AI media buying: what to hand to rules and what a buyer still decides

Advantage+ sales campaigns, formerly called Advantage+ shopping campaigns, set audience, placements, and budget automatically. In Meta's developer documentation, a campaign counts as Advantage+ only when all three automation levers are on.

• Budget is set at the campaign level with a supported bid strategy, and Meta moves it between ad sets.
• Placements have no targeting or exclusions at the campaign level, so every available placement is eligible. Exclusions set at the ad account level are allowed.
• Audience uses Advantage+ audience, or no targeting beyond location.

If you add targeting or placement restrictions outside those criteria, the campaign stops being Advantage+. That's the trade: you get Meta's optimization by giving up the controls a buyer used to spend the day on.

One control worth knowing about: the old setting that capped spend on existing customers isn't available for new Advantage+ campaigns. Meta's documented replacement is two ad sets, one aimed at a strict custom audience of existing customers with minimum and maximum daily spend limits, and a duplicate that excludes that audience. If new customer acquisition is the goal, this is the setting to check first.

Performance Max runs one campaign across all Google Ads inventory: YouTube, Display, Search, Discover, Gmail, and Maps. Google's documentation lists what it handles.

• Bidding, with Smart Bidding in each auction, against your conversion goals and an optional ROAS or CPA target.
• Budget, moved toward the inventory and placements performing best.
• Audiences, including new customer segments.
• Creative combinations, mixing your text, images, and video into new ads, with generative tools for variations.

You supply the inputs: conversion goals, asset groups, audience signals, search themes, brand exclusions, negative keywords, and whether Google can send traffic to other pages on your site through final URL expansion. Google says you're still responsible for reviewing the landing pages it picks.

Both systems optimize toward the signal you give them, inside their own walls. Neither one knows your margin, your inventory, or what the other platform is doing. That leaves four decisions with a person.

• The conversion signal. Optimizing to purchases, to purchase value, or to new customers produces different campaigns. If returns or a low-margin product skew your value data, the platform will chase the wrong orders.
• The offer and the creative. The platform picks among the ads you give it. Which angle, which product, and which offer go into the account is a strategy call.
• Budget across channels. Each platform will spend what you allow and report a good return on it. Someone has to decide how much goes to Meta versus Google versus email and retention, using your own sales data.
• Measurement. Platform-reported conversions are each platform's own attribution. Whether the spend caused sales that wouldn't have happened anyway is a separate question, covered below.

Both platforms let you write rules that act on the account without anyone logged in. Meta's Ad Rules Engine runs rules either on a schedule or when an ad's data changes. Trigger rules can pause or unpause ads and send notifications. Scheduled rules can also change ad set or campaign budgets, change bids, and rotate ads, with a cap on how many times a budget or bid change can fire per object and a minimum time between changes. Google Ads automated rules can change ad status, budgets, and bids, and send email alerts.

A starting rule set for a brand spending at scale might look like this. The thresholds are illustrative; set your own from your margins.

• A kill rule pauses any ad that has spent twice your target cost per purchase with no purchases.
• A fatigue rule pauses an ad when its cost per purchase over the last three days runs well above its lifetime average.
• A scale rule raises a winning ad set's budget by a fixed percentage once a day while cost per purchase stays under target, with an execution limit so it can't compound forever.
• A spend cap alerts a person, or pauses the campaign, when daily spend passes a dollar ceiling.
• A tracking alarm notifies a person when purchases reported by the platform drop to zero while spend continues. That pattern usually means a broken pixel or checkout.

Inside Advantage+ campaigns Meta already moves budget between ad sets, so keep budget rules at the campaign level and use ad-level rules for killing creative.

When targeting is automated, creative is most of what you control, so creative ad testing becomes the buyer's main job. A process that holds up:

1. Write the hypothesis first: which angle, hook, or format you expect to lower cost per purchase, and why.
2. Change one variable per test. Meta's own A/B testing guidance says to test one variable at a time and to start with creative.
3. Keep the test audience out of other campaigns running at the same time. Meta warns that overlap can skew results.
4. Let it finish. Meta's tool shows results after at least 100 events for the cost per result being tested, and recommends running a test for at least two weeks and up to 30 days.
5. Move winners into the main scaling campaign and log what won and why, so the next brief starts from it.

On Google, Performance Max experiments include asset tests that measure the uplift from different creative assets. A creative testing platform from a third party can speed up making variants and reading results. The call on what goes live, and what the brand will and won't say, stays with a person.

The platforms offer holdout tests for this. Google's Conversion Lift splits people into a group that sees your ads and a group that doesn't, and the difference in conversions is the lift the ads caused. It reports incremental conversions, incremental cost per acquisition, and incremental ROAS. Google says you need to contact your account representative to use it. Performance Max also has an uplift experiment that compares results with and without Performance Max running alongside your other campaigns.

Even without a formal lift test, compare platform-reported purchases against orders in your store each week. If the platforms together claim more purchases than you received, the attribution is overlapping, and budget decisions made on it will be off.

On the Zero Hire Org Chart, a made-up $50M subscription brand goes from four media buyers to one. Pacing, budget shifts, and killing losing ads run on rules. The remaining buyer designs the tests and makes the calls the rules can't. Creative strategists go from two to one, because briefs and angle research come out at testing volume, but someone with taste still decides what goes live.

That buyer is a harder hire than any of the four before. Run your own team through the org chart roast to see what the current setup costs, or plan the change in the headcount planning template.

Roast your org chart