The Publisher Ad Ops Audit: What to Examine, in What Order, and What to Do With What You Find
A publisher ad ops audit examines five areas in sequence: inventory and setup hygiene, yield and pricing configuration, the end-to-end workflow from pre-sales to invoicing, reporting and reconciliation, and the detection layer that tells you when any of the above breaks. The order matters, because each area rests on the one before it - auditing your floors is wasted effort if the ad units they apply to are misconfigured, and auditing your workflow is wasted effort if nobody can agree on what the numbers say. Run properly, the exercise takes about four weeks alongside normal work and produces a ranked list of what's costing you money, what it would take to fix, and what you should deliberately leave alone.
Most teams already suspect where their problems are. What they lack is a defensible account of which problems are worth the disruption of fixing, in what order, with what expected effect - the thing you need before you can ask a finance stakeholder for budget or a sales stakeholder for patience. An audit is not an accusation about how the setup got this way. Every one of these environments was built by competent people making reasonable decisions under deadline, and the accumulated result is nobody's fault in particular. The point is to see the whole of it at once, which almost nobody does in the ordinary course of the job.
What an Ad Ops Audit Is, and What It Isn't
An audit is a structured read of your current state against what the setup is supposed to be doing, producing a ranked list of gaps. That's all. It's diagnostic, not corrective - the fixing is a separate exercise with separate resourcing, and conflating the two is the most common reason audits stall halfway through.
Three things it specifically is not:
It isn't a technology evaluation. The question "should we replace our ad server" is a different project. An audit assumes the stack you have and asks whether it's configured and operated the way you intended. Nine times in ten the answer produces enough recovered value that the replacement question gets postponed, which is usually the correct outcome.
It isn't a performance review of the team. If your audit surfaces that one person is the only one who understands how the header bidding wrapper is configured, that is a finding about your documentation and coverage, not about that person. Audits that get framed as personnel assessments produce defensive answers, and defensive answers produce useless audits.
It isn't a one-time event. The useful output is a baseline you can re-run. Most of what an audit finds is drift - settings that were right when they were set and stopped being right as inventory, demand, and traffic changed around them. Drift doesn't stop after you fix it once.
The Five Areas, and Why the Order Matters
The sequence is load-bearing. Each area depends on the accuracy of the one before it, and auditing them out of order produces findings you'll have to throw away.
Inventory and setup hygiene comes first because everything downstream assumes ad units that exist, serve, and are mapped to what you think they're mapped to. Walk the units that carry your revenue, verify they render correctly across templates and on mobile, confirm sizes and mappings, and archive the accumulated clutter. Our checklist of 8 GAM settings that quietly leak revenue is the companion pass for this area.
Yield and pricing configuration second, because floors and pricing rules are only meaningful once you trust the inventory they apply to. Review floor performance by unit and channel against current auction behaviour rather than the assumptions in force when the floors were set. Look at unfilled patterns and where they cluster. Check that your demand sources are all actually delivering, and that none of them are quietly competing with your own direct-sold inventory in ways nobody intended.
Workflow third, and this is the area publishers most often skip, which is why it's the area where the largest recoverable time usually sits. Trace one campaign end to end - the actual path from a signed insertion order through trafficking, delivery, optimisation, reporting, and invoicing. Note every handoff, every place someone re-keys data that already exists somewhere else, and every point where the process depends on a specific person being available. That trace is usually uncomfortable to look at and is the single most valuable artefact the audit produces.
Reporting and reconciliation fourth. Can finance close the month from your numbers without a follow-up conversation? Do sales, ad ops, and finance agree on what "delivered" means? If your trafficking conventions are inconsistent, this is where it surfaces - naming and setup standards break reporting downstream far more often than they break delivery, which is why the damage stays invisible for so long.
Detection last, because you can only sensibly design detection once you know what's worth detecting. The question is not whether you have alerts. It's whether a problem starting on a Tuesday afternoon reaches someone who can act on it before Thursday - and whether the answer changes on a weekend, when a Friday problem has the whole weekend to compound. Most teams discover their detection layer is a person who habitually checks, rather than a system. That works until that person is on holiday.
The Four-Week Sequence for Running It Yourself
You do not need a consultant to run this. You need about six to eight hours a week for four weeks, and enough discipline to write findings down rather than fix them as you go. The discipline is the hard part.
Week one: inventory and yield. The two configuration areas together, because they share a data pull and the same people. Output is a list of every unit and pricing rule that isn't doing what you assumed.
Week two: the workflow trace. Pick one representative direct-sold campaign and one programmatic deal, and follow each end to end. Interview the people at each handoff rather than reading the documented process, because the documented process and the real one diverged some time ago and the real one is what you're auditing.
Week three: reporting and detection. Reconcile one closed month against what finance received. Then map what would actually happen, hour by hour, if a campaign started under-pacing on a Friday afternoon.
Week four: write it up and rank it. Findings only, no solutions yet. Ranked by revenue at risk, then by effort. This week is where audits die if you let it slide - a set of findings that never gets written up is indistinguishable from never having run the audit.
