Real-Time Bidding for Our Attention
@midjourneybot: /imagine: real-time bidding for human attention
Real-time Bidding for Your Attention
In 2012, Facebook paid one billion dollars for a company with zero revenue, virtually no assets, and just 13 employees. Old school investors accustomed to acquiring companies for multiples of revenue or book value were shocked. Facebook paid Instagram investors $55M dollars per employee, which is better human capital efficiency than the LA Lakers and the Dallas Cowboys and they get paid for harvesting human attention too. So what exactly was Facebook buying?
Engagement.
Engagement refers to the level of interaction between a computer system and its Target Audience. Engagement describes how much attention a software interface supplies. The most common measurement of engagement is the MAU, which stands for Monthly Active Uniques.
If 5 different people visit a website 500 times in a month, the website’s MAU is 5.
Private equity valuations for software companies are more highly correlated with MAU than revenue because once the software “habituates” the human attention, there are plenty of ways to “tax it”. When Facebook acquired Instagram, they simply “plug-and-played” their existing advertising engine into the Instagram newsfeeds to dramatically increase their Total Inventory. Facebook’s “total inventory” is the count and quality of human brains they can sell to the highest advertising bidder.
Human attention is much more “liquid” than other forms of capital so it’s fascinating to see it ebb and flow over time. Instagram had 50M MAU when Facebook acquired them in 2012. Just four years later, Instagram had 500M MAU. What other kinds of businesses could handle that many new customers in just four years? Today, 2.35 billion people use Instagram each month, which is 47.84% of all smartphone users worldwide. That’s a lot of “inventory” for sale to the highest bidder.
Here’s a 9-minute video showing the most popular websites in the world month-by-month since the beginning of the internet. You will see the rise and fall of AOL, Yahoo, several search engines before Google, and several encyclopedias before Wiki. You’ll see Amazon hang around forever. Eventually you’ll see Facebook rocket to the top with Instagram in tow. By the end of the video, you’ll see two different porn streamers that are both bigger than Netflix.
Here’s 78 more eye-popping Instagram audience engagement metrics:
Audience Engagement has loads of other metrics besides just MAU. User Experience (UX) Designers are the researchers and artists responsible for the way that software is presented to humans. To perpetually increase their audience engagement, UX designers have to measure everything. Every tap, touch, and swipe within an app or website is collected to create an Engagement Funnel for the UX designers to experiment on. An engagement funnel is the series of screens or pages that a user follows along their Customer Journey. Each of those screens will have a different Bounce Rate that shows the percent of users that abandoned their Session from that screen.
Don’t worry, you won’t have to remember any of these definitions.
In the apps on our phone, UX designers can see heat maps of exactly where our big, fat thumbs hit their skinny, little buttons. UX designers track likes, click-through rates, impressions, shares, comments, and scroll-speed. Generally, the slower you scroll past content in your feed, the more interesting it is to you.
The most important action that software designers want us to take is creating new content. For example, when we create a new playlist in Spotify it becomes that much more difficult for us to switch to Apple Music later. The reason social media makes it so easy for us to leave one-tap Emoji comments everywhere, is because they create new content. When you tap 🔥, the computer algorithm not only adds that comment to your friend’s Instagram post, but it also adds an alert about your new comment in their alerts tab. Everyone that has previously commented or liked that post will also get push notifications and alerts that there’s an update to the story. Comments are basically the pistons of social media engines.
There is an entirely different set of audience engagement metrics for the Advertisers. Advertisers don’t care how easy the software is to use. Advertisers don’t care about the bounce rates. Advertisers don’t care if users were tricked into that session from fake news or headline traps. The most important metric for advertisers is Reach.
Reach is the number of brains that consumed their advertising content.
Technically, that’s Actual Reach. There’s another metric called Anticipated Reach, which is the total possible audience that could consume their content. So if you divide actual reach / anticipated reach, you will get a Reach %. Here's why that matters to you. The people reading Instagram only have so much time in a day to look at photos of other people’s lunch. So Instagram can’t afford to show each of your posts to all of your followers because, technically, you are also an advertiser.
For example, let’s say you have 1000 friends on Facebook and 1000 followers on Instagram. When you post a photo of your dog being silly on both platforms, how many people will see your dog?
