A social feed contains a small number of advertising slots and a very large number of advertisers competing for them. The selection rule is an auction, but the ranking inside it is unusual.
Bids are combined with two predictions
The platform estimates how likely the viewer is to take the action the advertiser is paying for, and separately how likely the viewer is to find the advertisement worth seeing.
Those estimates are multiplied against the bid to produce a ranking score. An advertisement predicted to perform well can therefore outrank one that offered more money.
The logic is straightforward from the platform's side. Revenue per slot over time depends on people continuing to use the feed, so an irritating advertisement is expensive even when it pays well.
Negative signals carry unusual weight
Hiding an advertisement, reporting it or marking it irrelevant feeds directly into that second prediction. These actions are rare, which makes each one informative.
A small number of such responses can raise an advertiser's effective cost noticeably, because the platform now expects a similar reaction from similar viewers.
This produces a pattern advertisers often misread as a budget problem. Costs rise, delivery slows, and the underlying cause is the reaction rather than the bidding.
The learning phase is a data requirement
A new advertisement set has no history, so the platform's predictions begin as broad averages and sharpen as conversions accumulate.
Until enough events have been observed, delivery is comparatively erratic and results are unreliable as a guide. Judging performance during this period usually leads to premature changes.
Editing the set restarts the process, which is why frequent optimisation can leave a campaign permanently in the unstable stage it was being edited to escape.
Creative is a targeting instrument
Delivery systems find the people most likely to respond to a specific advertisement, so the advertisement itself shapes who sees it as much as the audience settings do.
Two creatives aimed at the same audience will reach visibly different subsets of it, because each one attracts different early responses that the system then extrapolates from.
This is why broad targeting with strong creative often outperforms narrow targeting with weak creative. The system is capable of finding the audience if the advertisement gives it something to work with.
Competition varies by objective, not only by audience
Advertisers chasing purchases compete for a much narrower slice of viewers than advertisers chasing reach, and prices in that slice climb accordingly.
Two campaigns targeting an identical audience can therefore face entirely different price levels, determined by which optimisation goal each one selected.
Changing the objective changes the market being entered, which makes it one of the most consequential settings available and one of the least examined.