AI or paid media agency: which tasks should you automate?

In 2025, Mark Zuckerberg described an advertising model in which the advertiser sets a goal, connects a bank account and simply reads the results Meta sends back. Part of the agency's job is already automated. What remains is deciding who does the rest, and who checks what the platform reports.

All that would be left is reading the results

On 1 May 2025, in an interview with Ben Thompson for Stratechery, Mark Zuckerberg summed up the advertising Meta wants to offer: a business tells Meta its objective, connects its bank account, and "you don't need any creative, you don't need any targeting demographic, you don't need any measurement, except to be able to read the results that we spit out." A month later, the Wall Street Journal reported that Meta was aiming for the end of 2026 to let brands fully automate the creation and targeting of their ads.

For an advertiser paying an agency, the promise looks like a saving. It glosses over one point, though: in this model, the platform that sells the ad space is also the one that sets the price of each bid and counts the sales it attributes to itself.

That is where we start when we split the work between platform, AI tools, expert and advertiser. Our grid rests on two criteria: who holds the information needed to do the task, and who has an interest in it being done well. The grid is ours and does not reflect any market standard.

Manual bidding is no longer billable

On a Google Ads campaign using Target CPA, Target ROAS or Maximize conversions, Smart Bidding sets the amount of each bid at the moment the auction happens. Google says it relies on signals available at auction time (device, location, time of day, remarketing list membership, the query typed), which no operator can use auction by auction. At Meta, Advantage+ campaigns likewise hand audience, placements and budget allocation to the delivery system.

An agency that still bills for daily bid optimisation is therefore selling work the platform already does, with signals the agency cannot see. Take an advertiser spending €40,000 a month on Google Ads and Meta and paying its agency 10% of that spend, or €4,000 a month. If bids, placements and part of the audiences are automated, those €4,000 have to pay for something else, and it is fair to ask what.

One detail of this fee model deserves attention. Indexed on spend, the agency's fee rises when the budget rises, just like the platform's revenue. And each platform optimises within the budget it is given; it will never tell you that part of that budget would be better used elsewhere, or not at all. If nobody at the table has an interest in proposing a cut, that proposal has to come from the advertiser.

A 70% score to maintain

Google reserves Google Partner status for agencies whose manager account reaches an optimization score of at least 70%. Google presents this score as an estimate of how well the account is set to perform: it moves as the recommendations suggested by Google Ads are applied or dismissed. The programme page states that an agency does not have to follow every recommendation and remains free to dismiss them based on its own assessment and its clients' goals.

So the system leaves room for refusal. Still, an agency running dozens of accounts under one manager account has a direct interest in keeping the score above the threshold, and the fastest way there is to apply. Common recommendations include adjusting CPA or ROAS targets, adding broad match keywords or extending delivery to new inventory: all changes that shift part of the budget.

This does not mean a Google Partner agency applies everything without review. The optimization score, on the other hand, measures the gap between your account and what Google recommends, nothing more. If your agency holds the status, ask how many recommendations it dismissed last quarter, and why. We explain which ones can be applied without approval in our article on automating Google Ads.

Margin lives in no ad account

Take an online retailer selling two products at €100. The first leaves a €60 margin, the second €15. If the conversion action passes the sale price as its value, a Target ROAS campaign values both sales identically, and will happily spend more on the second if it sells more easily. Observed ROAS can then rise while the margin the campaign generates falls.

Google Ads can calculate gross profit, provided it is given the cost of goods sold in the Merchant Center feed. The platform does not make it up: without that data, it optimises on the revenue it receives. The same goes for stock, the promotional calendar, return rates by product or the sales actually collected in the back office or CRM. This information belongs to the advertiser, and no AI tool plugged into the ad platforms will find it on its own.

Turning that data into settings is the expert's job, at an agency or in-house: choosing the optimisation event, defining the conversion value (price, margin, estimated value of a lead), setting a ROAS target consistent with the break-even point. We explain why a high ROAS can hide a loss in our article on the limits of ROAS.

Who does what, task by task

Here is the split we recommend for an advertiser investing on at least two platforms. It reads in two columns: who carries out the task, and who must approve it.

Task	Carried out by	Approved by
Amount of each bid	Platform (Smart Bidding, Advantage+)	Nobody, case by case
Headline, visual and format variants	Platform or AI tool	Advertiser, for brand and legal mentions
Monitoring of conversion tracking and spend pacing	AI tool, continuously	Expert, on alert
Weekly report and reconciliation with the back office	AI tool	Expert
Search term review and negatives	AI tool, as a proposal	Expert
Platform recommendations	Expert, one by one	Advertiser for those that extend delivery
Optimisation event and conversion value	Expert	Advertiser, who supplies the margin
CPA or ROAS target	Expert	Advertiser
Budget split between platforms	Expert, based on tests	Advertiser
Total budget, objectives, commercial calendar	Advertiser	Leadership

The dividing line follows how often a decision is made and how easily it can be reversed. What is decided thousands of times a day, can be corrected quickly and is measured with the platform's own signals belongs to the platform. What is decided once a month, commits the budget for several weeks and depends on data the platform does not have stays with someone who knows the account.

The budget split between platforms deserves a separate mention. Google and Meta can both attribute the same sale to themselves, each according to its own attribution rules. Only a test with a control group, such as a Meta Conversion Lift study or a geo test, measures what each platform actually caused. The platforms provide these tools, but the protocol (duration, regions, decision threshold) should be set by someone who is not judging their own case.

Where an AI tool saves time

Between the platform and the expert, AI tools (conversational assistants, agents, scripts) take over what used to fill a good part of an agency's time without involving a decision: pulling data from three platforms, aligning campaign names, comparing attributed conversions with back-office orders, spotting a conversion tag that stopped firing on a Friday evening. If your weekly report takes half a day of data handling, that half-day can go, and the expert spends it on what the report shows.

Two limits frame this use. A tool connected to the platforms takes on their attribution: if it compares two ROAS figures that partly count the same sales, its recommendation inherits the error. And a tool with write access can apply a mistake at scale. We would always start with read-only access, and we explain why in our article on AI agents in paid media. In the BRAIKE Suite, this work falls to Sentinel for monitoring and to CheckUp for audits.

Before renegotiating with your agency

Open the change history of your Google Ads and Meta accounts for the last 30 days, and sort each change by author: the platform (auto-applied recommendations), your agency, your team. If the agency's column is limited to budget tweaks and accepted recommendations, you know what its fees correspond to.

Then ask your agency three questions. Which recommendations did it dismiss, and why? Which source does it use to count sales: the platforms, GA4 or your back office? What incrementality test has it run in the past year to justify the budget split between platforms?

We do not recommend replacing an agency with an AI tool if nobody in-house can set a ROAS target from the margin or read a control-group test. Conversely, paying a percentage of spend for bids that Smart Bidding already sets no longer makes sense. An agency now earns its fee through its judgement on infrequent decisions and its ability to challenge the platform when its figures do not match your sales. Those two points are what we build our AI-augmented paid media agency service on.

Frequently asked questions

Can a small advertiser do without an agency?

Often, yes. On a single platform, with a Performance Max or Advantage+ campaign and reliable conversion tracking, the platform handles most of the execution. What you need to keep is a regular check of tracking and of the recommendations applied, which automated monitoring and a few expert hours per quarter can cover.

Should you bring it in-house rather than switch agencies?

It depends on whether someone on your side can turn margin into conversion value and read an incrementality test. If so, AI tools cover much of the remaining execution. If not, that skill is precisely what you are buying from an agency, far more than its hours spent in the interface.

Information checked on 13 October 2026.