Michael Mayer - Business Boss Battles - [Founder’s Field Guide, EP. 40]
What this covers
Patrick talks with Michael Mayer, founder of Bottomless, a company that automatically restocks coffee, about the tactical decisions behind building it. The starting point is that replenishing consumables is broken because sellers can't see when a household needs more. That makes the product a problem of demand legibility. Mayer explains why computer vision fails in messy homes where coffee bags hide their fill level, and why weight works. He argues for stock over flow measurement: a single missed flow event leaves a permanent error. He also says he would now start from a distribution trick and work backward to a product, and he explains why households make better first customers than restaurants that already run manual inventory.
The middle of the conversation is about money and operations. Mayer describes a failed early raise, made with about twenty unknown customers, and the later approach of treating the company as a series of proofs. He credits anonymous, interest-driven Twitter posting with building social capital that cold outreach couldn't match. Production moved from outsourced electronics and 3D-printed shelves to hand assembly and then a mold set up in China. On the supply side, USPS's own delivery estimates ranked only about fifth among the inputs in the team's predictive model.
The talk then widens. It covers a hiring process built to resist the substitution heuristic, the contested prediction that remote international hiring will be a major trend within ten years, the limits of founder management at roughly 20 to 30 people, the status cost of leaving a well-paid job, and how a person's information diet shapes their thinking.
Mayer argues that startups succeed by solving only the bottleneck directly in front of them, iterating on the real problem with cheap, mature technology, and staging proof of each risk, rather than planning for distant scale.
- Choosing the right unit of legibility (weight of stock, not vision) determined how the hardware had to be built
- Warm social capital built through public Twitter presence succeeded where cold fundraising failed
- Premature solving of future bottlenecks is a deadly mistake; the web stack and 3D-printed shelves were good enough for years
This asset isn't compiled yet
You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.
A hiring process with separate scored interviews, a work sample, and a pass threshold at each stage helps prevent substituting the easier question of whether one likes the candidate for whether they will succeed.
“You substitute an easier problem. Is this person going to be a great engineer on our team? You may substitute the problem of like, do I personally like this person?”
Talent is distributed fairly evenly around the world but opportunity is not, so remote international hiring gives access to highly talented engineers who would otherwise be stuck in unsatisfying roles, and Mayer predicts this will become a major trend within ten years.
“talent may not be evenly distributed, but opportunity is less evenly distributed than talent.”
Trying to solve a future bottleneck prematurely is a huge mistake; solving only the bottleneck immediately in front of you keeps the team working on real problems and avoids fake problems that can be deadly for a startup.
“Trying to solve a bottleneck prematurely is a huge mistake, and very tempting.”
The company never planned for 50,000 units; instead, at each step the founders solved whatever bottleneck was directly in front of them, such as outsourcing electronics assembly while keeping 3D-printed shells, which kept work tied to a real problem.
“What we did is we never really made the leap of saying, 'We're going to make 50,000 units'. We just solved whatever problem was right in front of us”
Investors require proof beyond the founder's own certainty; what it takes to get a yes is many times more than what it takes for the founder to feel confident, and founders cannot project their psychology onto investors.
“You really cannot project your own psychology onto an investor's psychology.”
Vision sensors that track items moving in and out miss events and become permanently off by one, so the right approach is to monitor the stock (how much remains) rather than the flow, and weight provides clean ground truth for stock.
“You need to monitor the stock, not the flow. That's some insight that we came up with originally”
Replenishing consumables is broadly broken because sellers have no reliable visibility into when a household or business needs more, so the core problem of restocking is one of legibility of demand.
“The core problem of restocking is legibility.”
Whatever information a person consumes generates thoughts automatically, so consuming sports, news, or unproductive feeds can later surface in idle moments such as showers, which undermines focus on what one wants to achieve.
“any information that you put into your mind is going to generate thoughts and the thoughts that occur to you are basically automatic.”
If starting over, Mayer would begin with a distribution trick and work backward to a product that fits it, because even a strong product still faces a constant customer-acquisition challenge.
“I would definitely start with distribution first and work backwards to the concept.”
