, and the conversation provides no independent comparison with other products.

The company now monitors how it appears in AI-driven search, alongside Google, although Franceschetti does not have a precise percentage of traffic from that source. He expects the contribution to grow. The conversation treats this as an operating observation rather than a measured industry-wide shift.

His channel mistake was an out-of-home campaign in a Middle Eastern market where the brand was not yet well known. The same format might perform differently in New York, he says, because the audience starts with greater awareness. He recommends monitoring both prompted and unprompted brand awareness when considering such campaigns.

Instead of buying more out-of-home advertising at the time of the interview, Eight Sleep has a demonstration truck with a bedroom inside. People can book a visit or encounter it at events and try the product. Franceschetti says the format has worked well because many potential customers wanted a way to experience it before purchase.

He puts the company's marketing budget in the hundreds of millions, without providing a detailed breakdown. One example of automation is email: after the lead of a two-person team planned to leave, co-founder Alexandra explored whether AI could take over parts of the work.

Franceschetti says Alexandra built multiple bots in three days and the dedicated email team is now zero. He attributes close to $100 million to email marketing, but the discussion does not establish an audited causal measure or explain how that revenue is allocated among channels.

He describes an email calendar informed by historical data, with agents suggesting what to send and when. He believes the system uses a third-party model, though he says he would need to check exactly which one. He counts many internal AI agents separately from the company's human employees.

For paid digital marketing, he says agents prepare a daily report suggesting changes and the human team approves or rejects them. He expects people and their personal agents to discover products differently as content volume rises, so brands may need to think about both audiences.

On engineering, Franceschetti says the company's engineers stopped writing code around a year earlier and now supervise many AI coding agents. He describes that as enabling a small team to support operations across 35 countries. The claim describes his company's workflow; the interview does not independently assess every engineer's work.

He says Claude is the main tool and that monthly spending is in the millions of dollars, below $5 million. He does not provide a precise current figure. The team still needs to manage the agents and the product, even if the tools generate much of the code.

Stebbings asks whether AI model spending will become a larger share of engineering budgets or fall as models become cheaper. Franceschetti expects usage to rise substantially while unit costs fall, with the net spending share eventually declining. This is his forecast, not a demonstrated budget outcome.

He compares the change to electricity becoming less expensive and more widely used over time. Both speakers expect productivity gains, but the size and distribution of those gains remain uncertain. Stebbings describes an agent he has asked to analyze competitors' social-media content and suggest stories.

Asked about a hypothetical Anthropic IPO, Franceschetti says he would probably buy and expresses optimism about AI companies and SpaceX. His comment is a personal view about a potential investment, not a confirmed IPO plan or advice to buy any security.

Franceschetti's broader theory is that AI could help companies become holding companies. As teams build tools to solve their own problems, he expects them to find adjacent businesses; Eight Sleep's current product might eventually become one business among several. That is a strategic possibility, not an announced set of launches.

He points to Chinese companies such as Xiaomi and BYD as examples of firms extending capabilities across product categories. His idea of focus is not necessarily doing the same thing forever, but applying distinctive knowledge to new markets. He believes AI could make that kind of expansion easier.

The two discuss Chinese automakers and potential competition for European manufacturers. Franceschetti praises the vehicles and customer experience he has seen on work trips, including design and in-car features. His comments reflect his experience and expectations, not a demonstrated prediction of any national industry's outcome.

He predicts car manufacturers could also become robotics companies and describes humanoid robots as a potentially enormous market. The speakers discuss demonstrations and sporting events, but their excitement and claims about records are not independently verified in the interview. Future market size and capabilities remain uncertain.

They also consider warnings about the speed of frontier AI development. Franceschetti worries about systems that improve themselves and speculates that capabilities inside labs exceed what users see. He does not present evidence that a particular system has achieved autonomous self-improvement or that control has already been lost.

Asked about geopolitical concerns, he declines to offer a broad judgment and says China is an important market and manufacturing location for Eight Sleep. He describes the practical challenge of entering it: the company had to learn new distribution and purchasing habits.

