Yield Audit: Where Your Session Revenue Actually Goes

The first number I ask for is not page RPM. It is revenue per thousand sessions, because RPM rewards showing more ads to fewer people and that is the direction most stacks have already drifted without anyone choosing it. The second thing I pull is the gap between what your SSPs report and what your ad server counted. Above roughly ten percent that is not rounding, it is a leak with a cause, and in a header bidding stack it is usually where the recoverable money sits. What I pull and read: Ad server reporting at line item and creative level, per-SSP bid rate, win rate and bid density, timeout distribution from your wrapper analytics, unfilled inventory by ad unit and device and geography, unified pricing rules set against your line item priorities, and your ads.txt and sellers.json entries traced to the parties actually reselling you. I want the wrapper configuration file and the last three months of payout statements, not a screenshot of a dashboard. The findings that repeat: Floors set once, years ago, applied blind to geography and device. The same SSP reached twice, once through the wrapper and once through server-side bidding, where you pay in latency for a bid you already had. Ads.txt authorising resellers nobody at the company remembers approving, so a buyer who sees your inventory on six paths buys the cheapest, which is not the one that pays you most. Wrapper timeouts copied from a US benchmark onto a stack whose bidders sit behind a different network path. House line items outranking programmatic demand because of a priority set during a launch and never reversed. What this is not: It is not a vendor selection exercise, and it is not a rebuild proposal wearing an audit's clothes. I take no revenue share, no referral fee and no commission from any SSP, exchange or wrapper vendor. If I recommend removing a partner, it is because the data shows the partner costs more in latency and duplication than it returns, and the arithmetic is in the appendix for you to check. What I refuse to do: Recommend adding demand partners as a first move. Nine times in ten the stack already carries more partners than it can auction inside a workable timeout, and the next one adds latency to every page for a marginal bid. I also will not produce a projected uplift figure. What a change earns depends on your demand mix and your readers, and a consultant's forecast has a way of becoming a target somebody else is measured against later. Who should not commission this: Publishers below roughly three million monthly page views, where the findings will be real but the money at stake will not repay the fee, and where an hour of conversation is the fairer trade. Sites whose traffic is purchased. And anyone who needs the audit to reach a particular conclusion for an internal argument, which I will decline to write and will decline politely. What arrives: A written diagnosis ordered by money at stake, each item carrying the evidence that produced it, the change required, who has to make it, and what it risks. Where the finding is a configuration value, you get the value. Where it is a commercial question, a reseller to drop or a floor regime to change, you get the argument to make and the number to make it with. Then a working session with your ad operations lead and, if you want it, the engineer who has to ship it, because findings delivered to a manager and not to the person holding merge rights tend to survive as a PDF and nothing else.