2000
1000
More than 2000
Less than 1000
136
The average Organic Reach on Instagram is only 9.34%, so your dog photo only appears in the Instagram newsfeed of 93 of your followers. If you received 50 likes for your dog photo on Instagram, that means more than half the people who saw it, liked it.
The average Organic Reach on Facebook is worse, at 4.32%. Your dog photo will only appear in the Facebook newsfeed of 43 of your friends. (So technically, 136 is artificially inflated by the number of friends that follow you on both platforms.)
Organic Reach means free. To have more than 9% of your Instagram followers see your post, you need to pay Emperor Zuckerberg. That goes for celebrities too. I analyzed the video view-counts of several celebrities a few years back and they only get 9% organic reach too. Matthew McConaughey, for example, had about 4.5 million followers at the time and his view counts were usually around 400,000. But posts that pushed his whiskey, his book, or other endorsed products got between 1.5M to 2M views. So Paid Reach can only get you to about half of your own audience. I can’t imagine what Christiano Ronaldo has to pay to talk to more than 9% of his 600M followers.💰💰💰
Paid Reach is one of the reasons many influencers are moving away from social media to platforms that let them Own their Audience instead of just Renting their Audience. Some of the most popular services for this are Substack, Twitch, and Onlyfans—which is diversifying away from its porn roots. Substack delivers your content directly to your audience's email so you don’t have to pay to talk to people who already said they want to hear from you. Twitch allows gamers and influencers to live stream their lives directly to their audience. Onlyfans does that for amateur work-from-home strippers. 🤣
Meta makes so much money renting everyone’s audience back to them. To get a better idea of what each brain in an audience is worth, let’s use some “cocktail napkin math” from Meta’s recent earnings data.
Meta, which owns Facebook and Instagram, earned $116 billion dollars of ad revenue in 2022. If we divide their total revenue by their average MAU (2.7 billion people), they made about $42 per brain that year. Dividing the revenue per brain into months gives us $3.50 per brain per month. If we divide that by the average time spent on Facebook and Instagram each month (14 hours), we get $0.25 per brain per hour from reselling our Human Attention. Meta’s gross profit margin is an incredible 78% because products made entirely out of electrons have almost no inventory costs and free shipping worldwide.
So Meta keeps 100% of the Money from advertisers and 95.7% of the Reach from its users. #greedy?
There are several ways digital advertisers are billed for our Human Attention:
The most simple way to pay is per impression. Impressions are so cheap, they are billed in batches of 1000, referred to as an eCPM (Cost per Milli). The average eCPM is between $2 and $10 depending on the country.
Advertisers can also pay CPC (Cost Per Click), which verifies that the user followed a link provided by the advertiser. These are much more expensive (between $0.50 to $5 each).
Advertisers can also pay CPA (Cost Per Action), which means advertisers only pay if the user follows a link and performs a desired action on their platform. CPA is why giveaways from influencers require you to follow all the steps from their sponsor to be eligible.
CPI (Cost Per Install) is even more expensive because the user has to actually install an app on their phone and App Stores have crazy high bounce rates. The average CPI is more than $5 per install in North America. That’s $5000 per milli.
The last billing method that’s notable is called CPCV (Cost Per Completed View). CPCV is how advertisers get charged for video ads.
Before the digital revolution we watched TV, but after the digital revolution the TV is watching us. Connected TV like Amazon Fire, Apple TV, Chromecast, and Roku not only watch what we stream, but when we do it, when we pause, when we rewind, the trailers we watch, and which descriptions we read. The value of this information is reflected in the average prices of the human attention sold on Connected TV. Here are the normalized average prices for a 30 second spot on various streaming platforms per thousand impressions:
Cable TV $20
YouTube $10
Hulu $30
Connected TV $35-65
Amazingly it’s worth almost 100% more to deploy that same content on a Connected TV service than Cable TV. The targeting is that much better and the viewers often can’t fast forward through commercials. 95% of Connected TV ads are viewed completely.
Those stats are amazing, but how many video ads do you actually pay attention to? There’s no double entry accounting for this process so digital advertisers have to trust the impression counts from their ad networks. But Google and Meta don’t care if you’re watching the advertisements or not. Meta can hardly distinguish which computers are operated by humans and which are operated by bots.
