The Second Auction: How Social Learned to Sell Interruption

On November 6, 2007, in a packed room on the sixth floor of a Manhattan warehouse, a 23-year-old took the stage in front of more than 250 marketing executives. Fortune’s reporter described a hush falling over the room, then Mark Zuckerberg saying — she noted, “haughtily” — that once every hundred years, the way media works fundamentally changes.

He then announced Facebook Ads.

The claim was overblown and the launch was partly a disaster. But the underlying idea was real, and it split digital advertising into two businesses that still operate on different logic. Search sells to demand that already exists. Social manufactures it. Understanding which one you’re buying is the single most useful distinction for anyone spending money on either.

What launched that day, and what blew up

Facebook Ads had three parts: Pages for businesses, Social Ads that attached your friends’ actions to advertisements, and Beacon.

Beacon was the problem. It broadcast what users bought on other websites back into their Facebook feeds. Someone bought a movie ticket at Blockbuster, and their friends found out. Someone bought shoes at Zappos, and their friends found out. Someone bought an engagement ring, and — this is the version that made the news — their partner found out.

MoveOn ran a petition. Zuckerberg apologized in December 2007, about a month after launch. The product limped along and was shut down in September 2009 amid a class action settlement.

Facebook’s own newsroom quote from launch day now reads as either prescient or ominous depending on your mood: for the last hundred years media had been pushed out to people, and now marketers were going to be part of the conversation. Being part of the conversation, it turned out, meant knowing what people bought.

Two weeks before all this, on October 24, 2007, Microsoft had bought 1.6% of Facebook for $240 million, valuing the company at $15 billion. Facebook’s actual ad revenue at the time was a rounding error against that number.

The near-death moment

Facebook went public on May 18, 2012 at $38 a share, a $104 billion valuation, and roughly $16 billion raised. The Nasdaq open was botched. More than 80 million shares traded in the first thirty seconds.

Days earlier, General Motors had publicly pulled about $10 million in Facebook advertising, telling the Wall Street Journal the ads didn’t move car buyers.

The deeper problem was in the filings. Facebook’s users were moving to phones, and Facebook made essentially no money on phones. There was no mobile ad product. The stock fell to around $31 within days and to roughly $24 by November.

The fix had actually shipped a month before the IPO. On June 5, 2012, Facebook launched ads that appeared directly in the mobile News Feed — not in a sidebar, not in a banner, but in the same stream as your friends’ posts. On the Q3 2012 earnings call, Zuckerberg said he wanted to dispel the myth that Facebook couldn’t make money on mobile.

Mobile share of Facebook ad revenue — from SEC filings
Quarter	Mobile share of ad revenue
Q3 2012	~14%
Q4 2012	~23%
Q2 2013	~41%
Q4 2013	~53%
Q4 2014	~69%

Total revenue went from $5.09 billion in 2012 to $7.87 billion in 2013 to $12.47 billion in 2014.

The lesson has nothing to do with Facebook. An ad format that lives inside the content people came for outperforms an ad format that sits beside it. That’s why your Instagram feed looks the way it does, why YouTube in-stream works, and why Google put ads inside AI Overviews in 2024.

The targeting stack

Between 2012 and 2014, Facebook built the machinery that made social advertising a performance channel rather than a branding experiment. The dates matter because they’re the thing later dismantled by privacy changes.

Custom Audiences, fall 2012. Upload your customer list — emails, phone numbers — and Facebook matches it to accounts. Your CRM becomes a targeting parameter.

Lookalike Audiences, March 19, 2013. Hand Facebook a list of your good customers and it finds people who resemble them. This was the genuinely new capability. Search can’t do it. Search waits for someone to type something; lookalikes go find people who haven’t typed anything yet.

The conversion pixel, 2013, unified into the Facebook Pixel in 2015. A snippet on your site reporting back what visitors did.

Facebook Exchange (FBX), revealed June 2012 and open by late summer with sixteen partners including AdRoll and Criteo. Real-time bidding for retargeting inside Facebook. Shut down in 2016.

Atlas, acquired from Microsoft in April 2013 and relaunched September 2014 as a cross-device measurement platform — Facebook’s attempt to track people across devices using logged-in identity instead of cookies.

