What this covers

Patrick talks with Auren Hoffman, CEO of SafeGraph and founder of LiveRamp, about how data businesses are built and why they behave differently from software. Hoffman sorts them with a 2x2 and a taxonomy: every data company is organized around one or more of the Four Nouns of Data, meaning people, places, organizations, or products, sometimes crossed with time or price. Examples recur throughout: FICO turning a prediction into a currency, Verisk starting as an insurer-funded co-op, Bloomberg bundling datasets, and LinkedIn versus ZoomInfo as two routes from similar data.

The middle of the conversation is about economics and craft. Purchased data lands in COGS even though it behaves like a fixed cost, so year-one margins can look terrible. Hoffman argues the best businesses see CAC fall as share rises, and defines a great business as one that survives firing its top 100 people, as Visa would. He also covers data quality (marketing tolerated about 50% accuracy, while compounding models need 90% to 99%+), why fast-changing data is worth more, and why Joinability matters, including the case for creating a shared key where an industry lacks one. Data co-ops can form top-down or bottom-up.

The more contested material comes late. Hoffman forecasts that homomorphic encryption, differential privacy, and synthetic data will let parties join sensitive datasets without exposing them. He also argues that a few firms hoarding data, such as Google, Apple, and Tencent, suppresses innovation. Practical asides include seeding fake records as watermarks to trace leaks. The episode closes with advice on hosting a dinner party.

Sharpest takeaway

Hoffman argues data businesses are slow, fixed-cost, truth-oriented businesses that win through market share and joinability, and that founders should commit to one of the four data nouns and one part of the data-to-application spectrum rather than mixing them early.

  • Data is hard to manufacture and cheap to copy once made, so one or two providers tend to dominate
  • Data is an ingredient rather than a finished solution, so buyers need engineering maturity and businesses need UI or application layers to scale across industries
  • Mixing data and application businesses early multiplies failure risk

The claims · ranked71 claims · weighted by value

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.

0.86

Data companies protect against theft by seeding fake records as watermarks, like the fictional cities mapmakers inserted, so a leaked copy can be traced back to the customer that breached the agreement.

causalhigh valueestablishednovelty 3/4durability 4/4· Auren Hoffman

“So they essentially will have some fake data that they deliver to their customers. So think of it as some sort of like watermark that's in your data.”

0.73

Privacy technologies such as homomorphic encryption, differential privacy, and synthetic data now make it possible to join datasets, including medical and IRS records, without any party seeing the underlying data, which was not feasible when SafeGraph started about six years earlier.

forecasthigh valuecontestednovelty 3/4durability 3/4· Auren Hoffman

“today, you have things like homomorphic encryption, you have better understanding of differential privacy. There's a lot of ways to create synthetic data on top of data.”

0.68

When a small number of companies such as Google, Apple, and Tencent hoard underlying data without sharing it, innovation declines and the economic rents from innovation accrue to those few firms, which is a world nobody wants, so open data access is the remedy.

causalhigh valuecontestednovelty 2/4durability 3/4· Auren Hoffman

“if data really is the most important thing, if it really is fueling all this innovation then you're just going to see scenarios where we're just going to have a lot less innovation”

0.68

In most countries credit data is held by one monopoly bureau or two duopoly bureaus, while the United States is rare in having three major bureaus, Experian, Equifax, and TransUnion, and such data collection is very hard to scale beyond three.

factualhigh valueestablishednovelty 2/4durability 3/4· Auren Hoffman

“In most countries, you have either one, which is like a super monopoly or you've got two, which is some sort of duopoly. U.S. is actually the rare case where you really get to three.”

0.60

Leaders of data businesses must accept being on the side of the customer rather than at the center, like archivists who maintain the Constitution without writing it, which requires humility and a supporting mindset.

normativehigh valuespeaker onlynovelty 3/4durability 4/4· Auren Hoffman

“The data business is usually like one of those guys on the side and the customer is at the center.”

