Zone and Placement Whitelist Audit, Twelve Weeks of Logs

A zone should not be cut until it has spent three times your target acquisition cost with nothing to show. Below that you are pruning variance, and the whitelist gets worse every week somebody touches it. Most accounts handed to us have been optimised daily against a fortnightly signal, which is how they end up with four thousand blacklisted zone IDs and an acquisition cost that has not moved since March. What we need from you: Log-level exports from your tracker, network reports covering the same window at zone and sub-ID granularity, and read access if you have it to give. Twelve weeks minimum. The two sides get reconciled first, because on three of the five large pop networks the zone-level spend in the report and the clicks in the tracker disagree by more than five percent, and which side is wrong changes what everything after it means. The scoring: Zones are scored on a Beta-Binomial posterior against your own account baseline, not against each other. That produces a probability that a zone beats your average rather than a rank, which is what stops a zone with nine clicks and one conversion sitting at the top of a spreadsheet. The cut threshold, the promotion threshold and the minimum sample are derived from your acquisition cost and margin, and they are written into the deliverable so your team can apply them without calling us. Zone ID recycling: Identifiers get reissued. On several networks an ID that was a Brazilian streaming site in January is a Turkish forum by June, and a whitelist nobody has re-scored is buying something other than what it was built for. Every entry is dated, tested for behavioural discontinuity in its daily series, and given a re-score interval derived from how fast that network's IDs actually turn over in your data. It lands between 30 and 60 days more often than not. Sub-ID laundering: Where a network exposes sub-IDs, the interesting failures live there: a whitelisted zone whose top three sub-IDs carry most of the spend and none of the conversions. Scoring splits at whichever level your network reports honestly. Where the sub-ID field is free text set by the publisher, we say so plainly rather than modelling noise and presenting it as a finding. What you receive: Ingest-ready lists in each network's own format, because they differ: Propeller and Clickadu take a CSV of IDs, Zeropark takes a target list, Adcash needs them pasted per campaign. A rules document of four to six pages. And a re-scoring script in Python that pulls your tracker API and reissues the lists monthly. It is roughly two hundred lines, it is yours, and it does not report anything back to us. What this is not: Not campaign management. We do not touch bids, creative or offers on this engagement, and we do not open network accounts. The audit is deliberately separable so you can hand the result to whoever runs your media, including an agency we have never met. It is also not fraud forensics. We will name zones behaving like bot farms, but if you need evidence for a chargeback dispute against a network, that is a different practitioner and we will give you a name. Who this is wrong for: Accounts spending under about 15,000 dollars a month on the channel. There is not enough data, and the honest recommendation would be to run three months longer before paying anyone to analyse it. Also wrong if what you actually want is someone to confirm a decision already taken. The scoring occasionally says the zones you like best are carrying the account's worst traffic, and we hand that over as written.