← Back to list
Colossus (Invest Like the Best / Business Breakdowns)Podcast14 Mar 2023Source: joincolossus.comHost: Patrick O'Shaughnessy

Auren Hoffman - A Deep Dive on Data - [Invest Like the Best, EP.320]

In plain words

This episode explains why data businesses are slow to grow but highly profitable once established. Auren Hoffman categorizes data companies into four types and argues that pure data businesses are like eating broccoli—hard to sell at first, but extremely sticky once customers integrate the data into their systems. He highlights ZoomInfo (the only pure data company to break out in 20 years), FICO (its credit score became the universal standard for loans), and Bloomberg (clients can't quit its terminals). He warns that the most common mistake in data startups is trying to build both data and applications at once.

AI SummaryAI-generated · may contain errors · verify against the original

Auren Hoffman (CEO of Safegraph) delves into the data business on the program, proposing a 2x2 matrix to classify data companies, with core dimensions including "religious" (high stickiness) and "application-based." He defines great businesses as those with high gross margins, network effects, and l

~10 min full read · 7 sections
Deep Analysis

At a Glance

Auren Hoffman (CEO of Safegraph, former founder of LiveRamp) deconstructs data businesses using a 2x2 matrix, with the core dimensions being "Past vs Future" (Truth vs Religion) and "Data vs Application". He argues that pure data businesses are like "eating broccoli"—extremely slow early growth, very difficult to sell, but once a moat is built, they become highly sticky, high-margin long-term businesses. Over the past 20 years, the only pure data company that has truly broken through is ZoomInfo.


Theme 1: A 2x2 Classification Matrix for Data Businesses—Four Quadrants, Four Business Logics

Auren Hoffman argues that all data companies can be placed into a 2x2 matrix along two dimensions: the horizontal axis is "Past Facts (Truth) vs. Future Predictions (Religion)," and the vertical axis is "Pure Data vs. Application."

  • Truth-Data Quadrant (e.g., Experian's credit data): Provides facts about what has already occurred. Hoffman points out that this type of business tends to form a "winner-takes-all" structure—"Data is just facts; once you have it, you don't need 100 vendors providing the same thing." Typically, there are only 1-2 core players.
  • Religion-Data Quadrant (e.g., FICO): Makes predictions based on facts. Hoffman believes that "religion-type companies face extremely intense competition, and no one holds a high market share," but FICO is an exception—it has turned predictions into a "currency." "Once it becomes a currency, it is easier to become the dominant player. The FICO score is packaged into loan securitizations and becomes the universal language of transactions."
  • Truth-Application Quadrant (e.g., Bloomberg, Second Measure): Builds UI and solutions on top of data. Hoffman notes that these companies "turn hard-to-use raw data into beautiful interfaces, enabling non-technical users to perform analysis."
  • Religion-Application Quadrant (e.g., Verisk): Provides both predictions and process tools. Verisk originally started as a data cooperative (co-op) for insurance companies, where "all insurers benefit from its existence, but no single one could do it alone."

Hoffman emphasizes that small companies (revenue <$50 million) can only focus on one quadrant; large companies can deploy across multiple quadrants.


Theme 2: The "Slow" and "Sticky" Nature of Data Businesses — Why It’s Broccoli, Not Sugar

Hoffman uses the "broccoli vs. sugar" analogy to characterize the growth profile of data businesses: data companies grow extremely slowly in the early stages but exhibit exceptionally strong long-term stickiness.

  • Growth Curve: "Data businesses rarely grow as fast as SaaS. They grow slowly, but they can become very, very good businesses, maintaining high margins over the long term."
  • Customer Stickiness: Data is an "ingredient," not a "solution." Hoffman uses a baking metaphor: "We sell high-quality butter to pastry chefs. Good butter doesn’t guarantee a good croissant, but without good butter, you definitely can’t make a good croissant." Once customers integrate data into their models, the switching cost becomes extremely high.
  • Market Maturity: Most industries are still in the early stages of data utilization. "Even in the hedge fund industry, there are no more than 100 firms that are truly adept at using alternative data. In the real estate industry, fewer than 5 firms can use data effectively."

Hoffman points out that early-stage data businesses may appear to have poor gross margins — because purchasing external data is accounted for as "cost of goods sold" (COGS), even if it is a fixed cost. "At $5 million in annual revenue, a $2 million data procurement looks like a terrible business; but at $100 million, it’s just a rounding error."


Theme 3: The Moat of Data Companies – Market Share, Brand, and "Frequency of Change"

Hoffman argues that the moat of data companies comes from three unique dimensions: high market share, brand effect, and the "frequency of change" of the data itself.

