Most paid media decisions are now being made by machines on both sides of the auction. Your bidding agent against Google’s, Meta’s, TikTok’s. Your creative against thousands of AI-generated variants. The humans in the loop are increasingly setting the constraints, not making the calls.
That changes what’s worth focusing on, and most paid teams are still spending their time on the wrong layer.
What the machines actually do
The platforms are running optimisation against signals you barely see. Performance Max, Advantage+, Smart Bidding, these aren’t features, they’re agents. They take inputs (creative, audiences, goals), run constant tests, and produce outputs (spend allocation, placement, bid).
What they reward.
Creative volume. More variants, more often. Static rotations die, and whatever the model can iterate against, it iterates against.
Signal density. Conversion events, audience signals, first-party data feeding the auction. Brands with clean data pipes outbid brands with the same budget but worse signal.
Landing experience. The agent reads your post-click behaviour and steers spend accordingly, so a great ad with a poor landing page is worse than a good ad with a strong one.
What’s stopped working
The old manual paid playbook, fixed audiences, fixed creatives, weekly check-ins, quarterly retainer review, is mathematically losing now. The agents can’t optimise against creative that doesn’t change, and the volume the platforms expect (50+ variants per campaign isn’t unusual) is impossible for a human team to produce alone.
So the gap is widening. Brands using AI to produce creative at platform tempo are pulling in front of brands still paying agencies to ship a static quarter’s worth of assets.
Where the leverage actually sits
Creative volume at brand quality. This is the hard part. AI can produce a hundred variants but most are unusable because they break brand or break taste. The win is a pipeline that produces a hundred variants at the floor of brand quality you accept, and that isn’t a tool, it’s a workflow.
Feed and signal hygiene. Your product feed, your audience definitions, your event taxonomy. The agents are only as good as the inputs, and most brands have not invested here in years.
Outcome design. What is the agent allowed to optimise for? Click, add-to-cart, revenue, LTV? The further down the funnel you can give the agent a real signal, the better the spend gets steered, and this is a strategic call rather than a media-buying one.
The negative space. What the agent doesn’t see (competitor moves, brand impact, channel cannibalisation) is where humans add value. The senior call is “don’t let the bidding agent buy you out of your own brand search” or “don’t let the optimisation drift the brand for a 2% lift in last-click ROAS”.
The shape of a good paid function now
A small operator team plus a creative AI pipeline plus clean data plus clear goals. The agency model where you pay 20% of media spend for a media buyer doing manual bid changes is dying.
Spend should compound. Each week the model should learn more, the creative should evolve and the signal should sharpen, so if your paid function isn’t getting smarter every week, something in the loop is broken.
The brands that win paid in the AI era won’t necessarily spend more. They’ll have better inputs, faster loops and senior judgement about what to actually optimise for, and the bidding agents will do the rest.