What is Account structure?

Account structure is how the containers are arranged: which campaigns exist, what sits inside each, and where the budgets and targets are set.

For most of the history of paid advertising this was largely a filing decision. Bids were set by hand at whatever level the advertiser chose, so a highly granular structure meant fine control and a coarse one meant less. The containers organised the work.

Under automated bidding and pooled budgets the same containers do something different. They decide what the platform chooses between, and where conversion data pools for the models to learn from.

That turns structure from an organisational preference into an allocation decision, and it is why advice about account structure written before that shift reads as reasonable and produces accounts that underperform.

Every structural question comes down to a tension between two facts.

Splitting divides the signal. Each container needs enough of its optimisation event for the bidding to predict with. An account fragmented into many small campaigns or ad sets gives each of them a fraction of the conversions, and several will sit permanently below the threshold, delivering erratically.

Pooling hands over the choice. Things in the same container with a shared budget are rivals, and the platform allocates between them on its own prediction. It cannot see reasons that are not in the objective: a committed budget, a different margin, an audience being kept warm, a market with strategic weight.

Neither force wins in general. The resolution is a test that can be applied to any proposed split.

Separate when the difference is one the platform cannot see. Pool when it is not.

Different ceilings, committed budgets, different objectives and strategic roles are invisible to the optimiser. Different audiences with the same economics are not, and it will handle them better than a manual split.

The highly segmented structures that dominated search advertising — one keyword per ad group, campaigns split by match type, ad sets per interest — were correct responses to conditions that no longer hold.

Match types were literal, so a keyword defined its queries and an ad could be written against that exact phrase. Bids were manual, so a container per keyword meant a bid per keyword. Conversion data pooling did not matter because no model was learning from it.

All three changed. Semantic matching means a keyword no longer defines its queries precisely. Bidding is per auction. And the models are the main thing consuming conversion data, which fragmentation starves.

What survived is the relevance argument: an ad speaks better to queries that share an intent, and that remains true. What did not survive is granularity for control, because the control moved.

A few checks say more than a long audit.

Optimisation events per container. Count them over the window. Where most containers are below the platform's own threshold, the structure is the problem and everything else will be blamed for it.

Brand mixed with prospecting. If branded traffic sits inside a campaign with non-branded, the account's reported performance is being carried by traffic that was arriving anyway.

Prospecting pooled with retargeting. Under a shared budget, retargeting wins and new-customer acquisition quietly stops.

Overlapping audiences in one campaign. The platform's allocation is being driven by a contaminated comparison.

Campaigns nobody can explain. Names referencing dates and tests long past, still spending.

Structural change is material: it resets the models and produces a period of unstable delivery. That makes a restructure worth doing deliberately and rarely.

Decide the shape first, on the separation test above, and write it down before touching anything.

Move rather than rebuild where the platform allows it, since history and learning are worth preserving.

Stage it. Changing everything in one week means the whole account is unsettled at once and nothing can be attributed.

Hold budget steady through the transition, so that delivery changes can be read as consequences of the structure rather than of the money.

Wait. At least a full conversion window before judging, and longer on accounts with slow cycles.

Campaigns separated by the invisible differences — brand against non-brand, prospecting against retargeting, markets with committed budgets or different economics.

Inside them, few containers, each broad enough to accumulate real volume.

Exclusions applied so the containers do not overlap.

Reporting dimensions carried by naming conventions and labels rather than by structural splits.

Testing done in the creative, where the remaining variance is, rather than in the targeting.

It is a simpler account than the discipline used to build, and the simplicity is what makes the automated parts work.

It is worth being clear about the limits, because restructuring is a satisfying activity that can substitute for harder work.

A wrong conversion definition. No arrangement of containers improves an account optimising toward the wrong event.

Weak creative. Auctions rank on expected value, and structure does not make an ad more compelling.

An unaffordable target. If no auction clears at the target, no structure makes it clear.

A market that is too small. Splitting a limited audience into more containers does not produce more people.

An account that has been restructured three times in a year and has not improved is usually an account with one of those four problems, and each restructure has cost it a period of unstable delivery on the way to not fixing it.

Occasionally the question is not how to arrange one account but whether to run several.

The reasons that justify it are administrative rather than performance-related: separate legal entities, separate billing, agencies or teams that must not see each other's data, markets with different compliance regimes, or a business selling under distinct brands that share nothing.

The reason that does not is a desire for cleaner reporting, because splitting accounts divides conversion data at the highest possible level and each account then has to learn independently. Where one entity is advertising one business, one account almost always performs better than two.