What decay curves reveal about real spending on push ads, and where every list stops earning around its third week

Last updated August 27, 2026 · Smith Jones, Performance Media Buyer

Subscription based notification inventory is sold by list freshness rather than by placement. The age of the base explains more about performance than any targeting option in the interface. A subscriber collected this week responds several times better than one collected three months ago, and the whole base decays continuously as people revoke permission or reset a browser. Anyone buying push ads is therefore renting a wasting asset, which makes renewal cadence and cost per acquisition over a fortnight the two numbers that decide whether a campaign works.

Where push ads come from

Publishers collect subscribers through a permission prompt, and its wording determines what the list is worth later. A base built by asking readers to allow notifications for article updates behaves nothing like one built by a prompt demanding a click to continue. Networks segment inventory by collection age. That segmentation is the first thing to ask about when buying push ads.

Permission prompts

Collection quality is invisible in a network interface and decisive in the result, since a prompt shown after two minutes on an article produces a subscriber who chose something, while a prompt blocking access to a video produces a subscriber who wanted the video and will treat every later notification as an accident. The second kind unsubscribes within days, so the buyer pays for a name that will not exist by the third send. Same base, same price. Ask which publishers supply the segment and what the prompt says, and let adult traffic tracking settle the rest when the answer is a refusal.

List size gets quoted constantly and audited rarely, and a network claiming forty million subscribers in one country is counting every permission ever granted, including devices that have not connected in a year, so ask for delivery counts instead of base counts. An undeliverable subscriber costs nothing to store and produces nothing either.

Send timing, delivery timing and the gap between them that quietly moves every reported number for push ads

Notifications queue while a device is offline and arrive when it reconnects, so send time and delivery time are separate events. A campaign scheduled for nine in the morning reaches part of the base at once and the rest across the following hours, so response curves for push ads run wider than the send window. Reading a report by send hour rather than by delivery hour attributes conversions to the wrong slot, much as anti-adblock traffic skews a device split, and the error is large enough to reverse a schedule.

Morning and late evening carry the strongest response in most markets, while the middle of the working day produces delivery without attention, and the distance between the best and worst hour is wide enough to justify running one offer at two different times before drawing conclusions.

Time to live, and what happens to a notification nobody was online to receive

Time to live settings decide what happens to a queued notification, where a short value discards anything undelivered after a few hours, which protects relevance on a time limited offer, while a long value maximises delivery at the cost of arriving once the reason for sending has passed. A notification about a two hour discount landing eleven hours late damages the sender more than a missed impression could, because the recipient reads it as carelessness. Most networks default to a value nobody chose deliberately, and most buyers never open the field. The right number follows the offer.

Multi country campaigns need scheduling per country. Running one schedule across several time zones sends the same message into breakfast in one market and into the middle of the night in another, the second group leaves, and a revoked permission cannot be asked for a second time. Permanently.

In page formats replacing classic push ads on locked systems

Platform restrictions created the second format. Web notifications never worked on some mobile systems the way advertisers assumed, and browsers began muting permission prompts for sites that abused them, which cut the collection rate on classic push ads sharply. The result was a supply problem rather than a demand problem, since advertisers still wanted the format and publishers could no longer build the bases to sell it. Prices rose on fresh cohorts and fell on stale ones. Scarcity. In page delivery replaced part of that loss. The unit imitates a notification interface inside a web page during an ordinary visit, closer to popunder ads than to a subscription, and no permission is requested.

Where in page delivery fits a media plan

No subscription is stored either, so the unit is a display impression in familiar clothing, available on every device including those where real notifications cannot arrive. Volume is abundant, intent is lower. In page inventory suits broad offers tested at low cost per click, while a subscription base stays better once an offer needs somebody who already agreed to hear from a sender. Treating the two as interchangeable inside one report produces averages that describe neither of them.

The distinction also matters for compliance. A subscription implies consent that a display impression does not, and privacy regimes treat the two records differently even when creative looks identical. Legal teams notice the difference long after the media team stopped caring.

Creative anatomy of push ads

The creative surface is tiny and rigidly structured, with an icon around ninety pixels square, a title truncated near forty characters on most systems, a body line of roughly a hundred, and on some platforms a larger image below it, which is the entire canvas available to push ads. Nothing about that layout is negotiable, and every system truncates at a slightly different point, a constraint native ads avoid, so a title fitting on one handset disappears on another.

The icon does the work of a thumbnail. Faces and single high contrast objects survive at that size, while a logo carrying readable text becomes a grey square on almost every handset. Test it at actual size on real hardware rather than inside a design tool. Unreadable.

Icon and title choices that survive truncation on the handset that matters most in a market

Title writing follows one constraint above the rest, since the first thirty characters must contain the reason to tap, because the remainder gets truncated on most devices and never read, and a title opening with a brand name spends its only visible space on information the recipient did not ask for. Body copy exists to confirm the promise rather than to extend it. Imitating system messages gets accounts closed, since creative resembling a delivery notice or a security warning is prohibited by every serious network, and a publisher losing subscribers over one advertiser drops that advertiser the same week.

Localisation is where cheap campaigns lose their margin, because a translated title running four characters longer than the original disappears behind the truncation point on the most common handset in that market, and nobody notices since the interface preview shows the full string. Check it on hardware.

Decay curves that force a rolling renewal schedule rather than an annual panic about push ads

Response falls week over week on any fixed base, and the curve is steep enough to make a single measurement misleading. A campaign returning target cost per acquisition in week one commonly runs thirty to fifty percent worse by week three against the same audience, which is the structural reason budgets for push ads have to move between bases rather than sit still. Two mechanisms drive the decline, since subscribers revoke permission or clear browser data continuously while those who remain grow accustomed to commercial notifications. Habituation.

The two mechanisms need different answers. Churn is fixed by buying fresher cohorts and habituation by changing the creative, and confusing the two wastes a month of budget on entirely the wrong repair, which is the commonest mistake made here. Delivery counts tell them apart.

Renewal cadence

Renewal works on a rolling schedule rather than as an annual rescue operation. Introducing one new base every fortnight while retiring the weakest keeps blended cost stable, and it avoids the pattern where a buyer notices the decline only after a month and then replaces the whole roster at once. Gradual replacement preserves the comparison data that makes every later decision cheap, exactly as a plan to buy website traffic needs a baseline before it starts.

Benchmarks for cohort pricing are easier to borrow than to build, and I looked at how freshness segments are described and priced on push-ads.io while setting an internal floor, where the structure matched what the last three campaigns had shown about where response collapses. Borrowed figures are always a starting point rather than a conclusion, and they stop being useful the moment a campaign produces numbers of its own from those very cohorts. Replace them with measured ones after the second week, because a figure copied from somewhere else describes somebody else's inventory rather than the cohorts actually bought here.

Measurement has to extend past the first day, whichever Buy Adult Web Traffic source supplied the base, for any of this to work. Counting conversions over a fortnight rather than twenty four hours reveals which cohorts produced buyers instead of clicks, and buyers judging push ads on immediate return keep cutting the sources quietly paying for everything else. The window costs two weeks.