0.60

If an industry lacks a shared join key, creating one, through taxonomy and ideally open-sourcing it, can be very valuable, because joinability unlocks the data's power, as with Placekey for addresses.

normativehigh valuespeaker onlynovelty 3/4durability 4/4· Auren Hoffman

“If there isn't joint keys in your industry or your particular thing that you're looking at, then thinking about creating one and thinking about open sourcing it could be really, really valuable.”

0.60

Verisk began as a nonprofit data co-op in which insurance companies pooled claims data to catch fraud, because no single insurer had enough market share and insurers could not share raw data with each other without risking price fixing, so a neutral third entity had to be created.

causalhigh valuespeaker onlynovelty 3/4durability 4/4· Auren Hoffman

“They couldn't actually share data with the other insurance companies because then you have the scenario you could be like price fixing and other types of things, so you needed another entity to go do that.”

0.60

Selling data is hard because data is rarely the solution itself; it is an ingredient, like high-quality butter that is necessary but not sufficient for a great pastry, so buyers still need other inputs, tools, and capabilities.

causalhigh valuespeaker onlynovelty 3/4durability 4/4· Auren Hoffman

“Data is just an ingredient. We like to say we're selling high-quality butter to pastry chefs just because you have a high-quality butter, it doesn't mean you're going to end up with like this amazing croissant.”

0.60

Data must be nearly perfect for modern use: marketing could tolerate data that is about 50% accurate, but models that multiply inputs together compound errors so that data must be at least 90% accurate and often 99% or better to solve certain problems.

causalhigh valuespeaker onlynovelty 3/4durability 4/4· Auren Hoffman

“if you start times 0.9 against each other, just a few times, you're going to a really small number really quickly. So you often have to be at least 90% accurate.”

0.60

The more frequently a dataset changes, the more valuable it is, while data about stable historical facts like the War of 1812 loses value over time as it becomes common knowledge, and per-element value declines even as customer spending rises.

causalhigh valuespeaker onlynovelty 3/4durability 4/4· Auren Hoffman

“The more frequent it changes, the more valuable the data is.”

0.58

The most common way data companies fail is trying to be in more than one of the four quadrants at once, such as being both a data business and an application business, because doing more at once multiplies failure risk, and companies should focus until around $300 million in revenue.

normativehigh valuespeaker onlynovelty 3/4durability 4/4· Auren Hoffman

“The most common way, I think, by far, is to try to be in more than one of the four quadrants at once.”

0.58

All data companies are organized around one or more of four nouns: people, places, organizations, or products, and these can be crossed with time or price, which is why a stock ticker dataset is an organization crossed with time and price.

definitionhigh valuespeaker onlynovelty 3/4durability 4/4· Auren Hoffman

“there's really four nouns of data. there's data about people, there's data about places. So that's what SafeGraph does.”

0.58

A great business is one that would still work amazingly well if you fired the top 100 people and replaced them with almost anyone, as Visa would, because its economics rest on structure rather than talent.

definitionhigh valuespeaker onlynovelty 3/4durability 4/4· Auren Hoffman

“To me, a great business is you could fire the top 100 people in the company and the business is still amazing.”

0.57

It is much easier to start an application business than a data business, and data businesses are a tough slog, but once a company builds an application without deliberately planning a data layer, it becomes very hard to add a data business later because customer contracts may prohibit data reuse.

causalhigh valuespeaker onlynovelty 3/4durability 3/4· Auren Hoffman

“The problem is once you start an application business, it is hard to start the data business because unless you've really been thinking about it all the way through it”

0.57

Competitive moats in co-op-based companies are established before the companies become standalone for-profit businesses, which makes early co-op formation a source of later equity value that investors can exploit.

causalhigh valuespeaker onlynovelty 3/4durability 3/4· Patrick

“It's almost like the competitive moat or advantage to these companies gets baked before their stand-alone for-profit businesses.”

0.57

FICO converted a predictive religion product into a currency, and once a prediction becomes a currency it is much easier for one player to become dominant and packaged everywhere.

causalhigh valuespeaker onlynovelty 3/4durability 3/4· Auren Hoffman

“In FICO's case, they converted that religion into a currency. Of course, once you become a currency, then it's much easier to become the dominant player.”

