What a push ad network sells, and what it only claims
A push ad network is a broker: it holds contracts with publishers whose visitors allowed notifications, runs the auction that decides which advertiser fills each slot, and reports what happened. Nothing it sells is made in-house. The list belongs to the publisher, the delivery path belongs to the browser vendor, and the click belongs to whoever tapped. What the broker adds is selection, targeting, fraud filtering and reporting, and the difference between a good one and a poor one lives entirely in those four functions. Judge them before the deposit.
What a push ad network actually owns, what it rents, and why the distinction sets the price of every click
Strip the marketing and a network owns three things: contracts, a bidding engine and a reporting database. Three. It does not own subscribers, because the subscription lives in the browser and is keyed to the publisher's domain, and it does not own delivery, because Google, Mozilla and Apple relay the message. It owns the decision about which advertiser gets which slot, and the record of what happened next. When you evaluate a push ad network you are evaluating the quality of that decision, and nothing else it says about itself.
The contracts decide inventory. A network with direct publisher agreements knows each site, its subscriber age distribution and its content, and can tell you which sources are weather pages and which are download portals, because someone there has actually opened them. I first saw that distinction drawn plainly on push-ads.io, where the sourcing of a list is treated as part of its price rather than a detail, and it changed how I read every rate card afterwards.
Publisher payout, inferred
The share a network pays its publishers is not published, but it can be inferred. A network paying well keeps the sites with fresh lists, because those sites have options, while a network paying poorly keeps sites with nothing else to sell, which are the ones with stale lists and permissive opt-in tactics. Ask what proportion of top sources have been on the platform more than a year. Supply-side retention is the nearest thing to a public payout figure that exists, and it is worth more than any percentage a sales deck offers.
Direct feeds, resold feeds and how a push ad network hides the difference
Not every network sells its own supply. Many plug into others through real-time bidding, taking a margin on inventory they have never seen, and the message still arrives, the report still shows a delivery, and the buyer pays two margins instead of one, one to the network in front of you and one to the network behind it. A push ad network that resells is not dishonest. Expensive. The problem is that the interface rarely says which is which.
Resold supply behaves in a recognisable way. Source identifiers change without explanation, because they are the upstream network's identifiers passed through, subscriber age is missing or bucketed coarsely, because the reseller does not receive it, and bid floors move by the hour as the upstream auction moves. A direct feed sits still. Comparing this against how push ads are priced on a direct panel makes the layering obvious within a day, and the difference shows in the report before it shows in the bill.
Reseller tells on the rate card
Look at the minimum bid by country. A direct network sets floors from its own publishers' expectations, so floors for a small market can be oddly low or oddly high, whereas a reseller inherits floors from upstream and they line up suspiciously well with the big platforms' public rate cards. Look also at the platform split. A network with genuine direct supply can tell you the Android to desktop ratio of each source, and a reseller can only tell you the ratio of its whole feed.
| Signal | Direct supply | Resold feed |
|---|---|---|
| Source identifiers | Stable, and the network can name the site behind each one | Opaque, and they change when the upstream partner changes |
| Subscriber age | Per source, in days | Absent, or three coarse bands |
| Bid floor movement | Weekly at most | Hourly |
| Platform split | Reported per source, so a single download portal with a desktop-heavy list can be excluded without touching the rest | Reported for the whole feed only, which tells you nothing about any one source |
| Creative vetting | In-house, before launch | Passed upstream, sometimes after launch |
Push ad network targeting minimums
Targeting is where a network earns its margin, and the minimum set is short. Country and platform. Obviously. Subscriber age in days or narrow bands, because the price of a click follows it, source identifier, so a good source can be isolated and a bad one excluded, and device with OS version, since Android 13 subscribers passed a different permission gate from earlier ones. Anything less and a push ad network is asking you to buy blind and call it reach.
Time targeting deserves a separate check. Scheduling in the advertiser's time zone is nearly useless when the list spans three continents, and local-time scheduling by subscriber is the feature that matters, which is rarer than it should be. I keep this list on The Albany Notebook because published claims drift faster than anyone updates them. Writing down what was offered on a given date is the only cure.
Exclusion lists matter more than inclusion lists once a campaign has run a week, because the value of a network is how easily you can cut the sources that produce taps without conversions. The cleanest description of that workflow I have read is on a self-serve panel selling push ads, where blocking a source is treated as part of the daily routine rather than a settings chore, and the block takes effect on the next auction rather than the next day. Daily. A block that waits for a settings page is a block that gets postponed.
Fraud, dead endpoints and how a push ad network filters both before they reach the invoice
Fraud on this channel takes two forms, and only one is the kind people expect, the bot click a publisher buys to inflate its own payout. The larger loss is dead endpoints: subscriptions that expired, phones that were wiped, permissions revoked on Android, and billing for those is not fraud in a legal sense, but the effect on your budget is identical. A push ad network is judged by how quickly it removes both.
Vendor responses give the network the tools it needs, and they arrive without anyone asking for them. When Google's relay returns an error for an expired endpoint, a well-run network drops it within hours, and when a source's click-through spikes while its conversion rate falls, a well-run network suspends it, usually before the advertiser has noticed anything. Neither is hard. What varies is whether the network has a reason to act, and a network billing per delivery has less reason than one billing per click.
What Chrome's enforcement did
From Chrome 86 onward, Google's crawler subscribes to sites, inspects the notifications that arrive, and blocks prompts on origins that send abusive content, warning the owner 30 days before it does. Publishers who let networks push fake system alerts lost their prompt, and with it their new subscribers, while networks that vetted creatives kept their publishers. The same lesson applies to push notification ads today. Creative vetting is a supply-protection function, not a formality, and a network that skips it is spending your list.
Testing a push ad network with a small budget, one country, one offer and a week of patience, before any of the answers above are trusted
A small budget answers most questions if it is spent on structure rather than on winning. Fund the minimum, pick one country, one platform and one offer, and split the budget by subscriber age band and by source. The aim in the first week is not profit. It is to learn whether the reports of this push ad network agree with what your landing page records, and how far apart the two are.
A first-week test plan
Day one, run at the floor bid with a daily cap and note delivered against clicks per source. Day two, block the top three sources by click volume that produced zero landing-page events, and day three, raise the bid only on sources under thirty days old, nowhere else. By day five you know whether fresh inventory exists at a price you can pay, and whether the exclusion tools work at all. Yours. The numbers from this exercise belong to you and to nobody's case study.
| Question | Acceptable answer |
|---|---|
| Is this feed direct or resold, and in what proportion | A number, per country |
| How is subscriber age reported | In days, per source, in the campaign report itself rather than in a sales deck |
| What does delivered mean on your invoice | Vendor acceptance, and they say so plainly |
| How fast are expired endpoints removed | Within a day |
| Block from report row | Yes |
| Who vets creatives and when | A person, before the first delivery, and again when a creative is changed, because the network's own prompt eligibility on every publisher site depends on it |
Compare the network's click count with your own analytics before trusting anything else, because a gap of a few percent is loss on the tap, while a gap of a third is a dead list or a redirect chain. Always. I first saw that comparison laid out step by step on a site selling push notification ads, and it is the single check that separates a usable report from a decorative one, whatever the panel looks like.
What you end up with is not a verdict on the format but a verdict on one broker. Another push ad network with different publishers will produce different numbers from the same offer, which is the whole reason the sourcing questions above come before the deposit rather than after it. Keep the test. Keep the exclusion list, and move the budget to whoever answered the questions in writing.