Being one of the roughly 500 accounts an investor follows can be worth more than a prestigious credential such as a Stanford degree, because the investor already has a personal memory of the founder.
“being one of those 500 is way more important than being a Stanford graduate”
A business-first product fails when the target already has a process, since partial automation adds little marginal value; households without a designated restocker are a better first customer than restaurants that already run manual inventory.
“unless you solve the whole problem, they're still going to be doing inventory every day.”
Anonymous, interest-driven Twitter posting built a network of investors who already knew the founder, turning a social audience into a fundraising asset that cold outreach could not replicate.
“Building your own social network is definitely quite a bit more powerful than I think people understand.”
Once customers signed up nationwide, the team found that the supplier side, specifically the time it takes for product to be produced and shipped, was as important as demand legibility, because delivery timing determines whether a product arrives before running out.
“this supplier legibility was equally as important as the demand legibility.”
A founder's personal performance and management and time-management skills must keep growing with the company, since each stage revealed the founder was not yet equal to the next level, a painful but necessary process.
“my personal performance as a founder was not up to the standard necessary almost at every step of the way.”
Tracking how much product remains (stock) is more reliable than tracking consumption events (flow), because a single missed flow event leaves a permanent error that must later be reconciled.
“You need to monitor the stock, not the flow.”
Because the chips come from the smartphone supply chain, weight sensors and web stacks are mature, almost anything that already exists can be figured out by stitching mature technologies together, but truly novel invention such as biotech drugs would be different.
“Almost anything that already exists, you can figure it out.”
USPS's own delivery time estimates proved nearly useless as inputs, and the estimate ranked only about fifth in importance among the model's parameters in the team's predictive model.
“the USPS estimate is like the fifth most important parameter. It's not even close.”
For a startup, the eventual outcome is the endpoint of a vector whose direction, chosen early, accounts for about half the outcome, while effort, luck, and partners supply the magnitude.
“The idea is the direction of the vector. The magnitude is how hard you work, how lucky you get, the partners that you have.”
Computer vision fails for home restocking because home environments are too messy and coffee bags hide their fill level, so only an automated weight measurement works without requiring users to manually reposition items.
“vision only works for items, where you can count the proportion of ideal stock in some way that it's obvious by looking at it.”
Mayer's team treated the company as a series of proofs, each validating one specific risk: first whether the concept works, then whether strangers would buy, then whether retention holds, before raising capital.
“We always saw the company as a series of proofs, sort of proving out risks.”
Raising too early, with about 20 unknown customers and a year of product work, failed completely, and that premature attempt set the company back, because the founder's confidence did not transfer to investors.
“And that was probably the biggest mistake in the history of a company.”
Production bottlenecks were solved in sequence: outsourcing electronics to manufactured boards, keeping 3D-printed shelves, then hand assembly, then flying to China to set up assembly and convert the 3D print into a mold.
“each step of the way we just solved the bottleneck.”
Hiring is fundamentally a legibility problem because there is no reliable source of truth for predicting how well a candidate will fit or perform, so the indicators are very poor.
“hiring really is a legibility problem. You really don't know.”
Quitting a well-paid developer job to start an unfunded company costs a large status hit, and founders need low sensitivity to others' opinions to tolerate it, a trait Mayer says he had to consciously adopt.
“I had to have fairly low tolerance or low sensitivity to status in order to even do it, to begin with.”
Society should treat starting a company and taking risk as high status, more than a junior investment banking role, because the subsidy of status offsets the very high personal risk of early-stage entrepreneurship.
“I honestly think starting companies and taking that risk and trying to build something new should be high status.”
A two-founder company can manage about five to ten people directly, but beyond roughly 20 to 30 people the founders must delegate and build organizational processes, because desire to get things done does not scale.
“desire just doesn't scale.”
Replenishment of consumables is broadly broken, but addressing it across restaurants, households, and industrial companies at once would be an impossibly broad mission, so the founders had to narrow to something they could actually build themselves.
