The price of reaching a given audience changes hour by hour, sometimes by a wide margin. The swing comes from two curves that are not aligned with each other.
Supply follows human routine
Available impressions rise and fall with when people read, watch and scroll. Commuting hours, lunch breaks and the evening produce distinct peaks in most markets.
Overnight the supply collapses, since fewer people are awake and browsing. The inventory that does exist skews towards audiences an advertiser may not have planned for.
None of this is under anyone's control. It is simply the aggregate rhythm of a population, and it repeats with enough regularity to be forecast.
Demand follows budgets and pacing rules
Buying platforms are told to spend a fixed amount over a fixed period, and most of them spread that spending to avoid exhausting it early.
Pacing algorithms therefore bid more aggressively when they are behind schedule and retreat when they are ahead. The bidding pressure on any given hour depends on what happened in the hours before.
Because many advertisers use similar pacing logic, these adjustments happen in rough unison, which amplifies the swing rather than smoothing it.
Budget resets create a predictable surge
Daily budgets reset at a boundary time, and a large number of campaigns become active again at the same moment. Competition spikes immediately afterwards.
The same pattern appears at the start of a month and at the start of a quarter, when new budgets are released and unspent money has to be committed.
Advertisers who can shift delivery away from those boundaries often buy the identical audience for meaningfully less, simply by not competing during the crush.
Not all hours contain the same audience
Cheap hours are cheap partly because the people available then convert less often. A low price on a poorly performing hour is not a saving.
This is why blanket instructions to buy overnight inventory tend to disappoint. The cost per impression falls and the cost per outcome frequently rises.
The useful question is whether a specific audience behaves differently at a specific time, which requires measuring outcomes by hour rather than assuming the pattern.
Dayparting is a blunt instrument
Switching a campaign off during expensive hours removes the competition but also removes the audience, and the machine learning behind delivery loses data it was using to improve.
Bid adjustments by hour are gentler, since they change what the campaign is willing to pay rather than whether it participates at all.
The stronger approach in most cases is to let the system optimise against outcomes and intervene only where a genuine, repeated difference in results by hour can be demonstrated.