- Google’s 30 conversions, and 50 for Target ROAS, describe how long a period to measure over. Not a bar a campaign must clear. [1]
- Google’s own Target CPA page says advertisers can start with no conversion history. [2]
- The real cost of splitting is arithmetic: you divide the evidence and multiply the things needing evidence.
- Three honest reasons to split: different economics, different budget owner, different measurement need.
- Brand and non-brand stay apart, because blended targets let cheap conversions subsidize expensive ones.
- Google defines an asset group as creatives on a theme or audience, and says nothing about how many. [3]
- Before-and-after readings of a restructure are observational, and observational estimates run high. [6]
Every account restructure in this industry gets argued from two numbers. Thirty conversions per campaign per month. Fifty optimization events per ad set per week. They appear in decks, in audits, and in the confident paragraph explaining why eleven campaigns are about to become three.
Go looking for where either number is stated as a requirement and the ground gives way.
What Google actually says
Google’s Smart Bidding documentation describes strategies that “use Google AI to optimize for conversions or conversion value in every auction,” and when it reaches numbers, it uses them for something specific: it advises “measuring performance over longer time periods that have at least 30 conversions, such as a month or longer (50 conversions for Target ROAS).” [1]
Read that twice, because the distinction is the whole article. Those figures describe the size of a window you should look at before you believe what you are seeing. They are advice about reading results. Somewhere between the documentation and the industry, they were promoted into a structural threshold that campaigns must clear, and then used to justify a decade of restructures.
The Target CPA documentation goes further in the other direction. It states plainly that advertisers can start using Target CPA with no conversion history. [2] Whatever else is true, the platform is not refusing to bid until you feed it thirty of anything.
Meta is quieter still. Its developer documentation on campaign-level budget says the system “automatically and continuously finds the best available opportunities for results across your ad sets and distributes your campaign budget in real time to get those results.” [5] That is a description of a mechanism. It is not a threshold, and it is not a promise about what happens when you run six ad sets on a small budget.
The heuristic is roughly right for the wrong reason
None of this makes consolidation wrong. It makes the argument for it different, and the different argument is sturdier because you can check it yourself.
Thirty conversions is approximately the point at which a month of data starts to mean something. Below it, the difference between a good month and a bad one is mostly noise, which means you cannot tell whether a campaign is working, cannot tell whether a change helped, and cannot tell whether the target you set is achievable. That is a real and serious problem. It is a problem about your ability to know things, not a gate the platform enforces.
Once you frame it that way, the cost of splitting becomes arithmetic anyone can do at a desk.
A campaign produces 60 conversions a month. Split it three ways and you have three campaigns at 20 conversions each. The spend did not change and the demand did not change. What changed is that you now hold three questions you cannot answer instead of one you could, each of them needing its own history before it says anything, each of them re-learning after every edit. Split it six ways and you have six.
That is the entire cost, and it is why the instinct to consolidate is usually correct. Not because a system refuses to work below a number, but because you have taken one legible thing and turned it into several illegible ones, in exchange for a report you could have produced with a filter.
The three reasons that survive
| Reason to split | Does it hold? | Why |
|---|---|---|
| The economics genuinely differ | Yes | A high-margin line and a thin-margin line cannot share one target without one of them being wrong |
| A different person owns the budget | Yes | A budget somebody must be able to protect needs a container the system cannot spend out of |
| The segments must be measured apart | Yes | If a decision depends on reading them separately, they cannot share a bidding unit |
| Reporting looks cleaner | No | That is a saved filter, not a campaign |
| It feels more controlled | No | The control is illusory and the fragmentation is real |
| It has always been that way | No | The account inherited a structure from a platform that no longer exists |
The first three are all business facts the platform genuinely cannot infer. That is the test worth applying to any proposed split: does this encode something the system could not work out on its own? Margin can. Organizational boundaries can. Measurement requirements can. Your preference for tidy naming cannot.
Brand and non-brand
This is the one split almost nobody should merge, and the reason is the same arithmetic seen from the other end.
Branded search converts at a rate prospecting cannot approach, at a cost prospecting cannot approach, from demand that mostly already existed. Put both in one bidding unit against one target and the cheap conversions subsidize the expensive ones. The campaign hits its target comfortably, the report looks healthy, and the prospecting you were actually trying to buy quietly gets less than it needs, because the system has an easier way to satisfy you.
You also lose the ability to read either one. Two populations with different economics averaged into one number produce a figure that describes neither, which is the same failure as any other blended metric and has the same fix: separate the things that behave differently, and only those.
Performance Max asset groups
Google’s definition is unglamorous and worth taking literally. An asset group is “a collection of creatives centered on a theme or target audience.” [3] The documentation offers no guidance on how many to run, which is itself informative.
The failure mode is using asset groups as a substitute for campaign controls they were never going to provide. If the reason for a second asset group is that the creative genuinely differs, that is what they are for. If the reason is to control budget, targeting, or reporting at a level the campaign will not give you, the asset group will not deliver it, and you have added a container that dilutes creative rather than one that concentrates it.