0.57

Purchased data is typically recorded as cost of goods sold even though it is a fixed cost, so early-stage data companies can show very poor or even negative gross margins in year one, which investors must be taught to see as temporary as revenue scales.

causalhigh valuespeaker onlynovelty 3/4durability 3/4· Auren Hoffman

“that goes above the line as a COG even though it should be a fixed cost in many cases, it's a fixed cost. So it makes your margins early on look really, really bad.”

0.57

Bloomberg keeps getting stronger because it spends millions on each new dataset and then bundles it for all users, so each new terminal user finds a different subset valuable, making it very hard for competitors to replicate the bundle.

causalhigh valuespeaker onlynovelty 3/4durability 3/4· Auren Hoffman

“they kind of like Netflix or something where they have these very, very high fixed costs, every time they want to get a new data set, they might spend millions of dollars”

0.57

A predictive score becomes a currency when it is incorporated into lending and loans are packaged into tranches by score band for resale, which gives the score a standardized role in financial markets.

causalhigh valuespeaker onlynovelty 3/4durability 3/4· Auren Hoffman

“if you wanted to sell those loans, well, okay, you say these are the loans for this kind of tranche of FICO score”

0.57

The best businesses have customer acquisition costs (CAC) that fall every year, and the higher a company's market share, the more its CAC tends to decline, while many successful software companies have rising CACs that make them merely interesting rather than great.

causalhigh valuespeaker onlynovelty 3/4durability 3/4· Auren Hoffman

“The best businesses have decreasing CACs over time. So a lot of really great software businesses, their CACs actually have been going up every year.”

0.57

Co-ops can start top-down, as with Visa or Verisk where a few companies coordinate, or bottom-up, where many individual users contribute their own data, such as a DSO benchmarking co-op built from small businesses' accounting software.

definitionhigh valuespeaker onlynovelty 3/4durability 3/4· Auren Hoffman

“And it could start top down. So that's where like Visa or Verisk started where it's very, very top down and you have a small number of companies getting together and doing that. But it also could start bottoms up.”

0.57

LinkedIn and ZoomInfo have similar data but do not compete directly because LinkedIn cannot sell raw data bluntly without losing user trust, so it builds applications like advertising and recruiting on top of its data, while ZoomInfo sells leads directly.

causalhigh valuespeaker onlynovelty 3/4durability 3/4· Auren Hoffman

“LinkedIn can't really be just like a data selling business that's a little bit too crass, and people might lose their trust in LinkedIn.”

0.55

Data is very hard to manufacture and takes substantial work to create, but once created everyone can buy it, so data markets tend to be winner-take-most with one or two core providers, unlike software, which comes with a UI and allows more differentiated use cases.

causalhigh valuespeaker onlynovelty 3/4durability 4/4· Auren Hoffman

“data is very, very hard to manufacture and takes a lot of work to go do it. But then once you do it, everybody can buy it.”

0.55

The more a dataset can be joined to other datasets, the more valuable it is, which is why standard keys such as tickers, dollars for price, and Unix time for time became dominant, and why people quote prices in dollars even when trading in other assets.

causalhigh valuespeaker onlynovelty 3/4durability 4/4· Auren Hoffman

“the more this data could be joined to the more valuable it is, even if you quoted prices in ether instead of in dollars or something like that”

0.54

The higher a data company's market share, the lower its customer acquisition cost, because customers default to the leader; many companies spending heavily on sales and marketing could cut that cost by about 80% with little effect on sales.

factualhigh valuespeaker onlynovelty 3/4durability 3/4· Auren Hoffman

“Most of these companies that do spend a lot on sales and marketing probably could reduce their cost by 80% and barely that would even affect much of their sales at all?”

