Meta's $16 Billion Scam-Ad Problem Creates a New AI Service Business: Deepfake Brand-Impersonation Monitoring Can Catch What a 95% Confidence Bar Won't.

Meta's own internal numbers, surfaced by a Reuters investigation, say the company isn't just failing to catch scam ads — it has priced the failure in.

The numbers, exactly as reported

• Meta projected it would earn about $16 billion in 2024 from ads pushing scams and banned goods — roughly 10% of the company's total revenue that year
• Within that, Meta's own estimate put the annual value of "higher risk" ads carrying obvious fraud markers at $7 billion, out of about 15 billion such ads shown to users every day
• The enforcement bar: Meta bans an advertiser only once its systems are 95% certain he's a crook. Below that threshold, he just pays a higher rate to keep running
• Mark Cuban's reaction, after seeing his own face and voice used to sell a fake diabetes cure: he called Facebook "the worst of the worst" for knowingly running the ads
• Meta's response: spokesman Andy Stone called the leaked internal estimate "rough and overly-inclusive," saying it swept in many legitimate ads. The company also said it removed 134 million scam ads in 2025 and that user complaints about scam ads fell 58% over 18 months

The service business hiding in this story

A 95%-confidence threshold is, by definition, a gap. Below it, an advertiser can keep running an ad that uses someone's face, voice, or brand — as long as they're willing to pay more for the privilege. That gap doesn't close because Meta tightens a model next quarter. It closes because someone outside the platform is watching the ad inventory the platform itself isn't confident enough to pull.

That's a monitoring and takedown business:

• Continuous scanning of ad libraries for face, voice, and brand-mark matches tied to a specific client
• Evidence packaging the moment a match is found — ad ID, screenshot, landing page, timestamp — so a complaint can be filed same-day instead of after a customer flags it
• A standing relationship with the platform's impersonation and IP complaint channels, since knowing which form actually gets a human review is half the job
• A monthly retainer, because ad inventory refreshes constantly and a one-time scan is already stale by the time it's delivered

Who to sell this to

• Public figures and creators whose likeness is valuable enough to fake — Cuban is the visible example, but any recognizable spokesperson is exposed to the same pattern
• Consumer brands in categories scam ads gravitate toward: supplements, investment products, "as-seen-on-TV" style goods
• Legal and comms teams who'd rather pay a monthly fee than explain to a customer why a fake ad using their brand was live for a week

Bottom line

Meta didn't just fail to stop scam ads — Reuters' reporting shows the company modeled what tolerating them is worth, in dollars, at a specific confidence threshold. That threshold is a published gap. Someone is going to get paid to watch it.

Source: https://www.reuters.com/investigations/meta-is-earning-fortune-deluge-fraudulent-ads-documents-show-2025-11-06/