Franceschetti says Eight Sleep's sales rely heavily on local platforms such as WeChat and RedNote rather than the direct website model it used elsewhere. His description is simplified and based on his company's experience; it should not be read as a claim that no consumers in China use websites for e-commerce.

On payment security, he says his own shopping bot uses a browser where Amazon already holds his card, rather than giving card details directly to an AI system. He expects trust in agent payments to increase, comparing current hesitation with the early days of online credit-card purchases. That is a personal forecast, not a security guarantee.

In a speculative example, he imagines an agent buying and operating an autonomous taxi on a person's behalf. He also hopes AI and health technology will produce major advances. The examples are future possibilities he proposes, not available capabilities or established medical outcomes.

Franceschetti says Eight Sleep hopes its data and technology will support detection of health risks and sleep-related conditions. These are company ambitions and claims in development; the interview does not establish diagnostic accuracy, regulatory status, or that the product can prevent disease or save lives. Such claims are not medical advice.

Timing is a recurring lesson for him. A channel such as retail or out-of-home advertising is not simply good or bad: it may require brand recognition, operational capacity and other conditions to work. He expects retail to become more meaningful for Eight Sleep after believing it would have been premature earlier.

He says some Chinese founders told him they build recognition abroad before selling at home, because a known brand may command a premium. He believes Eight Sleep's international presence and public endorsements helped its own launch. This is his account of conversations and strategy, not a rule for every Chinese consumer or product.

Stebbings asks whether hardware companies must be the cheapest or have the most features. Franceschetti accepts that low price can work, but says his preference is a strong product and brand, rather than features alone. He uses the iPod as an example of packaging and positioning shaping demand.

He argues that Silicon Valley can underestimate brand, although he says a good brand cannot rescue a bad product. A strong product comes first; brand then helps customers understand its appeal and can support a price premium. The discussion concerns positioning and consumer perception, not a guarantee of product performance.

Franceschetti tells the story of Formula 1 driver Charles Leclerc, whom he says already owned the product before they met. Franceschetti had tried to sponsor him earlier in his racing career. After they eventually connected and discussed the company's direction, he says they agreed to work together and Leclerc invested.

He says Eight Sleep has learned from several years of sports partnerships, including cycling, tennis and padel. His two requirements for an ambassador are that the person has tried and understands the product, and that he can have a direct relationship with them. He sees genuine use as more valuable than a purely financial endorsement.

The founder says some athletes use the product without publicly disclosing their relationship. He offers an anecdote about a footballer who replaced a custom mattress after trying it, and another about an actor buying units for family members. These are his unverified customer stories, not evidence of clinical effectiveness.

He also tells stories of wealthy customers buying many units for homes or hotels. Asked whether the product is too expensive, he says the company's ambition is to reduce prices and broaden access over time. He does not announce a specific lower-cost product or date.

Stebbings asks how Franceschetti works with his wife and co-founder Alexandra. He says they separate work and personal communication using Slack and WhatsApp, and respect her preference not to discuss non-urgent work while relaxing in the evening. He may still send a message for later rather than interrupting her.

He says the early years required learning to remain professional even when demanding the same standards from someone close to him. Alexandra asked to be treated like any other executive in the way he communicated. He says their working relationship became more finely tuned over the following years.

Franceschetti reflects on uncertainty as an immigrant founder who studied law in Italy and did not come through the usual Silicon Valley institutions. Early on, he wondered whether he could build a product that people beyond friends and crowdfunding supporters would buy and sustain as a business.

He recalls moving to China for several months after fundraising because prototypes had not translated into reliable manufacturing. None of the founders had manufacturing experience, and he returned after working on the problem. He says the company's longest-tenured employee joined during that period.

Asked how the couple would handle similarly intense travel if they had a child, he says they would need to work it out together. He does not offer a tested solution. The question exposes the tension between a founder's commitment to a business and the time a family may need.

He says he and Alexandra had not taken a conventional vacation since starting Eight Sleep and even scheduled their wedding around a crowdfunding campaign. He emphasizes that he is describing their preferences, not suggesting everyone should live the same way. He says he becomes restless after a short period away from work.

