Dayparting is dead, except where it is not
Algorithmic bidders already price time of day, so manual dayparting mostly loses. It still earns its place where the constraint sits on your side.
Dayparting used to be one of the first levers a buyer learned. Turn the campaign off overnight, push bids during the evening commute, hold budget for the weekend. On the networks that now take most game spend, that lever is not just weaker; it is usually harmful. The bidder already knows what hour it is. What it does not know, and cannot know, is anything about your side of the business, and that is the only place dayparting still has a job.
The thesis: scheduling ads by hour is dead as a performance tactic on self-optimising networks, and alive as an operational one. Teams that still daypart for CPI are fighting the model. Teams that daypart for reasons the model has no visibility into, such as a live event window, a server capacity limit or a support desk's hours, are using the lever correctly, and there are more of those reasons in games than in most categories.
Why the model beats you at time of day
A performance bidder is predicting, for every impression, the probability of the optimisation event. Hour of day and day of week are among the cheapest, most stable features it has. If your game converts better at 9pm than at 9am, the model has learned that from thousands of conversions across every advertiser in your genre, and it is already bidding more at 9pm on your behalf. When you additionally raise bids at 9pm, you are stacking a manual adjustment on a learned one and overpaying in the hour the model already favours.
Turning campaigns off overnight causes a different problem. Pacing algorithms spread a daily budget across the hours they are allowed to spend in. Cut the window in half and the same budget concentrates into fewer hours, which pushes the bidder into more expensive auctions to spend it, and often into lower-confidence users because it has run out of high-confidence ones in that window. CPI rises, the buyer blames the evening inventory, and the schedule gets tightened further.
There is also a relearning cost. Each time a schedule changes, the model's recent data becomes less representative of what it is about to face. Teams that adjust dayparts weekly are permanently keeping their campaigns slightly confused.
The second-order effect on competitors
An effect that receives almost no attention is what happens when a whole genre dayparts the same way. If most 4X and survival advertisers concentrate budget into the evening because their dashboards show evening conversions are cheaper, the evening auction becomes more crowded and the morning one thinner. The buyer who leaves the schedule open picks up the morning inventory at a discount and lets the model decide how much of it is worth having. In a category where everyone follows the same playbook, the open schedule is the contrarian position, and it is usually the cheaper one.
Where dayparting still earns its place
Dayparting works when the constraint is on your side of the install. The model optimises to your event; it has no idea what happens to users after they arrive, or what your studio can handle. Four cases come up repeatedly in games.
Live events with hard windows. If a limited-time event opens Friday and a user who installs on Thursday night sees it on first session while one who installs Saturday afternoon does not, the two cohorts have different first-session experiences and different day-7 retention. The model cannot see this. Shifting budget toward the window's opening hours is a legitimate use of scheduling, and the way to check it is to compare retention of cohorts installed inside and outside the window rather than CPI.
Capacity and support. A multiplayer launch with a server ceiling, or a title whose payment support operates in one time zone, has a real reason to slow acquisition overnight. The cost of a first session that fails is higher than the discount on the overnight impression.
Channels without a bidder. Apple Search Ads exposes scheduling and does not optimise to your deep events in the way a performance network does. Influencer and creator placements land at fixed times. On these, the buyer still owns the clock.
Genre session patterns as a hypothesis, not a rule. Commute genres such as word and solitaire titles do skew toward short daytime sessions, and evening-heavy genres such as 4X skew late. That is useful for planning creative and live ops. It is not a reason to daypart the performance campaign, because the model has the same information with better resolution.
A three-question test before you touch the schedule
Ask these in order. If the answer to the first is yes, stop.
- Is the reason for this schedule something the network's model can already see from conversion data? Hour-of-day CPI differences, weekend lift, payday effects. If yes, leave the schedule open and let the bid do it.
- Is the reason something that happens after the install and that the model cannot see? Event windows, server limits, support hours, first-session content that varies by time. If yes, daypart, and measure the effect on retention or first-session completion rather than CPI.
- Is this a channel with no algorithmic bidder? If yes, the old playbook applies and scheduling is one of the few levers you have.
How to run the one daypart that survives
For a live event window, an illustrative setup: two campaigns on the same network, identical in every respect except that one has a budget uplift of around 40 percent for the twelve hours after the event opens and a matching reduction in the twelve hours before. Judge them on day-3 retention and event participation rate, not on CPI. If the uplifted campaign's cohort shows materially higher participation and retention, the schedule paid for the higher CPI it almost certainly incurred. If participation is flat, the model was already routing those users correctly and the schedule bought nothing. The numbers are illustrative; the comparison is what matters.
The buyers who keep dayparting for CPI are using a 2018 tool on a 2026 auction. The buyers who stopped entirely have handed the network a lever it cannot use well. The narrow middle is a schedule that encodes what only you know, checked against what happens after the install rather than before it.