0.54

Some successful join keys are decreed top-down, such as the metric meter set by Napoleon, while others emerge bottom-up, and a powerful authority can impose a join key that works across a large region.

factualestablishednovelty 2/4durability 4/4· Auren Hoffman

“that was decreed by Napoleon that we were going to use this measure. Sometimes it's helpful to be the emperor all of Europe to decree your joint key”

0.53

A bottom-up co-op could pool accounting data such as days sales outstanding (DSO) from QuickBooks users, so that a buyer choosing between suppliers could see which counterparties pay slowly, for example General Motors at 90 days versus Ford at 30 days, which would pressure slow payers to improve.

forecasthigh valuespeaker onlynovelty 3/4durability 2/4· Auren

“I would love to see a co-op where, hey, you can off your QuickBooks”

0.53

Most companies are not ready to buy external data because they still struggle to use their own data; only the most sophisticated players such as Starbucks, Walmart, or Target are far enough along the maturity curve, and even large industries have few capable buyers, with about 100 hedge funds and fewer than five real estate firms able to use alternative data well.

factualhigh valuespeaker onlynovelty 3/4durability 2/4· Auren Hoffman

“there's really like 100 hedge fund tops that are really good at using alternative data today. And if you think of real estate, there's like less than five that really can use data.”

0.52

Product velocity, meaning how quickly a product changes, is a barometer of improvement, and a small company with product velocity ten times faster than an incumbent can eventually become competitive with it.

causalhigh valuespeaker onlynovelty 2/4durability 3/4· Auren Hoffman

“And so the way that small companies can compete in any scenario is velocity? Is there product velocity. If your velocity is 10x faster than the incumbent”

0.52

When evaluating a data vendor, a buyer should check whether the schema is public, whether bugs are transparently reported and fixed, and how much of the buyer's expensive engineering time the data will consume, favoring formats that push directly into systems like Snowflake or APIs.

normativehigh valuespeaker onlynovelty 2/4durability 3/4· Auren Hoffman

“Is the schema public? Is there a transparency around the data when they have bugs, how do I even let them know about the bugs and how do they fix the bugs.”

0.52

LinkedIn functions as a data co-op in which users keep their own profiles current because everyone else does, producing a valuable, mostly accurate database that the company can monetize for both consumer and business use cases.

causalhigh valuespeaker onlynovelty 2/4durability 3/4· Auren Hoffman

“Well LinkedIn is a data co-op essentially. The reason why I submit my data LinkedIn in is because everyone else is”

0.52

Because data lacks a UI, data markets can support one or two dominant players with very high market share, as with Stripe and Twilio, whereas SaaS leaders typically hold about 25% share because workflow and UI differentiation sustains competition.

factualhigh valuespeaker onlynovelty 3/4durability 3/4· Auren Hoffman

“In the data business, it can make more sense to have just one or two players because it's just data. So you don't have that UI side.”

0.52

The first axis for categorizing data businesses is whether they focus on past facts (truth), such as whether a loan was repaid, or on predictions about the future (religion), such as a credit score.

definitionhigh valuespeaker onlynovelty 3/4durability 3/4· Auren Hoffman

“So I think of this as truth versus religion.”

0.52

Most religion (prediction) data companies become good businesses but rarely great ones, because many competitors each build their own algorithm for the same question, such as fraud scores or climate disaster predictions, which makes a large sustainable business very hard to build.

causalhigh valuespeaker onlynovelty 3/4durability 3/4· Auren Hoffman

“Most of these religion companies do not become currencies, and they become good businesses, but they rarely ever become great businesses because there's so much competition”

0.52

Having hundreds of thousands of data sources is more valuable than relying on one, because some sources may be over 90% bad data but still contain valuable nuggets that can be sorted out, and the source count reduces dependence on any single provider.

causalhigh valuespeaker onlynovelty 3/4durability 3/4· Auren Hoffman

“we have sources that come in, which are over 90% of the data that comes in from these sources is bad, like the data is wrong.”

0.51

Second-time founders who want to have impact often pick problems they are passionate about, such as climate change or nuclear proliferation, but that passion does not mean they have the capability to address the problem, so they must honestly assess what they can actually help with.

normativehigh valuespeaker onlynovelty 2/4durability 4/4· Auren Hoffman

“it doesn't mean that their particular capabilities line up to solve that problem. So you also, I think, need to look yourself in the eye”

0.51

The most important property of data is that it is true; for backward-looking facts like historical stock prices, accuracy can be near 99.99%, and a data company should aim for 100% accuracy even though data changes make perfection unattainable.

normativehigh valuespeaker onlynovelty 2/4durability 4/4· Auren Hoffman

“the very most important thing about data is that it's true. If you're selling fax, if you're selling something that's backward looking, it should be true.”

