What is Campaign?

A campaign is the top-level object in an advertising account. Beneath it sit ad groups or ad sets, and beneath those the ads.

What lives at the campaign level is the set of decisions that bound everything underneath: the objective, the budget and whether it is shared, the bid strategy and its target, the schedule, and the broad location and language settings.

Those are not settings in the ordinary sense. They define what the platform is trying to do and what it is allowed to spend doing it, which means changing one is a material change that puts the campaign back into a period of unstable delivery while its model rebuilds.

Every platform asks what the campaign is for before anything else, and the answer determines more than the reporting.

It decides what the bidder optimises toward, which inventory the campaign is eligible for, which formats can be used, which optimisation events are available, and in several cases which reports exist at all.

It is also, in most platforms, not changeable afterwards. A campaign built for traffic cannot become a campaign built for sales; a new campaign has to be created, which means the history is lost and the learning starts again.

The consequence is that the objective deserves more thought than it usually gets. A campaign set to traffic in an account that wants sales will find people who click, faithfully and cheaply, and the mistake will not be visible in any of the numbers that campaign reports.

Under manual bidding, structure was mostly organisational. An account could be split a hundred ways and each bid set by hand, so the containers were for reporting and control rather than for the algorithm.

Under automated bidding and pooled budgets, structure decides what competes with what.

Two things in the same campaign, with a shared budget, are rivals for the same money and the platform chooses between them. The same two things in separate campaigns are not. That turns a filing decision into an allocation decision, and it is the single most useful thing to understand about modern account structure.

The rule that falls out is short: put things in the same campaign when you genuinely want the platform to choose between them, and in separate campaigns when you do not.

The instinct that served accounts for years — split everything, control everything — now has a cost that did not exist before.

Every campaign needs enough of its optimisation event for the bidder to learn from. Ten campaigns producing a handful of conversions a week each are ten campaigns permanently short of signal, and each will deliver erratically and report a cost figure that swings for no reason. Consolidating them gives one model something to work with.

The limit is that consolidation hands allocation to the platform, and the platform cannot see reasons that are not in the objective. It does not know that one market has a committed budget, that one product line carries half the margin of another, or that a campaign is being kept alive to warm an audience for a launch.

So the useful test for whether something deserves its own campaign is whether the difference is one the platform can see. Different margins, different ceilings, committed budgets and strategic roles are all invisible to it. Different audiences within the same economics are not.

Brand and non-brand. Branded search converts at rates no other traffic approaches, and mixing it with prospecting lifts the averages enough to hide a prospecting problem entirely. This is the most consistently worthwhile split in search advertising.

Prospecting and retargeting. Different jobs, different economics, and in one campaign the platform will fund retargeting because it converts more cheaply — which quietly stops the account acquiring anybody new.

Markets with their own budgets or their own economics. Where a customer is worth a different amount, the ceiling differs and a shared target misprices both.

Product lines with materially different margins, unless profit is being sent as the conversion value, in which case the bidder can weigh them itself.

Anything with an agreed budget that must not be reallocated.

Splitting by device, by small audience variations, by creative format, or by geography inside one market with the same economics. Each divides the conversion signal and buys control that automated bidding has mostly taken over anyway.

The reporting reason for splitting — wanting to see the numbers separately — is worth resisting specifically, because reporting can be segmented without structural separation, and the cost of fragmentation is paid in delivery rather than in reports.

Because the campaign holds the objective, the budget arrangement and the bid strategy, most changes at this level are material.

A new target, a large budget move, a change of strategy or a structural edit all reset what the model is predicting against, and the campaign delivers unevenly for a stretch afterwards while it rebuilds.

Two practices follow. Changes at this level should be batched rather than dripped, so the campaign pays for one unstable period instead of four. And they should be spaced far enough apart to be read, which in most accounts means at least a full conversion window between a change and the judgement of it.

The objective in particular cannot be changed at all on most platforms, which makes it the one decision worth slowing down for at the outset.

A campaign list says a great deal about how an account has been managed, and the patterns are recognisable.

Many small campaigns, each with a handful of conversions. Built for manual control and never consolidated. Every bid strategy in the account is working on too little data.

Brand mixed into prospecting. The account's reported performance is being carried by traffic that was arriving anyway, and the prospecting problem underneath it is invisible.

Campaigns named for dates long past. Nobody has retired anything, and some of them are still spending.

One campaign holding everything. The opposite failure: committed budgets and different ceilings all pooled, with the platform free to fund whichever converts cheapest.

None of these is fatal and each points at a specific piece of work, which is why the list is worth reading before the reports are.