  • Market Share: "In the data business, it makes sense to have only 1-2 players, because data has no UI differentiation. The higher the market share, the lower the customer acquisition cost—clients will choose you anyway."
  • Brand and Trust: Taking Twilio as an example, "As a developer, I don't know if Twilio is really better than its competitors; I don't even know the names of its competitors, so I use Twilio first." The same applies to data companies—the brand becomes the default choice.
  • Frequency of Data Change: "The more frequently data changes, the higher its value. Facts about the War of 1812 no longer change, so their value is low. Data is like the AWS business—customer spending is increasing, but the amount of computing they receive is also increasing significantly. The cost per data element continues to decline, making it extremely difficult for new entrants to compete."

Hoffman also emphasizes that "joinability" is key: the more easily data can be linked with other datasets, the higher its value. "Dollars, Unix timestamps, and stock tickers are all excellent join keys. If your industry lacks a good join key, creating and open-sourcing one could be highly valuable."


Theme 4: Common Failure Modes in Data Entrepreneurship and Founder Traits

Hoffman points out that the most common failure in data entrepreneurship is "trying to occupy multiple quadrants at once"; successful founders need the humility to "play a supporting role."

  • Failure Mode: "The most common way to fail is trying to do both data and applications. Starting an application business is already difficult; starting a data business is even harder. If you do both at once, you double the probability of failure. Wait until revenue reaches $300 million before adding another business line."
  • Founder Traits: Founders of data companies need to "put the customer at the center and themselves on the sidelines." "We are archivists, not constitution drafters. Madison and Hamilton were innovators, standing on pedestals; we are archivists maintaining the constitution, and no one knows our names."
  • The Trap of Moving from Applications to Data: Hoffman uses Salesforce as an example—it possesses vast amounts of data, but "every contract prohibits it from doing so. If it wants to change the terms, it needs to go back and renegotiate with each customer, which distracts the sales team and causes revenue to decline."

Three companies Hoffman recommends studying: Experian/FICO (understanding the credit data ecosystem), Bloomberg (the moat of high fixed costs plus bundling model), and ZoomInfo (the only pure data company to break through in the past 20 years).


Mentioned Positions

Position Analyst View Key Data
FICO Bullish ("phenomenal business") Converts predictions into "currency," becoming the universal standard for loan securitization
ZoomInfo Bullish ("the only pure data company to break through in the past 20 years") Strong brand effect—"I don't know if their data is actually better than competitors, I just use ZoomInfo"
Bloomberg Bullish ("increasingly incredible") High fixed costs + bundling model; "could fire the top 20 employees and the business would still be fine"
Verisk Bullish ("worth studying for everyone") Started as an insurance data cooperative, transitioned from non-profit to for-profit
Experian / Equifax / TransUnion Neutral (industry structure analysis) The U.S. has 3 credit bureaus (rare), most countries have only 1-2
Twilio Neutral (brand case) Brand becomes the default choice—developers don't know competitors' names
Stripe Neutral (analogy) High market share, customer acquisition costs keep declining
LinkedIn Neutral (data cooperative case) User-maintained database, but cannot directly sell data (trust issues)
CoStar Neutral (application layer case) Real estate industry data maturity is low, must build applications
G2 Bullish (Hoffman investment) "Yelp for software," application layer + proprietary data
Second Measure Neutral (acquired by Bloomberg) Built UI on top of Yodlee's raw data
Salesforce Risk warning (data monetization limited) Contract terms prohibit data sharing, cannot transition into a data business

Judgments Worth Remembering

1. Hoffman defines a "great business": one that would still perform well even if the top 100 employees were fired. The best example is Visa—"Even if my deceased grandmother were running it, it would probably still function well." Core characteristics: customer acquisition costs decline year over year, and high market share.

2. Data companies are "archivists" rather than "constitutional drafters"—they must accept a supporting role. "Madison and Hamilton stand on the pedestal; we are the archivists maintaining the constitution. No one knows our names, but we are still important."

3. The frequency of data change is a core metric for value. "Facts about the War of 1812 no longer change, so their value is low. Data is like AWS—customer spending increases, but the amount of computing they receive also rises significantly, while the cost per data element continues to decline."

4. The "joinability" of data determines its value ceiling. "The more connectable data is, the higher its value. The dollar, Unix time, and stock tickers are all excellent join keys. If your industry lacks one, create and open-source it."

5. The most common failure in data entrepreneurship: attempting to occupy multiple quadrants simultaneously. "Wait until revenue reaches $300 million before adding other business lines. Until then, stay extremely focused."

6. Early-stage gross margins for data companies may look poor—because data procurement is recorded as COGS, even if it is a fixed cost. "At $5 million in annual revenue, $2 million in data procurement looks like a terrible business; by $100 million, it is just a rounding error."

7. Most industries are still in the very early stages of data utilization. "Even among hedge funds, no more than 100 are truly adept at using alternative data. In the real estate industry, fewer than 5 can use data effectively."

8. Hoffman's "Four Nouns of Data" framework: all data companies revolve around one of four nouns—people, places, organizations, or products. "Almost all data companies center on at least one of these nouns, sometimes intersecting, and usually crossing with time or price."