0.50

A great dinner party appeals to both introverts and extroverts, and it is better to have one shared conversation than several side conversations, which requires good acoustics, a home setting, and a group of about eight people.

normativehigh valuespeaker onlynovelty 2/4durability 3/4· Auren Hoffman

“having one conversation is always better than having multiple conversations. So if you're just talking to your neighbor, it's not that interesting”

0.50

A dinner party needs a skilled moderator to keep one conversation flowing, and even with the best moderator the maximum group is about 14 people, with eight being ideal; moderators must draw in everyone, cut off dominant talkers, and add humor.

normativehigh valuespeaker onlynovelty 2/4durability 3/4· Auren Hoffman

“even with the best moderator, let's say, maybe the maximum is 14 and maybe the ideal number is eight.”

0.50

A dinner moderator must balance contributions, cut off people who dominate, and inject humor; dinners should end early, around three hours after starting, so guests leave wanting more.

normativehigh valuespeaker onlynovelty 2/4durability 3/4· Auren

“all dinner parts should end early. Keep people wanting for it.”

0.50

When buyers in an industry like real estate are not ready to use raw data, a data company may have to build an application on top of the data, which is why CoStar built applications rather than selling raw data to many customers.

causalhigh valuespeaker onlynovelty 2/4durability 3/4· Auren Hoffman

“So CoStar is an incredible company, and they've built applications because it'd just be too hard to just sell like raw data to a lot of folks.”

0.50

Data businesses usually do not own end demand, so the key question is whether the market is fragmented or concentrated and whether a single vertical or multiple verticals are being served, which affects go-to-market cost and the feasibility of selling raw data.

causalhigh valuespeaker onlynovelty 2/4durability 3/4· Auren

“Sometimes you're selling across industries. Those data businesses are going to be harder because you're going to have different go-to-market efforts”

0.50

A new data business should choose one of the four nouns to focus on, usually the one the founder is most passionate about, because each noun has trade-offs and attempting multiple at once increases failure risk.

normativehigh valuespeaker onlynovelty 2/4durability 3/4· Auren

“You have to choose to noun that you are most passionate about.”

0.50

Data businesses are the opposite of a sugar rush: they take much longer to build and grow slower than SaaS, but they are much stickier over the long term and can stay high margin for a very long time.

causalhigh valuespeaker onlynovelty 2/4durability 3/4· Auren Hoffman

“They take a lot longer to build. So they rarely grow as fast as SaaS companies, but they're usually much more stickier long term.”

0.50

Assigned seating and advance notice of the conversation topic help introverts thrive, because assigned seats remove the awkwardness of choosing where to sit and let people prepare thoughts before arriving, while cocktail parties are the worst format for introverts.

normativehigh valuespeaker onlynovelty 2/4durability 3/4· Auren Hoffman

“So just assigning someone you're sitting here, it's really easy. We're going to have one conversation, then it's just going to be so much better. These like cocktail things that are out there, cocktail parties are just the death for introverts.”

0.49

SafeGraph covers just over 40 million physical places worldwide, roughly the same count as the world's population, but if trees and other small features are counted as places there are far more places than people, so 40 million is not nearly enough.

factualhigh valuespeaker onlynovelty 2/4durability 2/4· Auren Hoffman

“Today, we cover a little bit more than 40 million places around the world. And then we want to update. We want to make sure we can update those places at least every month”

0.49

Data businesses are starting to look more attractive because more buyers now want to buy external data and modern tools like Snowflake and Databricks mean buyers do not need to be elite engineers to ingest data.

forecasthigh valuespeaker onlynovelty 2/4durability 2/4· Auren Hoffman

“You're seeing a lot more buyers that are looking to buy data today and they weren't looking to buy data 5 years ago”

0.49

Data companies prioritize which columns to add by asking customers what they want, checking competitors and data quality, estimating build cost, and favoring cheap, quick additions when the cost of capital is high because millions in investment must be recouped.

