Google Ad Manager Rebuild: Inventory, Pricing, Reporting

The first number we produce on an Ad Manager engagement is the gap between what your network reports and what your demand partners remit. On the publisher networks we have inherited it has run between 6 and 19 percent, and almost none of it was theft. It was a network in which no two reports could be joined: ad units named after the sales team that sold them, key-values added by three developers over four years with nobody owning the dictionary, and a reporting habit built on the Ad Manager interface, which samples long date ranges and quietly excludes late-arriving data. We rebuild in the order the dependencies actually run, which is not the order the interface presents them. Inventory: The ad unit tree is the schema of every report you will run for the next five years, so it maps to page templates and slot positions, never to accounts or to the current org chart. Three levels is usually the right depth. Anything the sales team cares about, sponsor, section, series, becomes a key-value instead, because a key-value can be renamed on a Tuesday and an ad unit cannot be renamed at all. We write the key dictionary as a document your engineers and your ad ops team both sign, with the allowed value cardinality per key, because Ad Manager truncates key names at 20 characters and Prebid's hb_ keys collide with hand-made ones more often than anyone expects. Pricing: Unified Pricing Rules replace whatever floor logic is scattered across your line items and your Ad Exchange settings. Floors come from your own bid-level data, by inventory and by geography, because an Indonesian session should not carry a Singapore floor and an exchange's recommended floor is biased downward by construction. Protections get the opposite treatment: inherited networks almost always block more than they should, category blocks stacked over years by people who have left. Where the objection is quality rather than legal or contractual, we reprice instead of blocking. Proof: Data Transfer goes on in week one, delivering NetworkImpressions, NetworkBackfillImpressions, NetworkRequests and NetworkClicks to a Cloud Storage bucket in your own project, loaded to BigQuery on a schedule you control. Every report we hand you afterwards is a query you can read, in a dataset you own, joined against the remittance files your partners send. If you stop working with us the reporting stays and keeps running. This is also the only way we know to settle a discrepancy argument with an exchange in one email rather than six weeks. On the page, GPT runs single-request with fetch and render margins set from your own scroll-depth distribution rather than the defaults, PPID is wired if you have a logged-in audience worth addressing, and refresh, if you run it, is gated on Active View rather than on a timer. Not included: Creative production. Direct sales. CMP procurement, though we will tell you plainly if yours is misfiring and we will not build on top of a broken consent signal. Video player selection. We will set up Multiple Customer Management for your child publishers, and we will not become your MCM parent, because a partner who takes a cut of your child inventory should not also be advising you on it. Who should not buy this: Networks under roughly 20 million monthly ad impressions, where the fee is a larger line than the money we can find. Publishers buying their traffic. Publishers unwilling to fund a Google Cloud project, which for a mid-sized network runs a few hundred dollars a month. And anyone who wants us paid on a share of the uplift. We quote a flat fee agreed before the work starts, because a share of uplift pays us to add demand partners and adding demand partners is usually the wrong move.