Knowing whether the restructure worked
Here is the part that gets skipped in every restructure deck ever written.
A before-and-after comparison of a rebuilt account is observational. It sits inside seasonality, inside whatever else changed that quarter, and inside the fact that a restructure is usually accompanied by fresh attention from people who were not paying attention before. Research across 663 large-scale advertising experiments found that even sophisticated non-experimental methods substantially overstated true effects, turning median lifts of 29%, 18%, and 5% into estimates of 83%, 58%, and 24%. [6]
Applied here, that means the confident chart in the post-restructure review is the least trustworthy artifact in the process. If a restructure matters enough to do, it matters enough to stage: move part of the account, hold the rest, and give yourself something to compare against that is living in the same week as the change.
The review, in an hour
Open the account and answer four questions per campaign, in order. How many conversions did this produce last month, and is that enough to read a result from? What business fact does its separate existence encode, in one sentence, without using the words reporting or control? If it were merged into its nearest neighbor, what decision would become impossible? And when was this structure last chosen deliberately, rather than inherited?
Google publishes the structural ceilings that genuinely constrain you, and they are far higher than anything you will hit for the right reasons. [4] If a limit is not in that document, it is not a limit, it is a habit.
Most campaigns fail the second question. A few fail it and should still survive, because the honest answer is that nobody has the appetite to merge them this quarter, and a structure you will actually maintain beats an optimal one you will abandon. That answer is fine, as long as it is said out loud rather than dressed up in a threshold that was never a threshold.
The thing to stop doing is quoting the number. Whatever your account needs, it does not need thirty of anything. It needs each unit to produce enough evidence that a human can tell what is happening, and that bar is one you set, from how quickly you need to know, and how expensive it is to be wrong. Then the value you send and the creative you feed it start to matter, which is where the work was always going to be.
Sources
- Google Ads Help · About Smart Biddingauction-time bidding, and the recommendation to measure over periods holding at least 30 conversions, 50 for Target ROAS
- Google Ads Help · About Target CPA biddingadvertisers can start using Target CPA with no conversion history
- Google Ads Help · About asset groups in Performance Maxan asset group is a collection of creatives centered on a theme or target audience, with no stated guidance on quantity
- Google Ads Help · About Google Ads account limitsthe published structural ceilings, as distinct from recommendations
- Meta for Developers · Marketing API: Advantage campaign budgetcontinuous real-time distribution of one campaign budget across ad sets
- arXiv · Close enough? A large-scale exploration of non-experimental approaches to advertising measurement663 large-scale Facebook experiments: median true effects of 29%, 18%, and 5% estimated as 83%, 58%, and 24%
Frequently asked questions
Does Google publish a minimum conversion volume for Smart Bidding?
Not as a structural requirement. Its Smart Bidding page recommends measuring performance over periods holding at least 30 conversions, and 50 for Target ROAS, which is guidance about your evaluation window rather than a bar a campaign must clear. Its Target CPA page states that advertisers can start using Target CPA with no conversion history at all.
So is the thirty-conversion rule wrong?
It is a reasonable heuristic wearing a badge it never earned. Thirty conversions is roughly what you need before a month of data says anything, so a campaign below it is genuinely hard to manage. That is an argument about how confidently you can read a result, not a switch the platform flips.
What does splitting a campaign actually cost?
You divide the evidence and multiply the things that need evidence. Split one campaign producing 60 conversions a month into three, and you now have three campaigns at 20, each of which needs its own history before anyone can tell whether it is working. The spend did not change. The number of questions you cannot answer went up threefold.
When is splitting genuinely justified?
When the economics differ, when a separate person controls the budget, or when you need the segments measured apart. Those are business facts the platform cannot infer. Everything else, including reporting convenience, is usually a request for a filtered view rather than a reason for a separate campaign.
Should brand and non-brand stay separate?
In almost every account, yes. They have different intent, different marginal value, and different response to spend. Mixing them lets cheap branded conversions subsidize a target that then quietly starves the prospecting you were actually trying to buy, and it destroys your ability to read either one.
How many asset groups should a Performance Max campaign have?
Google does not say. Its documentation describes an asset group as a collection of creatives centered on a theme or target audience, and offers no guidance on quantity. The workable reading is that an asset group should exist when its creative genuinely differs, not as a substitute for the campaign-level controls it cannot provide.
Does campaign-level budget beat per-ad-set budgets on Meta?
It changes who makes the allocation decision. Meta’s documentation describes the campaign budget continuously finding the best available opportunities across ad sets and distributing spend in real time. That is only an advantage if your ad sets are genuinely comparable, because the system will move money toward whichever one currently looks best, whether or not you meant them to compete.
How do I know whether a restructure worked?
With difficulty, and this is the part everyone skips. Before-and-after comparisons of a restructured account are observational, and research across 663 large-scale advertising experiments found non-experimental methods substantially overstating effects. If a restructure is big enough to be worth doing, it is big enough to deserve a holdout or a staged rollout.
Mikołaj Salecki
Writes about media, tech, and AI business for people who actually run digital. Former agency lead. Skeptic of frameworks that read better than they perform.