“replenishment of consumables was broken broadly and that would have been sort of an impossible space to tackle”
A self-taught software developer with no hardware training built a working scale prototype by iterating through tutorials and examples, showing that the iterative learning approach from software carries over to hardware.
“It was just a process of looking at a tutorial, making my own example and iterating on it until I had something that actually worked.”
Bottomless does more productive data science than many companies that brand themselves as AI, and it does not market itself as an AI company, which reflects focusing on the end goal rather than the technique.
“nowhere in their materials does it say Bottomless is an AI company.”
Removing sports, news, and then Twitter from his feed was the highest-return personal change Mayer made, leaving him more focused, and he recommends pruning inputs to those that build useful mental models.
“for me, probably the highest ROI thing that I've done is slowly peel out.”
Vision-based restocking only works if users manually rearrange items so cameras can see them, and people will not do that, so the solution must be fully automated or people will keep their existing habits.
“You need something that's actually automated or people are going to stick with what they're already doing”
Posting about genuinely interesting topics without the goal of fundraising produced an audience that later helped fundraising; the benefit came from giving value to the network, not from strategic self-promotion.
“I was trying to tweet things that people enjoy, and I was trying to provide a service to the network.”
Early signs that friends used the product were insufficient to convince skeptics, so the founders needed unknown paying customers who kept buying, which made them move from friends-and-family tests to paid online ad acquisition.
“We acquired 10 customers that we had never met using online ads that we had purchased ourselves. And to our surprise, these people actually used it more than our friends and family”
The median income of the destination zip code is highly predictive of how long a package takes to arrive, though the model shows correlation without causation and the direction of the effect is unclear.
“The income of the zip code, where the coffee's going to. It's highly predictive of how long it's going to take.”
Cold investor outreach from unknown founders is largely ignored regardless of data, so the founder's existing relationship and reputation with the investor matters more than the metrics alone.
“who you are to these people matters a lot.”
Strong business problems are fractal, offering new levels of depth as one zooms in, so a narrow wedge like consumer coffee subscriptions contains an effectively infinite problem set to keep solving.
“The best, most interesting problems are just fractal.”
The device had to run at least six months on power, because a battery lasting only a month would require monthly recharging that users would not tolerate just to restock coffee.
“who wants to recharge something every month just to restock their coffee? We had a target, but it had to be at least six months”
Mayer used a confidence interval on about 20 customers' six-month revenue retention, finding even the lower bound exceeded typical subscription retention, which convinced him to fundraise, though the small sample was a limitation.
“at the 99% bottom of that confidence interval, we were still above what I knew to be the revenue retention for a normal subscription.”
A server Mayer built as a throwaway prototype, without server experience, still runs essentially unchanged at about 10,000 current connections, showing that mature web tooling can handle far more load than feared.
“now we have, what? 10,000 current connections or something, and it's still essentially the same server.”
The perception that starting a company is high status and easy to fund applies only to a narrow group who already have substantial social capital, such as Stanford CS graduates, so the typical founder still takes a status loss.
“The perception of it being high status and it just working very quickly for people is only happening for a very narrow subset of the population.”
Patrick notes that the Nintendo lesson of building new products from mature, withered technology parallels the company's approach of stitching together existing technology in new ways.
“combine old pieces of technology. It's like that Nintendo story, new products with withered technology”
Restaurants resupply by having staff walk through every item daily, write counts on paper, and compare them to par values before reordering, yet they still run out of stock constantly.
“They have somebody walk through every single item in the restaurant and put it on like a little piece of paper every single day and compare it to a par value and reorder every day”
Products should minimize user effort so that the customer never has to think about the process and the product simply arrives when needed, which Patrick and Mayer treat as the guiding goal for the scale.
“You kind of want the user to never think about it and never have to do anything and just have the thing show up when they need it.”
The company's north star is making the user never think about restocking and have the product show up when needed, so convenience justifies going to extreme lengths to keep users out of the process.
“You kind of want the user to never think about it and never have to do anything and just have the thing show up when they need it”
Silicon Valley's founders and investors were already talking more online than in person, so relationships formed through social media precede and can substitute for physical proximity.
