The Signal
What changed.
AI and the ad platforms keep marching toward full automation. Performance Max, Advantage+, Andromeda, GEM: the bidding, targeting, and budget work media buyers once did by hand is being automated toward zero, and Meta says that by the end of 2026 you'll just hand over a budget and a goal.
The Hypothesis
The emerging thesis we're exploring.
The machine ate the low-leverage work: the manual bids, the keyword wrangling, the endless micro-optimizations. So what’s left? The hard stuff: narrative, creative, data, measurement, behavioral science. In other words, the stuff that's always mattered most in advertising.
So no, AI didn't commoditize paid media, and it sure didn't make cracking it easier. If anything, the bar has risen enormously. The companies that bought the commodity story, that figured they could hand Meta the keys, churn out a pile of creatives, and sit back while the money printer hums, are getting crushed. The winners are taking the opposite approach: they invest in deep expertise and treat paid media like the high-stakes discipline it's always been.
The Big Picture
The structural shifts behind the hypothesis.
- Hard and high-leverage were never the same thing. Yes, manual bidding and targeting optimizations were nonstop, tedious work. And yes, the algos largely handle those now. But being really good at those things was never a major edge to begin with. At most, a band-aid solution. For example, for big spenders, ad ops could be a real edge over equally funded but sloppier rivals. But it was a secondary edge, and it covered for weak fundamentals.
- Shared automation is nobody's advantage. When every advertiser runs the exact same automated baseline, that baseline can't be your edge. Everyone has it. The advantage moves to what can't be copied out of the box: your strategy, your data, your judgment, your storytelling. Automation didn't kill the edge. It just freed up more time for advertisers to invest even more in the areas that have always been responsible for real advantages.
- "All that's left is creative" undersells it too. Creative matters enormously, but it's just one pillar. The best advertisers also know how to obtain and feed the algorithms proprietary data, measure true incremental impact, and craft beautiful stories & experiences across the full customer journey. Concluding that “creative is all that matters” is honestly just a confession that someone has no clue what it actually takes to win in advertising.
- The result is a moat, not a commodity. Automation wiped out the small edges that let weaker advertisers squeak by. The gap between the best and the rest didn't shrink, it widened, in a pretty massive way. The top operators are sitting on a moat they may not have had in years, maybe ever, precisely because the cheap ways to brute force it are gone.
- In the wild, the mistake looks like outsourcing your growth to the lowest bidder. Convinced it's all automated now, we’re seeing so many founders treat paid media like a cost center. They’ll compare a seasoned operator at $10k/mo versus someone with two years' experience at $3k, and they pick the $3k because "how hard can it be, the AI does it." For a startup, that's not too different than delegating product-market fit to an intern. Growth is the highest-stakes thing you do, and they're trying to spend as little as possible on it. Then the ads don't work, they take to LinkedIn to declare paid media a scam, and the cycle repeats. Bad inputs, bad outputs. Tale as old as time.
What's Working
Tactics, experiments, and observations from the frontier.
- Stop trusting the ROAS your platforms report. Every platform grades its own homework, and it grades generously. Across 299 brands, Google brand-search ROAS collapsed from 19x to 5.7x once anyone bothered to measure incrementality. The sharp operators run holdouts to see what their ads truly caused: Advantage+ over-reports against a manual control, and blended efficiency, total revenue over total spend, never runs through platform attribution at all. Knowing your real numbers is the unglamorous work most teams skip.
- Data engineering is the lever, not the pixel. The algorithm is only as smart as the signal you feed it, and that signal is yours to control. Feed it closed-won revenue, not form-fills, so it chases customers who pay. Send it a lead-quality score so it bids toward your hot leads, not your tire-kickers. Feed it item-level margin so it stops happily scaling your thinnest products.
- The click is the start, not the win. Everyone fusses over the ad, but the ad's only job is to earn the click. The landing experience carries far more of the conversion, and the recent obsession with creative volume isn’t helping. Match the page to the buyer's stage, and if you're running ten ad angles, give each its own page instead of dumping them all on one generic PDP. What good is high-velocity creative testing when you're sending traffic to a generic landing page that hasn’t been updated in 12 months (and was probably built in 30 mins with AI)?
- Budget by the next dollar, not the blended average. A healthy account-wide ROAS hides which dollar is doing the work. Spend toward the next dollar's return, not the average, and move your efficiency target when you discount, because a 20% promo can put you underwater while your team is celebrating a "high ROAS."