By November 2028, the best Meta ads takeover proposals will lead with a do-not-touch list: the campaigns, ad sets, audiences, and tracking the incoming provider agrees to leave alone before it promises to optimize anything.
As execution gets cheaper, restraint gets more valuable. A proposal that names what it will not change recognizes the learning you have already paid for. A promise to build more ads does not.
Consider one DTC toy seller’s account. The business was spending roughly $500 a day at 1.7 to 1.9 ROAS when an agency took over on August 18. Nineteen days later, the owner’s audit counted:
- 1,004 logged changes, including 20 new campaigns and 50 ad sets.
- 102 ads built, with 74 subsequently switched off.
- A median of 31 hours between ad creation and shutdown.
- One ad set that lasted 9 minutes and one campaign that lasted 40.
- Two staff members reversing each other’s decisions: one switched an ad set on at 11pm; another switched it off at 1:23am.
This is an unverified owner’s account, not a dataset. It cannot tell us how often takeovers go wrong. It does show the operating problem behind my forecast: cheaper execution makes it easier to change an account faster than anyone can judge the changes.
Meta’s learning phase makes that problem expensive. Meta documents significant edits, including changes to creative, targeting, optimization events, and bid strategy, that can send an ad set back into learning. Its guidance describes about 50 optimization events, usually within a week of the last significant edit, as the volume needed to complete that phase.
Every unnecessary reset asks the account to pay again for something it had already learned. Making the edit is cheap. Absorbing the uncertainty afterward is not.
When a change took forty minutes of manual setup, the work itself imposed some discipline. Scripts, bulk tools, and AI operators remove much of that friction. They do not remove the need to decide whether the change belongs in the account.
That is the mechanism behind all three predictions. The scarce service is no longer the ability to act. It is knowing when not to.
Prediction 1: takeover proposals will name what stays untouched
By November 2028, serious takeover scopes will name the campaigns and ad sets protected for the first 30 days, specify what may change during that period, and state the condition that justifies an exception.
I expect a schedule attached to the proposal, not a paragraph about “respecting what works.” A do-not-touch list has names on it. A reassuring paragraph has adjectives.
The mechanism is financial. Buyers can inspect what happened to inherited campaigns after a takeover. As those audits become part of judging a provider, damage to working campaigns becomes harder to bury beneath a list of new deliverables. Proposals without exclusions will lose on the cost of disruption, not just on principle.
The toy seller’s audit is the warning sign today. The owner could reconstruct what had been created, paused, and reversed in nineteen days. The agency’s activity was visible. The rationale was much harder to find.
A technical reply in that thread described another intervention: duplicating a winning ad set, switching the duplicate to cost-cap bidding, and pausing the mature original. The reply reported roughly 70% higher CPM on the duplicate. That figure belongs to the same unverified account, but the decision is the point. Replacing something that works with something that needs to establish itself is not automatically an optimization.
Working campaigns also carry context that a performance column does not explain. A budget rule may reflect a stock constraint. A campaign that looks inefficient may serve a business role the incoming team has not yet understood. Discovery has to come before surgery.
The proposal worth signing identifies the roles the provider understands and the campaigns it is deliberately leaving alone. The one I would reject promises to optimize everything it finds. That promise treats discovery as free and sends the buyer the bill later.
What would prove me wrong: by November 2028, agency-selection guides and takeover scopes still sell volume rather than protection. Flighted’s 2026 guide describes tiers that include under $4,000 a month for a freelancer, $4,000 to $8,000 for media buying and twenty-plus ads a month, and $8,000 upward for strategy. If the market still leads with ads-per-month tiers rather than published exclusions, this prediction has not landed.
Ask for the protected inventory before you ask how much new work the provider can produce.
Prediction 2: providers will disclose intervention rules before receiving access
By November 2028, credible proposals will include a written intervention policy before the provider receives editing access. It will name the metrics that matter, the evidence required to pause or scale, and the response when the data is ambiguous.
Not a methodology deck. Rules for when someone is allowed to spend your money differently.
The mechanism is the gap between visible activity and defensible decisions. A change log can show that someone worked. It cannot, by itself, show that the work was necessary. Once buyers can inspect activity, the next question is what governed it.
There is evidence of that scrutiny already, even outside Meta. A business reviewing twelve months of a $10,000-a-month Google Ads account reported agency activity on an average of 3.6 days per month. Of 219 changes, 70 were budget adjustments. One month contained five changes in a fifteen-minute session.
The agency reportedly claimed that changes made through its internal software would not appear in the platform’s change log. Commenters disputed that explanation. Again, this is an owner’s report, not an industry benchmark. The useful signal is the buyer’s question: what, exactly, was being managed?
Low activity is not proof of neglect, just as 1,004 changes are not proof of diligence. Both counts need a decision policy behind them. Otherwise, the provider can defend silence as patience and churn as optimization. Convenient business model.
For Meta, that policy has to reflect conversion volume. An account generating 200 purchases a week and one generating 20 do not have the same room to absorb repeated learning-phase disruption. A generic instruction to “test more” ignores that difference.
