Chaos Monkeys by Antonio García Martínez

An insider’s story about Facebook in the years leading up to its IPO. It’s surprisingly candid — it names names and exposes internal Facebook discussions that were never meant to be public.

An engaging read, but the narrator is painfully obnoxious.

What I Liked 🔗︎

• Story is well-written and engaging
• Interesting to see such a revealing view inside Facebook
• Does an excellent job of explaining technology and finance terms that are approachable to the layperson

What I Disliked 🔗︎

• I found the author pretentious and abrasive
  • He boasts about getting out of a ticket after being pulled over for drunk driving.
  • “I got down to the serious business… of trying to bang my product marketing manager [his Facebook colleague].”
  • He seems proud of the fact that he was so dedicated to his startup that he neglected his responsibilities in raising his children.
  • He spends several paragraphs explaining why his tastes are too sophisticated to know who Lady Gaga is when she shows up at his office.
• Author took part in plans to monetize user data and is irritatingly dismissive about the implications on user privacy
  • Paints what feels like a disingenuous representation of Facebook as brave guardians of user data. He claims that advertisers are the ones that are the real bad guys as far as privacy is concerned.
• Lots of “hustle worship.”
  • He sneeringly declines a job offer from Twitter because they encourage their employees to go home at dinnertime, whereas Facebook encouraged all-night hackathons and weekend work.

Key Takeaways 🔗︎

• You can understand credit default swaps by thinking of them in terms of car insurance
  • Credit default swaps are insurance against someone failing to pay you money they owe. If you buy a bond from someone and it defaults, the seller of a credit default swap will reimburse your losses. Similar to how if your car gets damaged, the insurer will pay for you to get it back to normal state.
  • The difference with credit default swaps is that anyone can buy the policy, so it would be like if your neighbor could buy an insurance policy and collect money if your car got damaged.
  • Further, with credit default swaps, anyone can sell a policy, so it would be like if a person across town could sell an insurance policy for your car.

• an incumbent in a market dominated by a few, with total information asymmetry, and the ability to make prices on the market rather than just take them, has little incentive to increase transparency.

  • Explains why Goldman-Sachs preferred to offer credit default swaps through its own private channels rather than bringing them to public exchanges.
  • Also explains why Facebook never set up its own ad exchange
• Early 2000s advertising on the Internet was primitive
  • You couldn’t target which users see your ads and you had no information about whether viewing your ad led to a purchase
• Right Media (acquired by Yahoo! in 2007) created the first programmatic media buying system (ad exchange)
  • It was the first system where ad purchases happened via automation rather than people talking to each other and making an individual deal for an advertisement.
• Programmatic ad buying shifted power from publishers to advertisers.
  • Advertisers now had more information about the users because the advertisers could track users across the web, whereas the publisher only saw the user on their own site.
  • This imbalance of information gave the advertisers the ability to set prices for advertisement when previously the publishers controlled the price.
• “Brand advertising” vs. “direct response”
  • direct response advertising is designed to make the customer take immediate action in response to seeing an ad (e.g., get $5 off this pair of sneakers if you click this ad)
  • brand advertising is designed to create a more long-term awareness of the brand rather than perform some immediate action (e.g., Pepsi spends millions on a Super Bowl ad with the expectation that any resulting purchases are only loosely connected to viewing the ad).
• If you say that you’re in “stealth mode,” VCs will think that you’re an amateur
  • Ditto if you ask them to sign an NDA
• How many miracles have to happen for your startup to succeed?
  • If it’s zero, it’s just a regular business (*e.g., a trucking company)
  • Startups require miracles but can only depend on 1-2 to have a reasonable shot at success
    • i.e., if your startup can only succeed if several miraculous events occur, you’ll probably fail
• Return on advertising spend (ROAS)
  • Basic metric that marketers use
  • If you have to spend $0.75 per click on an ad, but you earn $1.10 per click, your ROAS = ($1.10 / $0.75) - 1 = 47%
• As of 2011, highest-cost search term was “mesothelioma” at $90 per click
  • Bid up by lawyers trying to gather plaintiffs for class-action lawsuits related to asbestos
• In 2011, only large companies could buy advertising through Google’s Ad Exchange (AdX)
  • Google’s native tools were too crude and hard to use for big marketers, so they built their own tools on top.
  • Google’s native tools were too confusing for small businesses without marketing expertise.
• AdGrok (author’s startup) was a layer on top of AdX that small businesses could use
• AdGrok messed up by dividing equity equally among three co-founders
  • Hard to make any decisions because three people have an equal say.
  • Author recommends having a single decider with >=51% equity so that important decisions don’t get bogged down with everyone’s input.
• AdGrok was successful at content marketing mainly by posting provocative and controversial opinions about the tech scene.
  • e.g., “New York will always be a tech backwater, I don’t care what Chris Dixon or Ron Conway or Paul Graham say”
• “Chaos monkey” is an open source tool from Netflix that tests a system’s resiliency to random outages.
  • Imagine a monkey loose in a data center randomly disconnecting servers.
• “the cap”
  • A startup’s first round of funding is the “seed round.”
  • In exchange for investing during the seed round, investors receive a convertible note
    • Convertible note is debt that turns into equity in the company during later funding rounds.
    • e.g., If an investor puts in $100k of seed money, they get $100k of equity when the series A funding happens. If the series A raises money at a $10M valuation, the investor gets 1% equity in the company ($100k / $10M = 1%)
    • That’s a bad deal for seed investors because they invested earlier in the company’s lifetime when the risks were higher, so the “cap” corrects that imbalance.
    • Cap is the maximum valuation at which the seed money can be converted to equity.
      • e.g., $100k of seed money with a cap of $3M means that if the valuation by series A is higher than $3M, the seed investor still gets equity proportionate to a $3M valuation (e.g., valuation at series A is $10M but cap is $3M, the seed investor still gets 3.3% of the company for their initial $100k investment ($100k / $3M = 3.3%)).
• At the seed stage (raising money with debt), different investors can invest at different valuations (different caps), but when raising with equity (e.g., series A), all investors must agree to a single valuation.
• “cap table” is the capitalization table.
  • A table that lists every equity owner in a startup and what percentage ownership they hold
• Paul Graham seems like a great guy
  • When a larger company initiated a frivolous lawsuit against AdGrok, Graham used all of YC’s resources and connections to force them to drop the suit.
  • There’s also this scene with Paul Graham and Jessica Livingston that I found cute for how it humanized these seemingly larger-than-life characters:

    Before we got too deep into the conspiring, Jessica, PG’s wife and YC copartner, came out to discuss lunch. What ensued was a minor squabble over some leftover pasta, and why it was gone, and who had planned on eating it that afternoon. PG looked a little miffed.

• Chris Sacca comes across badly - petulant and self-serving.
• In an acqui-hire, two separate companies can buy a startup by picking and choosing the employees they want
  • In AdGrok’s case, Facebook acqui-hired the author, while Twitter acquired the company itself and its two developers.
• In an acqui-hire, the acquiring company has a lot of freedom to structure the deal so that it benefits the founders, the investors, or splits the benefit evenly.
  • The acquiring company just wants to do the deal for the lowest possible cost per employee. They don’t care whether the money goes to the investors or the founders.
  • To favor the founders, the acquiring company can move more money into the employment compensation packages.
  • To favor the investors, the acquiring company can pay more for the equity of the acquired company.
  • In AdGrok’s case, a $10M acquisition, Twitter offered $9M in employee compensation packages and only $1m for the company itself.
    • This was a bad deal for investors, who expected at least $2M.
• Nobody owns startups when they first form
  • Startup ownership is typically structured so that founders split ownership but their stake is vested over time.
  • If each founder owned 30-40% of the company from day one, any founder could instantly kill the whole company by walking away with their share of the equity.
• Facebook didn’t monetize well until ~2013 (shortly after their IPO)
  • “[Facebook’s per-user revenue was] comparable to what you’d monetize your Star Wars blog at if you ran AdSense.”

• a billion times any number is still a big fucking number.

  • Despite poor monetization, Facebook had high revenue by virtue of its enormous user base
• Pre-IPO, the Ads team didn’t have much power and was forced to try to squeeze money from whatever the product teams built (without having any voice in the product direction)
• Nobody on Facebook Ads had any real advertising experience except for some of the people poached from Google
  • They weren’t aware of standard practices for user targeting
• Facebook often secretly tested new features in New Zealand.
  • It had the advantage of being English-speaking like Facebook’s largest markets.
  • New Zealanders largely knew other New Zealanders, so gossip about new features was slower to spread from New Zealand to other countries.
• “data on-boarding”: the ability to match someone’s offline identity to their online consumer profile
  • Companies like Datalogix, Neustar, and LiveRamp buy data from second-tier social networks and dating sites so that they can associate email addresses with browser cookies.
  • Facebook and Google are especially good at this because of how much personal information they have about their users.
    • Facebook’s Custom Audiences can match up to 90% of offline identities to Facebook profiles.
• A weak opening day IPO is actually good for the company going public
  • If a stock jumps 20% on IPO day, it means that the investment bank underpriced the shares and caused the company to sell off huge chunks of equity below their market value.
  • If a stock drops 20% on IPO day, the people who lose out are the investment banks and the buyers who got exclusive deals to buy shares early. The company wins out by selling early shares at a premium to their true market value.
  • Investment banks have financial incentive to underprice IPOs so that they capitalize on the undervalued shares and don’t get stuck holding the bag if the stock drops in price.

• Don’t be deceived by my withering treatment of Facebook in this book; inside every cynic lives a heartbroken idealist.

  • As a once-loyal and later-bitter Googler, I related to this sentiment.