During a weekend in the Maldives, a hardware leader called to say he was leaving. Franceschetti spent much of the trip trying to retain him. He says the employee eventually stayed after further discussion and an adjusted offer, despite another company offering substantially higher compensation.

Asked whether a startup should match a major lab's pay, he says candidates should consider the very different jobs, stages and working environments, not only salaries. He describes a case where an employee's reason for considering a move involved travel and pressure as well as compensation. This is his hiring philosophy, not a universal rule about candidates.

He recounts seeing some candidates accept an offer and soon leave for another opportunity, while cautioning against generalizing from his experience. He thinks employees should understand what they and their families need, including whether they value potential equity upside or more certain near-term cash.

As an example, he describes promoting an internal candidate to vice president of engineering after an external search. Before accepting, he asked the employee to discuss the anticipated workload and commitment with his spouse. The conversation is his account of management practice, not a prescription that every family should make the same choice.

Franceschetti says Eight Sleep has about 160 employees and compares its revenue per employee with larger companies, without disclosing revenue. Looking three years ahead, he proposes a target of roughly 250 employees and $1 billion in revenue. Those comparisons and the target are company claims and forecasts, not independently verified results.

He describes small, flat teams, sometimes with two people overseeing a large area while using AI tools. A separate internal AI team works closely with data engineering so agents can access the right country-level revenue and acquisition-cost data. He says accurate data connections are a central operational challenge.

He does not predict that every job category disappears; instead, he expects fewer people to cover more functions. He says paid media and finance are examples where automation helps the company keep teams small. The reported impact is his assessment of Eight Sleep's operation rather than a measured causal estimate.

In a quick-fire round, Franceschetti gives his personal sleep advice: he emphasizes sufficient and consistent sleep, and says eating before bed affects people differently. He describes changes in his own heart-rate and variability readings after pizza. These observations do not establish an individual diagnosis or a treatment recommendation.

He says he has come to value consistent basic habits over an excessive focus on products and supplements. His view reflects his own approach to health; individual sleep needs and medical concerns vary. The interview does not demonstrate that any particular routine or Eight Sleep product prevents illness.

Franceschetti predicts that children born today may not need to work for money by adulthood because of AI. He imagines people pursuing activities and relationships for other reasons. This is a speculative forecast about technology and society, not a certainty about future employment.

He describes a personal framework for well-being built around health, relationships and purpose. Stebbings asks about university and universal basic income; Franceschetti questions the value of university for some aspiring Silicon Valley entrepreneurs and expects society may need new ways to distribute the gains from automation. These are his opinions, not general education or policy advice.

On whether children should use social media, he says he has not yet settled on an answer. To prepare for possible parenthood, he has begun asking parents of successful people what they believe worked and what they would change. Stebbings shares his appreciation for a mother who spoke openly about mistakes.

The conversation turns to family and ambition. Franceschetti says his father set a high standard and may still influence his drive, though he is unsure. His father died while he was a lawyer in Milan, before he moved to the United States, became an entrepreneur or married Alexandra.

His mother worried about the risk of entrepreneurship because she had limited means, he says. He took comfort in believing a former law-firm employer might take him back if the venture failed, although he had recurring dreams about returning to law. Both speakers acknowledge that financial security shaped their early career choices.

Asked what advice he would give a young person, Franceschetti says he cannot reduce his path from Ferrara to Silicon Valley to a simple formula. He followed opportunities and connections over time. He says he did not speak English until his twenties and knew little about computers before then.

He recounts learning about international law firms and deciding he needed excellent grades and English to work at one. His father took a loan that helped him study English in San Diego; he later joined major law firms before becoming an entrepreneur. He credits a clear goal with changing how he approached school.

Stebbings asks where that drive came from and describes his own physical injuries as a possible cost of pushing hard. Franceschetti says he does not know exactly what triggered the change, but once he has a goal he finds it hard to stop. He recognizes that the same intensity can make relaxation difficult.

He closes with an image of a man enjoying a meal alone without checking his phone. Franceschetti knows nothing about the man's life, but admired the apparent calm because it is something he struggles to find himself. Stebbings thanks him for the candid conversation.