Nor does creating another ad set make the learning requirement disappear. The question is whether the proposed intervention earns its cost, not whether the provider can find somewhere new to run it.
An intervention policy worth signing spells out:
- The trigger: the conversion or cost-per-result evidence that warrants a pause, scale, or budget move.
- The learning cost: how the account’s event volume affects the decision to make a significant edit.
- The ambiguous case: what stays unchanged while the evidence is inconclusive.
This is where restraint becomes a service rather than an excuse. “We left it alone” needs a reason just as much as “we rebuilt it.”
What would prove me wrong: by November 2028, proposals still describe optimization as a deliverable without stating intervention thresholds. If buyers continue accepting “we monitor and optimize” in place of rules, I have overestimated the market’s appetite for accountability.
Ask the provider to explain when it will act, not just what it knows how to do.
Prediction 3: one person will own conflicting decisions
By November 2028, credible takeover proposals will name one person accountable for conflicting decisions and the account’s change log. A name in the contract, not a roster in the deck.
The mechanism is coordination. Two operators can each make a defensible local decision and still produce an indefensible account. Without a shared policy and a clear decision-maker, the last person to click wins.
The toy seller’s reported 11pm switch-on and 1:23am switch-off is the problem in miniature. Both people may have been working hard. Nothing in the process established whose decision should stand.
The warning is not that short pauses automatically erase an ad set’s accumulated learning. It is that contradictory interventions make the account harder to interpret and leave the buyer unable to distinguish a deliberate decision from an internal disagreement.
One owner changes that. Specialists can argue about the right move, but the account needs one accountable decision when their recommendations collide.
Daily monitoring does not require daily account surgery. Watch the account every day. Edit when the evidence policy says the expected gain justifies the disruption. Watch daily, edit rarely is an operating discipline, not permission to ignore the account.
That distinction is also what I would defend in groas’s operating model: a named human strategist owns direction and accountability while specialized models handle continuous execution. Logged actions and their reasoning make the work inspectable. Faster execution should make ownership clearer, not dissolve it into “the system.”
A busy change log without a governing decision is still just a busy change log.
I am taking a position some good buyers will reject: a team of specialists with independent authority to change the same account is a liability in the first 30 days of a takeover. Talent does not resolve conflicting instructions. A documented policy and one accountable owner do.
There are early signs of movement. One agency-selection guide already frames the choice as a governance and operating-model decision. Other agency roundups still emphasize generic red flags. The vocabulary is moving. The contract is the harder test.
What would prove me wrong: takeover contracts in November 2028 still offer a roster without a single decision owner, and those teams preserve mature ad sets through the first 30 days as reliably as single-owner teams. If shared authority produces the same protection, my coordination argument loses.
Count the accountable names. The answer should be one.
What I would put in the next takeover contract today
A forecast that cannot fail is a horoscope. These three have a deadline and visible tests: published exclusions, stated intervention thresholds, and one accountable owner. I am betting that credible providers will sell those protections explicitly by November 2028.
You do not have to wait until then to buy them. Before handing over editing access, I would ask the next Meta provider for five things in writing:
- A do-not-touch list. Named campaigns, ad sets, audiences, and tracking protected for the first 30 days, plus the condition that justifies an exception. “We preserve what works” is not a list.
- An intervention policy. The evidence threshold, expressed in the account’s own conversion volume and performance metrics, that triggers a pause, scale, or budget move. Not a best-practices slide. Your numbers.
- One named decision owner. The person who resolves conflicting recommendations and answers for the change log. Specialists can contribute; accountability cannot be a group project.
- Change-log rights. Your right to inspect the account’s change history, plus the provider’s obligation to record the reason behind each change. A timestamp tells you what happened. The explanation tells you whether anyone was steering.
- Scoped access and a clean exit. Change rights where the provider needs them, view-only access where it does not, and revocation when the engagement ends.
The last point needs a distinction the contract cannot gloss over: account permissions are not a campaign-level do-not-touch list. Meta’s ad-account roles distinguish editing from viewing, but that does not mean you can give the same provider editing rights on new test campaigns and enforce view-only rights on mature ad sets within that account through those roles.
The freeze therefore needs to exist in the operating policy and the contract. Do not mistake a promise for a technical lock.
When requesting shared access, review Meta’s partial-access options rather than treating full control as the default. Establish where the account lives, who controls it, and who can remove access. Permission changes take effect immediately, so the exit process should name the person responsible for making them.
None of this is exotic. It separates ownership, permission, and decision authority instead of pretending they are the same thing. It also gives the buyer something more useful than a promise of full transparency: a way to identify who acted, why they acted, and whether they stayed inside the agreed scope.
If I am right, the providers worth hiring in 2028 will make the most uncomfortable promises on paper: leave these alone, act only under these rules, this name owns the outcome. If I am wrong, you will still have five protections in writing and be harder to take advantage of than the buyer who signed for optimism.
The toy seller’s audit counted 1,004 changes. That is not the number I would want a takeover proposal to celebrate. Spend an afternoon building your do-not-touch list instead. It costs less than one relearning phase.