“Silicon Valley had been uploaded into the cloud already. And all of the investors and all of the entrepreneurs were already online, talking more than they were in person”
Early on the team manually checked every in-progress order daily on the USPS portal, and then built a custom tracking system because e-commerce tracking lacks any predictive delay detection that the model needed.
“We've actually built out a custom system for tracking orders as they're moving through the postal service.”
The narrower the focus, the better a creator performs online, a lesson Patrick says is repeatedly taught and which Mayer's niche Twitter experience supports.
“The more narrow your focus, strangely the better you do on the internet.”
After a successful pre-seed round built on warm investor leads, the company went through Y Combinator, where many people already knew the founder from Twitter.
“After this pre-seed round, we ended up going through Y Combinator.”
A fixed delivery-time constant in a replenishment model guarantees being wrong, since shipping times vary between roughly two and seven days, so the model must predict arrival day for each order.
“If it's going to be seven and we have a fixed constant of two in the model, we're always going to be late, and vice versa.”
The company monitors whether sellers pick up coffee every day, since pickups do not always happen on time, and calls sellers when a pickup is missing to find out where the coffee is.
“we actually are monitoring that the coffee's picked up and every seller every day, and we know what time they typically pick up because it doesn't happen”
The early founders spent a full day making five scales, sanding and painting them, until they learned customers did not care about finish, showing that early polish effort was misdirected.
“Eventually we dropped that when we realized nobody cared because we would put so much time and effort into these scales.”
E-commerce companies rarely use delivery-stage delay data to detect likely non-delivery and proactively replace orders, which Mayer says should be a basic building block of e-commerce.
“they should know, 'Okay, if it's delayed three days at this step specifically, that it's unlikely to be delivered and send you a replacement.' But nobody does that.”
Because cold applications arrive daily, Mayer frames hiring as a decision-making and sorting problem rather than a top-of-funnel sourcing problem, and this difference shapes how they design hiring.
“more as a sorting problem, decision-making problem than a top of funnel problem”
After investors rejected 15 to 20 customers as proof, the founders decided they needed about 150 customers, and turned the apartment into a hardware assembly line to reach that level of belief.
“We have to get 150. And that's what we did.”
Each solved bottleneck exposes the next, and Mayer's team has been moving up the chain, so future success may be about arranging an array of bottlenecks to solve rather than planning the whole path.
“you've gotten better and better at sort of almost leveling up the bottleneck”
Some countries may have only about 50 people employed in interesting startups, while the US has around 10,000 venture-backed startups running at any time, so the US remains the center of talent but has much richer opportunity.
“there might only be 50 people employed in interesting startups in some countries. Whereas in the US we have like what? 10,000 venture backed startups”
Founders should describe the end goal rather than the technology they use, and crash through bottlenecks repeatedly, because the means are secondary to what the company accomplishes.
“They're like, don't talk about the means. Think about the end that you're trying to accomplish, and just crash through bottlenecks over and over and over again”
The founder tweeted to improve his followers' feeds rather than to promote himself, treating the account as a service to the network, and believes generosity in public sharing pays back indirectly.
“I was trying to tweet things that people enjoy, and I was trying to provide a service to the network”
Personally doing two to three hours of work per customer made it impossible to build the platform, e-commerce, onboarding, and vendor relationships at once, a constraint that persisted for years.
“if you need to personally do two to three hours of work per customer, it just becomes extremely difficult”
The team narrowed scope step by step from industrial restocking to consumer households, then to a single product, and finally to premium coffee, because the founders were personally always out of fancy coffee and each step was testable.
“Consumer first and then to single product next and then to coffee specifically.”
When the founder gave friends a homemade scale and saw their weight graph jump after an order, he realized weight data could trigger ordering, an insight he had not intuitively anticipated.
“I still viscerally remember the time when I saw their weight graph, jump up”
The company launched only in the Seattle area to keep shipping simple, assuming a fixed two-day delivery, but signups quickly came from across the country, exposing that shipping variability was a core problem.
“When we started, we only started in the Seattle area to keep this problem easy”