What is Cost cap?

A cost cap tells the bidder to keep the average cost per result at or below a stated figure. Within that constraint it is free to pay whatever an individual auction requires. TikTok names the control Cost Cap; Meta reached the same arrangement through a control it once called cost cap and now labels a cost per result goal. The behaviour is what this page describes, under whichever label the platform in front of you uses.

That freedom is the whole design. An expensive auction that produces a result is acceptable if a cheaper one elsewhere balances it. A control that refused the expensive auction — a bid cap — would give up the result and the balance with it.

The trade is that the advertiser cannot say what any single result will cost. Some will cost a multiple of the cap. A campaign hitting its cap on average with a wide spread underneath is working correctly, and the spread is not a defect to be tuned away.

Three controls, three different promises, and they are frequently discussed as though they were strengths of one setting.

Bid cap — no single auction may be bid above this figure. Binds every auction. Strictest, least volume.

Cost cap — the average cost per result may not exceed this figure. Binds on aggregate. Looser, more volume, no per-result guarantee.

Spend cap — total spend may not exceed this figure. Binds on the total and says nothing about the price of anything.

An account that wants a ceiling on what any one result may cost has not asked for a cost cap. An account that wants a predictable cost per result on average, with volume, has.

The characteristic behaviour of a cost cap is that delivery flattens before the budget is exhausted, and understanding why prevents most of the frustration with it.

The cheapest results in any market are bought first. Each additional increment of budget reaches people who are further from converting and more expensive to reach. At some point the bidder cannot buy more without dragging the average past the cap, and it stops buying.

Spend plateaus. The average sits at or just under the cap. Volume does not grow with additional budget.

That is not a malfunction and it is not a bug to be worked around by raising the budget. What has been discovered is the amount of volume available at that price. The options are to accept it, to raise the cap because the business can afford more than it thought, or to widen the audience so there is more cheap volume to find.

The option that does not work is raising the budget, which is nevertheless what most accounts try first.

From the business, as with every ceiling.

Take the contribution a customer produces. Multiply by the rate at which the result being bought becomes a customer. That product is the most a result can cost before the transaction stops paying for itself, and the cap belongs under it by however much of the contribution the business needs to keep for everything else.

Two common sources that are both wrong. A benchmark describes businesses whose margins are invisible. The campaign's own recent average describes a period rather than a ceiling, and setting the cap there removes the headroom that makes an averaged control useful — the campaign can no longer buy the expensive result that a cheap one would have paid for, and it behaves like a bid-capped campaign nobody intended.

The diagnostics differ from an uncapped one.

Spend against budget is the first number, not cost against cap. A campaign well under budget is telling you the cap is binding, which is information about the price of volume rather than about efficiency.

The spread of costs around the average is expected to be wide. A narrow spread in a capped campaign frequently means the cap is set so high it is not binding at all.

Volume trend against cap changes. Raising the cap by a modest amount and watching what happens to volume is the cheapest available measurement of the shape of the market the campaign is buying in. If volume jumps, there was a cliff just above the old figure. If it barely moves, the constraint is the audience rather than the price.

Learning behaviour. Like every automated control, a cap change resets the model's predictions and produces a period of unstable delivery. Changing it repeatedly in response to that instability keeps the campaign permanently unsettled.

A real ceiling with a need for volume. This is the case it was built for: the business knows what a result can cost and wants as many as that price allows.

Scaling with discipline. A campaign growing against a cap will show exactly where the volume stops being affordable, which is a more useful answer than an unbounded campaign's rising average.

Mixed auction conditions. Where the cost of results varies a lot by audience, placement or time, an averaged control captures the cheap opportunities and pays for the dear ones, which a per-auction cap cannot do.

When a hard per-result ceiling is genuinely required. Use a bid cap and accept the volume loss.

When conversion volume is too thin. An average over a handful of results is not an average; it is noise, and a cap on it will produce erratic delivery.

When the conversion is mismeasured. A cap holds the average cost of whatever event was nominated. If that event is not the one the business sells, the cap is protecting a number nobody should care about.