Demand Stack Audit and Rebuild for Multi-SSP Publishers

The useful question is not how many demand partners you have, it is which of them pay for themselves. Across the last eleven stacks we rebuilt, the median site ran 14 header bidding adapters and four of them were negative once timeout loss, page latency and duplicate supply paths were counted. Switching those four off raised revenue per thousand sessions. That is the whole argument of this engagement, and it runs against how most stacks get assembled. How a partner earns its place: Holdout testing, one adapter at a time. The adapter is disabled for a randomised half of sessions for 14 days and we read revenue per thousand sessions on both arms. We do not accept the partner's own dashboard as evidence of its contribution. A bidder that wins auctions it was going to win anyway reports beautifully and adds nothing. Two to three holdouts run per month, so a stack of fourteen takes a quarter to resolve properly. There is no faster honest version of this. Where the losses actually sit: Timeout first. A 1000 ms client-side timeout drops late bids a 1500 ms timeout would have taken, and that extra 500 ms costs INP on mid-range Android, which is most of Brazil. The trade gets measured, not assumed. Match rates second: on Safari and iOS, user sync fidelity collapses and an adapter that pays on Chrome desktop bids into thin air on iPhone. Prebid Server recovers part of that and gives up other things, and we will tell you which of your partners belong on the server and which do not. Supply path: Most stacks sell the same impression to the same DSP through three routes and let the buyer take the cheapest. We map every path in your ads.txt against the sellers.json of each partner, find the reseller lines that exist only to take a margin on demand you already hold directly, and cut them. This is usually the fastest single gain in the engagement, and it costs nothing but willingness to have an awkward conversation with a partner. Reconciliation: Monthly, partner-reported impressions and revenue against your ad server. Five to eight percent discrepancy is ordinary. Above fifteen is a conversation, and we will have it with the partner on your behalf. We also read payment behaviour: net terms actually honoured, minimum payout thresholds, and how a USD remittance to a Brazilian entity gets handled. A partner that pays 90 days late is a financing cost dressed as a yield gain. If you also run an app: In-app mediation is a different auction with a different failure mode. Moving an AdMob or MAX waterfall to in-app bidding is work we will take, but only where you have a release cycle we can ship into and a QA path. We do not do it blind through a remote config and hope the crash rate holds. We refuse: To add a partner because their rate sheet reads well. To run a stack whose auction logs we cannot read. And to take a share of uplift as our fee, because attribution of that uplift is precisely the judgement you are paying for and we will not be the party grading our own paper. The fee is fixed and stated before work starts. We take no money from any demand partner and will sign that. Not included: The migration off AdSense, refresh and lazy load policy, direct sales pricing, adblock recovery. Separate engagements, all on this profile. Not for you if: Your stack was built in the last 90 days. There is nothing to audit yet, so run it. Or if what you want is the partner count maximised, which the partners themselves will do for you at no charge.