Retargeting had its own parallel track. Criteo, founded in Paris in 2005, pioneered dynamic product retargeting. AdRoll made it self-serve. By around 2013, “ads that follow you around the internet” had become a mainstream complaint rather than an industry term — the first broad consumer backlash against behavioral advertising, and an early signal of everything covered in Part 10.

How the second auction actually works

Here’s the part worth understanding as mechanism, because it’s less different from search than people assume.

Facebook does not simply award impressions to the highest bidder. It ranks by what Meta calls total value, combining three things it names publicly: your bid, estimated action rates — the predicted likelihood that this specific person takes the action you’re optimizing for — and ad quality.

That should look familiar. It’s the same architecture as Ad Rank. A relevant ad from a smaller advertiser can beat a bigger bid from a worse one, because the platform makes more money serving ads people engage with. Facebook even shipped a Relevance Score on February 11, 2015 — a 1 to 10 metric, explicitly described as based on expected positive and negative feedback from the target audience. It was Quality Score with a different name. It was replaced by three separate diagnostics in 2019.

The real difference isn’t the auction, it’s the input. Search auctions are triggered by a query — a person announcing intent in their own words. Social auctions are triggered by a person existing in a feed, with the platform inferring from behavior whether they might want what you sell.

That difference determines everything downstream. Search converts faster because the intent is already there. Social has to create the want first, which means creative carries far more weight, the funnel is longer, and last-click attribution systematically undercounts it.

The other platforms

LinkedIn launched self-service ads in July 2008 as DirectAds and added Sponsored Content in July 2013. Twitter launched Promoted Tweets on April 13, 2010, followed by Promoted Trends and Promoted Accounts the same year.

Both built auctions on the same principles. Neither reached Facebook’s scale, for the same reason Microsoft never caught Google: the auction is only as good as the data and the inventory behind it, and Facebook had more of both.

What this means for your account

Don’t run social the way you run search. A contractor whose Google Ads work well will often try Facebook, run the same offer with the same copy, get nothing, and conclude social doesn’t work for the trades. The mismatch is structural. Nobody scrolling Facebook is currently thinking about their water heater. What works is a reason to think about it — a before-and-after photo, a financing offer, a seasonal tune-up, a neighborhood you just finished a job in.

The one exception where social beats search on intent: your own list. Custom Audiences built from your CRM are the single most reliable social play for a service business. Past customers who bought a repair three years ago are a real market for replacement, and they will never search for you by name until something breaks.

Creative is the bid. Because the auction weights estimated action rate, a scroll-stopping ad literally costs less to deliver than a boring one. In search you optimize keywords; in social you optimize the thing people see. Agencies that treat social as a media-buying exercise rather than a creative one tend to underperform.

Expect attribution arguments. Social assists conversions it doesn’t get credit for under last-click, and it also claims conversions it didn’t cause under view-through. Both are true simultaneously. Judge it with holdout tests and overall lead volume, not the platform’s self-reported ROAS.

Carryover: the vocabulary that survived

Terms established in this era and the instruments they became.

Social Ads · 2007: Facebook’s product attaching friends’ actions to ads. Meta Ads; the social ad auction generally.
News Feed ads · June 2012: Ads placed inside the content stream, launched on mobile. The dominant placement format across every social platform.
Custom Audiences · fall 2012: Uploading a customer list for targeting. Meta Custom Audiences; Google’s Customer Match is the same idea.
Lookalike Audiences · March 2013: Finding people who resemble your customers. Meta Advantage+ lookalikes; Google’s similar segments and audience signals.
Conversion pixel · 2013: Site snippet reporting visitor actions back to Facebook. The Meta Pixel, now paired with the Conversions API.
Relevance Score · Feb 2015: Facebook’s 1–10 ad quality metric. Split in 2019 into quality, engagement rate, and conversion rate rankings.
Estimated action rate · 2012–14: Predicted likelihood a user takes your desired action. Still one of the three components of Meta’s total value calculation.
Retargeting · 2007–14: Re-showing ads to prior site visitors via cookies. Remarketing lists; now heavily dependent on first-party data and consent.
FBX · 2012: Facebook’s real-time bidding exchange for retargeting. Shut down 2016; the function moved inside Meta’s own auction.
Cross-device measurement · Atlas, 2014: Tracking users across devices via login instead of cookies. Enhanced Conversions, Conversions API, and modeled conversions.