normativehigh valuespeaker onlynovelty 2/4durability 2/4· Auren Hoffman

“maybe I don't want to invest millions of dollars for this particular thing or maybe this is something I can do for hundreds of thousands of dollars”

0.48

In the last 20 years the only pure data business to become huge is ZoomInfo, while roughly 1,000 SaaS companies became unicorns, so being an entrepreneur or investor in SaaS has been far better than being in a pure-play data business.

factualhigh valuespeaker onlynovelty 3/4durability 2/4· Auren Hoffman

“in the last 20 years, the only data business that's really, really broke out pure data business that has become huge is ZoomInfo. So there's been like 1,000 SaaS companies that have been unicorns”

0.48

AOL's release of anonymized search history to academia showed that simple anonymization fails, because individuals were quickly de-anonymized, which made the risk of exposing sensitive health information from search data very real.

factualestablishednovelty 1/4durability 4/4· Auren Hoffman

“AOL, they released all their search history to academia. And they try to anonymize the people who are doing the searches, but very, very quickly, people were able to de-anonymize”

0.47

Every company sits somewhere on a continuum between pure data provider and application that uses data as part of a solution. Small companies under about $50 million in revenue can only occupy one place, while multibillion-dollar companies can run several bets along it.

definitionhigh valuespeaker onlynovelty 2/4durability 3/4· Auren Hoffman

“If you're a small company, let's say, below 50 million in revenue, you can really only be a one place, in my opinion, on that.”

0.47

Truth-application companies take data that is hard to use, like raw credit card transaction streams, and build a usable interface on top, so non-technical users can extract insight without strong engineering teams.

definitionhigh valuespeaker onlynovelty 2/4durability 3/4· Auren Hoffman

“Sometimes they're just taking data, which is harder to use and then they're building some sort of UI on top of that data.”

0.47

Data businesses resemble AWS: customer spend rises while cost per data element keeps dropping, which makes them winner-take-most because new entrants cannot match the falling unit cost of the incumbent.

causalhigh valuespeaker onlynovelty 2/4durability 3/4· Auren Hoffman

“this is why some of these data businesses become winner-take-most because the cost per gate element keeps dropping over time.”

0.47

Any data business has three steps: ingesting data from many sources, reducing and cleaning that data into something accurate (where machine learning can fill small holes and resolve conflicts such as two different opening times), and delivering it to customers with credible documentation.

definitionhigh valuespeaker onlynovelty 2/4durability 3/4· Auren Hoffman

“If you think of data, there's often like three steps to any type of data business. There's the first step where you kind of ingest”

0.45

If a data company collects and manufactures its own data, it looks more like a SaaS company on the income statement, whereas buying raw ingredients in bulk produces a COGS-heavy profile that is hard to read.

factualestablishednovelty 2/4durability 3/4· Auren

“And in that case, it kind of goes below the line and maybe your business looks a little bit more like a SaaS company.”

0.44

The data truth quadrant has the fewest new companies because the market is hard to understand, attracts less venture capital, and has few large outcomes beyond ZoomInfo.

causalhigh valuespeaker onlynovelty 2/4durability 2/4· Auren Hoffman

“that data truth quadrant, we just have the fewest number of companies being started. I think there's good reasons because the market is not as good in that quadrant.”

0.44

SafeGraph covers just over 100 attributes per place, while roughly 10,000 relevant attributes exist, so it covers only about 1% of the total attribute space, and a go-to data company must expand both rows and columns.

factualhigh valuespeaker onlynovelty 2/4durability 2/4· Auren Hoffman

“today, we cover a little over 100, let's say, attributes about a place. That's interesting. That's a start, but there's probably 10,000 relevant attributes.”

0.39

Data about constantly changing facts, such as store hours, can be true one month and false the next, so a data company's goal is to approach truth while accepting that it can never be perfect, and it must update places roughly monthly.

factualestablishednovelty 1/4durability 3/4· Auren

“In SafeGraph we have data about store hours. So even if we were true a month ago, that doesn't mean we are true today.”