Meta removes 3 billion fake people from its social networks every year.
Even after “digitally terminating” all those bots, that still leaves 2-3 billion human brains for Meta to coin-op. With that much inventory, Meta is basically “the Walmart of Human Attention”. Meta can sell you any kind of brain you want. You can target people by household income or by education level. That’s easy. You can target people with an “Anniversary within 30 days” or “Close friends of women that have a birthday in 0-7 days”. You can target people “Currently away from hometown” or “Currently within X miles of any location”. Meta will even target the employees of a specific company for you. You can target people based on what they like, what they read, or what they buy. You can even use the emails from your own customer lists and have Meta’s algorithm target people exactly like your customer. That’s called a custom “look-a-like audience”. It’s just incredible what Meta knows about us. 🕵️♀️
The hunger for human attention creates some perverse incentives for all social media, not just Meta. Recall that the organic reach on Facebook is just 5%, so that means their EdgeRank Algorithm has to guess which 5% of our friends’ posts are most important to us. It doesn’t do that at night while we’re sleeping. When we use the Facebook or Instagram app, right when we scroll up a new “card” that is just out of view, the app “lazy loads” new content. That way the Meta EdgeRank Algorithm has the highest optionality at the moment your content is actually consumed.
So imagine we are your Facebook app at this exact lazy load moment. The Facebook app asks the EdgeRank Algorithm for your next best content (likely to keep you engaged in the app). Let’s say EdgeRank scans the remaining 95% of the potential content that you haven’t seen and notices that one post has 10 times as many comments as every other post.
Doesn’t it make sense that you would want to see that post next?
Those comments are created by people most likely to be your friends. Those comments are also likely to be unique in all of human history. Once a sentence has more than 9 words, that sentence is more likely than chance to be unique. Don’t you want to see what is making your friends so fired up?
Of course you do, we all do.
Now, here’s a list of possible emotions from an “emotion wheel” used in Psychology. If any of your possible content cards has 10 times the comments of every other post, which emotion do you think those people were likely feeling when they typed those comments?
Anticipation
Joy
Trust
Fear
Surprise
Sadness
Disgust
Anger
Do you think all those comments are from people debating how much they trust each other? Do you think they are all opening up about how much fear and sadness they share? No, anger and disgust are the emotions that put the “pep in our step”. Anger moves us. So when the Facebook EdgeRank algorithm looks to grow its 16 hours of attention per brain per month, which card is it going to choose?
The one you want to see most.
Social media is a technology that’s making us more of who we already are. If we all avoided contentious material and only consumed joyful and peaceful content, that’s what their algorithms would prioritize out of corporate greed.
But that’s not who we are.
When’s the last time you watched a Hollywood movie that didn’t have a fight at the center of its story? See for yourself. Here’s IMDB’s 100 most popular movies right now. How many of them have two sides fighting over something?
What percent of those movies don’t have a murder? Here’s the 100 most popular television shows. How many of them have two side fighting over something?
@america: Social media is causing our country to experience its Second Civil War. The Second Civil War might be a “cold” land war, but it’s a “hot” financial war, and a “nuclear” attention war. Instead of bullets and cannons, the second civil war is fought with memes, fake news, and trolling comments.
@america: The net result of social media algorithms cherry-picking the best performing 5-10% from all our content is that we designed all our own individual realities. Everyone got pushed away from the middle because the middle is bad for business. Of course we aren’t going to agree with the other side, we don’t even share the same facts. Half our country thinks the 2020 election was stolen, while the other half knows the other side was tricked. 🤣
@republicans: George Washington may have never told a lie, but Donald Trump may have never told a truth. You know how people use phrases like “honestly”, “to be honest with you”, and “here’s the truth” when they are lying to you? Donald Trump named his entire social network Truth Social. That guy lies so much that he lies too much for regular social media, which is full of lies. I was a republican before Trump, but now he’s just making your party look like fools. Actually, he’s making our entire country look like fools. 🇺🇸
@youngpeople: Listen to your girl Tay-Tay. These aren’t your dad’s Republicans. They aren’t Abraham Lincoln’s republicans. These Republicans are vitriolic, vindictive, and…
@ChatGPT: which words that start with v mean “sore losers”?
ChatGPT: A word that starts with "V" and means sore losers is "vexatious." While it may not specifically refer to sore losers in every context, "vexatious" describes individuals who tend to be difficult, annoying, or overly litigious when faced with defeat or unfavorable outcomes. They may exhibit a persistent and unreasonable resentment or refusal to accept defeat gracefully.
I don’t know how human society will overcome all the individual truths manufactured by our social media. Humans have never been that great at compromise. Keeping our democracy together was challenging before the internet, I just hope it survives social media. For more information about how social media shapes our individual truths check out the documentary, “The Social Dilemma” on Netflix.
While you’re annoyed at social media, remember that quantum mechanics is also creating an individual version of reality for you right now too. And that reality is also heavily influenced by your beliefs. 🫠
Now let’s get back to Meta’s revenue stream.
For Meta to generate their $3.50 per brain per month of revenue, they have to occasionally “lazy load” a Sponsored Post into our newsfeeds. The percentage of sponsored posts has slowly crept up over time and is currently 21%. The reason I explained the lazy load earlier is because what happens in the lazy load of a sponsored post will blow what’s left of your mind. Instead of asking the EdgeRank Algorithm to guess which post you will interact with the most, your feed simply asks a Real-Time Bidding Engine (RTB) for content from your highest bidder.
Billions and billions of seconds of human attention are auctioned every day, just like stocks, using Real-Time Bidding Engines. The stock markets for attention don’t keep Wall Street banker hours either, they trade all day and all night. Just as you are scrolling the first pixel of a sponsored post into view, the entire auction takes place in milliseconds.
The RTB engines of Google, Meta, and Amazon currently earn 50% of all digital advertising dollars on Earth.
To help their customers (rich corporations) decide which products to buy (brains), RTB engines need to know as much about their brains as possible. Facebook is free because we inform them of our friends, our likes, and our relationship status. But arguably the most important datapoint when bidding for any brain, is Location History.
Where your phone goes is who you are.
For example, my hometown of Austin has several major grocery stores that I’ll list starting with the most expensive: Central Market, Whole Foods, HEB, Randall’s, and Fiesta. If my phone goes to Whole Foods every week instead of Fiesta, that implies a lot about my purchasing behavior. If my phone goes to a church on Sunday that means something about me as a person. If my phone goes to a little league baseball field or skiing in the winter, those locations all mean something different about me as a person.
Location history isn’t just harvested by our phones, it is also harvested by individual apps installed on our phones. If you have an iPhone, tap Settings -> Privacy and Security -> Location Services and then scroll down through that list. A gray arrow beside an app means it has sampled your location in the past 24 hours. A purple arrow means that app is using your location right now. Previous versions of iOS had a hidden map in Settings -> Privacy -> Location Services -> System Services -> Frequent Locations that would show you every “GPS dot” of your entire life, and how many minutes you were there each day. I loved that map as a data researcher.
@cybernerds: I started my first location data company when Steve Jobs put a GPS antenna in the iPhone 3. I used a dev app to record everywhere my phone went all day every day. If my phone was stationary for more than 10 minutes I used the Google Places API to pull all the merchants that were “in view” of my location. My plan was to make some kind of “CPC for the real world” where merchants could give extremely personalized discounts to nearby foot traffic that would compete against Valpak (which still has no targeting). After a few months of collecting data, I had my entire life loaded into a SQL Server database. I was curiously running queries on myself one day and noticed the time it took for me to drive from my exit on the freeway to my house was 6 mins 13 secs. When I summed it by how many times I took that trip each year…I spent 43 hours per year just driving to the freeway. That’s a whole workweek. I walked downstairs and said, “honey, we gotta move”. 🚛
@cybernerds: We couldn’t get marketing agencies that interested in location data back in 2011 so we pivoted the location tracker to create a parenting app called MamaBear, which was a best new innovation finalist at the TechCrunch “Crunchie” Awards in 2015. We failed to secure funding despite the fact that 20% of new users became power users with over 7 app sessions per day. Building the right cap table is even more difficult than building the right team, which is more difficult than building the right product, which is more difficult than building the right customer base.