0.34

Dinner parties should be off the record so that people can play with unfinished ideas, because in a world where everything goes on Twitter, people need a safe space where they assume positive intent and can refine ideas without fear of judgment.

normativespeaker onlynovelty 2/4durability 3/4· Auren Hoffman

“in this world that we live in, where everything is on Twitter, it's very unsafe to play with an idea”

0.34

The kindest thing anyone did for Hoffman was a negative example: a highly unethical, charismatic person whose orbit he fell into as a teenager, and whose life he now uses as a model of what not to become.

factualspeaker onlynovelty 2/4durability 3/4· Auren Hoffman

“to this day, I actually often think about this person as what not to become. So in some ways, I think the kindest thing someone has done for me, it's the lesson that the least kindest person I've ever met gave me.”

0.34

Selling data is very difficult because it requires different salespeople than typical SaaS, longer sales cycles, and ongoing customer success work, since data vendors often do not know how customers use their product.

factualspeaker onlynovelty 2/4durability 3/4· Auren Hoffman

“One thing is it's just really hard to sell. So selling data is really, really difficult, and you need a different salesperson often that are selling typical SaaS”

0.32

Hosts should introduce each guest in advance and send all attendees contact information the next day, because people forget names after a dinner and reaching out afterward turns a one-time meal into lasting connections.

normativespeaker onlynovelty 2/4durability 3/4· Auren Hoffman

“after people go to the dinner the next day, send them everyone's contact information.”

0.30

Transparency about supply chains, labor rules, and data quality is becoming required for businesses because customers cannot verify every fact and want to see how a company is changing over time.

normativespeaker onlynovelty 1/4durability 3/4· Auren

“I think transparency is becoming really important.”

0.30

Brand becomes important for data and middleware companies only after they reach scale, as developers default to a known provider such as Twilio without knowing its competitors, though early-stage companies cannot yet build much brand equity.

causalspeaker onlynovelty 1/4durability 3/4· Auren Hoffman

“So I'm probably going to try Twilio first, then maybe if they're really having a problem and they're not servicing my needs”

0.29

Food is the least important element of a dinner party because guests remember the quality and content of the conversation rather than the meal.

normativespeaker onlynovelty 2/4durability 3/4· Auren Hoffman

“And the one thing that is probably the least important thing in dinner parties in my opinion, is the food. Nobody remembers the food in the end.”

0.29

Product data is the hardest noun to join because, unlike books with ISBN numbers, food products have no shared identifier, so a hamburger patty recorded differently in two datasets cannot easily be matched.

factualspeaker onlynovelty 2/4durability 3/4· Auren

“for books, there's like an ISBN number. But for most cases, it's very, very hard.”

0.28

Market makers such as Citadel Securities may function somewhat like co-ops, which suggests that investors can think about co-op-like structures beyond simply buying co-op equity.

factualspeaker onlynovelty 2/4durability 2/4· Auren

“It's possible that some of these market makers look a little bit like co-ops. So like the Citadels of the world could look a little bit like a co-op.”

0.26

People are generally good and want to be inspired and to do the right thing, but modern life often lacks inspiration, and seeing someone truly excel at their craft, even a tour guide, is a reliable source of enthusiasm.

normativespeaker onlynovelty 1/4durability 2/4· Auren Hoffman

“I do think that people are generally good. There's a small number of really bad apples out there. But I think people are very good and they want to be inspired.”

0.26

Google competes with almost every company on almost everything, so a company that does not compete with Google effectively does not exist, because Google wants all information and users on its platform.

factualspeaker onlynovelty 1/4durability 2/4· Auren

“Google is in everything business. So Google competes with everyone about everything.”

0.23

The most impactful people to host at a dinner are religious leaders such as Jesus and Muhammad, because they could explain their core ideas, and AI might one day enable such conversations.

normativespeaker onlynovelty 1/4durability 1/4· Auren

“One of the most impactful people in the world are religious leaders.”

0.21

SafeGraph's largest customer category is local search, and it also serves logistics for warehouses and power plants, retail site selection, out-of-home advertising, and macroeconomic analysis of store openings and closings.

factualspeaker onlynovelty 1/4durability 2/4· Auren Hoffman

“But the biggest industry for us is local search. So I want to search for cafes near me or something like that”