🤣😭🤣😭
To learn more about privacy and location data, I attended the RSA Security Conference in 2015. One of the many events was a panel discussion that included the Chief Privacy Officer of Facebook, Chief Privacy Counsel of Google, an Architect from Microsoft, and some guy from Mozilla. They spoke to about 1000 conference attendees in the Moscone Center in San Francisco. When the time came for audience Q&A, I raced to be first to the mic. I announced, “This question is for Keith Enright from Google. Given what Google is doing with all my web data, can I expect you guys to harvest and monetize all my location data from google maps?” His response (glibly delivered) was, “If you don’t think we’re going to treat your location data like we treat all your web data, then you need to elevate your digital literacy”. This is the guy who Google chose to protect our privacy.
Digital literacy elevated. 😡
Eight years later, when I checked my location data permissions to write this story, I noticed Google Keep had sampled my location in the background in the past 24 hours despite the fact I haven’t used that app in 3 years. Google Keep is a note taking app. I didn’t even remember I had Google Keep installed. But here’s the rub. The reason my YouTube and Google News recommendations are so dialed into my interests in life is because Google snoops in every corner of it. Now I kinda want Google to know everything I’m thinking because I want their recommendation engines to be that much more informed.
Digital literacy elevated. 🙏
Google bought Nest thermostats just so they have a reason to listen to the conversations in our homes. Google’s “devices chief” actually recommends that we all inform each of our houseguests that their machines are listening to their conversations. That’s preposterous. Here’s a link to the article about it in BusinessInsider titled, “Google exec says Nest owners should probably warn their guests that their conversations are being recorded”
Today, Google and Apple don’t even need GPS to know where we are. Google Maps now includes a feature called StreetView, where you can see the actual street view of almost every location of the civilized world. It’s really useful. Google sent cars all around the world with cameras on the roof to take those photos. Google eventually got sued all the way to the Supreme Court though, because those cars were also equipped with GPS and powerful Wi-Fi Sniffers. Google and Apple have constructed their own private maps of all the world’s “Wi-Fi bubbles” that get created around our houses and apartments. Even when our phones don’t connect to foreign Wi-Fi networks, the phones can still see the name and connection details. So instead of our phones asking satellites in outer space where we are, our phones simply ask their manufacturer where we are based on the Wi-Fi networks that are “in view”.
This will be easier to understand with an example. On your phone or computer go to the Wi-Fi setting as if you were going to connect to a different network. See all those Wi-Fi networks in that list? Even if there is only one in the list, that list is a unique combination of Wi-Fi networks in all of the world. That combination only exists at your current location right now. Each Wi-Fi network has a unique hardware identifier called a MAC address that you can’t see, but Google and Apple have already sniffed them all out like a truffle pig. 🐽
Wi-Fi sniffers are good at knowing where you are, but they are nowhere near the accuracy of Bluetooth Beacons. Think about how fast your phone automatically connects to your car when you get in it. That’s because several times per second your phone is invisibly asking the entire world, “hey do I know any Bluetooth devices in range?” Bluetooth Beacons listen for this invisible request and don’t even need to respond. Walmart, Home Depot, CVS, and pretty much every other big box retailer places Bluetooth Beacons every few meters along every aisle.
The beacons compare their signal strength to each other to triangulate where you are within a meter. This allows Home Depot to know how fast or slow you walked past the DeWalt drills, just like Facebook knows how fast or slow you scroll past each of your friends.
Your location history, your identity, your job title, your income, your web history, what you say in your home, and everything else about you is Context for RTB Auctions. For RTB engines to get the most money for your brain, advertising networks need to offer their bidders as much of this context as they can, which has drawn the criticism of digital privacy regulators in the EU. The Irish Council for Civil Liberties published a 12-slide pitch deck with some truly frightening RTB statistics. The headline on one slide reads:
Biggest data breach ever. Repeated daily.
Here are a few quotes from the “Key Insights” slide:
RTB is the biggest data breach ever recorded. It tracks and shares what people view online and their real-world location 294 billion times in the U.S. and 197 billion times in Europe every day.
On average, a person in the U.S. has their online activity and location exposed 747 times every day by the RTB industry.
In Europe, RTB exposes people’s data 376 times a day.
Europeans and U.S. Internet users’ private data is sent to firms across the globe, including to Russia and China, without any means of controlling what is then done with the data. The RTB industry generated $117+ billion in the U.S. & Europe in 2021
This is happening TO YOU every day. The slide deck is worth a look:
More incredibly, the authors of the study consider their estimates to be conservative. They write, “The industry figures on which we rely do not include Facebook or Amazon RTB broadcasts.” Huh?
TechCrunch has a good article about the Irish Council for Civil Liberties report. It reads, “Per the report, Google, the biggest player in the RTB system, allows 4,698 companies to receive RTB data about people in the U.S., while Microsoft — which ramped up its involvement in RTB in December last year when it bought adtech firm Xandr from AT&T — says it may send data to 1,647 companies. That too is likely just the tip of the iceberg since RTB data is broadcast across the Internet — meaning it’s ripe for interception and exploitation by non-officially listed RTB ‘partners’, such as data brokers whose businesses involve people farming by compiling dossiers of data to reidentify and profile individual web users for profit, using info like device IDs, device fingerprinting, location etc to link web activity to a named individual, for example.”
You can read the whole story here: Report spotlights vast scale of adtech’s ‘biggest data breach’
I have seen several demos from Data Brokers and most are scarier than you would imagine. Data brokers, “whose businesses involve people farming”, don’t just buy data from the phone company. They buy from all the apps on your phone. They have your pharmacy card data and your grocery store data. They even know what kind of mail you get because they can buy that from Pitney Bowes. Even Meta buys data from Pitney Bowes because Meta buys data from everybody. For example, there are 25 million women that use the Flo Period and Ovulation Tracker app each month. Meta pays Flo to perpetually monitor the ovulation cycles of Facebook and Instagram users. All of this context helps increase the Click-through Rates of their RTB engines. Here’s the write up about it in the Wall Street Journal.
My favorite demo from a data broker was from a company called GroundTruth. I was looking to target phones that go in and out of apartment buildings for research into “in real life” social networks. During the demo the sales guy started with a map of the United States, then zoomed into Washington, DC and then zoomed in again to the Pentagon. He used his mouse to trace a blue pentagon around the Pentagon building, and then tapped “Search”. Over the next two minutes, their location sniffing database found over 30,000 phones that had been “location sampled” inside the Pentagon within the past month. The sales guy randomly scrolled down through this enormous list of phones and randomly clicked on a single row. The map immediately changed to a residential neighborhood to show where that phone was last seen—which seemed to be that person’s home.
How can 30,000 of the most security-minded people in the world not know they are broadcasting their locations all over the internet? Because it’s the “Biggest data breach ever. Repeatedly daily.” All it takes is a pentagon employee using their phone to read one stupid news article with one RTB advertisement to create that “digital breadcrumb”.
If you don’t pay for the product, you are the product.
Continue reading
⬅️: The War for Human Attention
Receive months of free VIP access by referring friends to this story:
@influencerswith100kfollowers: Tell your audience what I said, just make sure to mention funfreq.com. Also, I want to give away (100) free years of VIP access to your (100) favorite fans. Just email me* a spreadsheet containing only their email addresses without identities to ihave100kfollowers@funfreq.com
*spam is attention theft. If I think you’re spamming me, I will ban your email address from my life.
@influencerswith1Mfollowers: I want to be a guest on your podcast, radio show, tv show, Facebook, Weibo, Twitter, LinkedIn, Instagram, TikTok, Twitch, or YouTube. Email me* at ihaveonemillionfollowers@funfreq.com
@sponsors: I don’t supply ad networks, but I will create sponsored mentions for people and products I like. If you want me to drive traffic to you, email me* at wewanttosponsor@funfreq.com
As an Amazon Associate, I earn from qualifying purchases. All intellectual property rights, including copyright, relating to the funfreq.com domain shall remain vested in John Stuart Kime, unless otherwise stated. Any extraction, translation, reproduction, and distribution, in any form, or by any means, electronic, mechanical, photocopying, or otherwise, requires the explicit prior written permission of John Stuart Kime. (That means if the page has a share button then humans can share it, but ai can’t train on it).
I reserve the right to disagree with everything I’ve said.