Two rules that make the difference between a useful audit and a wasted month. Write findings down instead of fixing them in the moment. The instinct to fix a broken thing the second you see it is a good instinct that ruins audits, because you lose the pattern across findings and you spend your best attention on whatever you happened to notice first. Include things that are working. An audit that only lists problems reads as an indictment, gets received defensively, and understates how much of your setup you can safely leave alone - which is itself one of the most valuable things the exercise can tell you.
What the Findings Usually Look Like
Findings tend to sort into four types, and the type determines what you do about it more than the severity does.
Things that are simply broken. A unit not serving, a rule not applying, a report scheduled to someone who left. These are satisfying, cheap, and usually the smallest category. Fix them immediately - they don't need to go into the plan.
Things configured for a reality that no longer exists. Floors set for last year's demand, targeting built for a site section that got redesigned, a demand partner still weighted for a relationship that has since changed. This is usually the largest category and the one with the most recoverable value, because drift accumulates quietly and nothing ever fires an alert about it.
Things only one person knows. Not a defect until that person is unavailable, at which point it's the most expensive category on the list. The remedy is documentation and cross-training, not a technical change - and it's why a real handover document is worth more than most tooling.
Things nobody would notice for weeks. Detection gaps. These rarely feel urgent during the audit and are consistently the ones that cost the most between audits, because their cost is a function of how long the next problem goes unseen rather than of anything currently wrong. This is the area where a daily monitoring baseline earns its place - it's the job ProOps Ads Tracker does, checking the network every morning so the gap between a problem starting and someone knowing about it stops depending on who happened to look.
Turning Findings Into a Plan You'll Finish
A ranked list of twenty findings handed to a team of three is not a plan. It's a source of guilt. Sort into three buckets and commit to them differently.
Fix this fortnight. Small, contained, no cross-team dependency. Owner and date against each. If it needs a meeting to schedule, it belongs in the next bucket.
Fix this quarter. Real work needing coordination - floor restructuring, naming convention migration, reporting rebuilds. These need a named owner, a rough estimate, and a sequence, because they'll compete with everything else on the roadmap and lose unless someone is accountable for them.
Accept, and revisit at the next audit. The bucket most teams skip, and the one that makes the other two credible. Some findings aren't worth fixing given your size, stack, or roadmap. Writing down that you've considered and declined something is a legitimate outcome, and it stops the same finding being rediscovered as new next year.
One benchmark worth holding yourself to as you rank: delivering 95% of committed media budgets should be your standard, and consistently achievable. If your audit shows you routinely below it, that's the number the plan should be built around, because it's the one your advertisers experience directly.
If the plan is mostly bucket two, you have a resourcing question rather than an ad ops question, and that's worth naming explicitly to whoever holds the budget. Our piece on what publishers should keep, delegate, and automate is the framework for that conversation.
When to Bring in Outside Help
Run it yourself if you have someone who can hold the whole picture, the calendar space to do it properly, and enough internal trust that the workflow interviews get honest answers. Those three conditions are more often met than people assume.
Bring in outside help when one of these applies. Nobody has the whole picture. Highly specialised teams are common now, and if your header bidding person and your direct-sold person can't each describe the other's area, an internal audit will have blind spots exactly where the handoffs are. The findings will be politically difficult. If the honest answer implicates a decision a current stakeholder made, an internal auditor is being asked to do something unfair, and you'll get a softened result. You need it to carry weight externally. An assessment going to a board, a buyer, or a parent company needs to be defensible by someone with no stake in the conclusion.
We run this audit as a paid engagement, and we also run a shorter version free - a working session across these five areas that produces a written summary of where the value is and what it would take to get it. Most publishers who take the free version go on to run the full thing themselves with the framework in hand, which is a perfectly good outcome and the one we'd suggest if your findings are mostly in bucket one.
If that's useful, the free ad ops audit is here. If you'd rather just start on your own this month, the four-week sequence above is the whole method - there's nothing withheld from it.
FAQ - Publisher Ad Ops Audits
What is a publisher ad ops audit?
A structured read of your current ad operations setup against what it's supposed to be doing, across five areas: inventory and setup hygiene, yield and pricing configuration, the end-to-end workflow from pre-sales to invoicing, reporting and reconciliation, and the detection layer. It's diagnostic rather than corrective - the output is a ranked list of gaps with the effort to close each one, not the fixes themselves.
How long does an ad ops audit take?
About four weeks running alongside normal work, at roughly six to eight hours a week. Week one covers inventory and yield configuration, week two traces one direct-sold campaign and one programmatic deal end to end, week three covers reporting reconciliation and detection, and week four is writing up and ranking the findings. The write-up week is the one teams skip, and skipping it makes the previous three weeks worthless.
Can we run an ad ops audit ourselves or do we need a consultant?
Run it yourself if someone on the team can hold the whole picture end to end, you have the calendar space to do it properly, and there's enough internal trust that the workflow interviews get honest answers. Bring in outside help when no single person has the whole picture, when the findings will implicate a decision a current stakeholder made, or when the assessment needs to carry weight with a board, buyer, or parent company.
What does an ad ops audit typically find?
Findings sort into four types: things that are simply broken (smallest category, cheapest to fix); things configured for a reality that no longer exists, such as floors set for last year's demand (usually the largest category and where most recoverable value sits); things only one person knows (not a defect until that person is unavailable); and things nobody would notice for weeks, which are detection gaps and consistently cost the most between audits.
How often should publishers audit their ad ops setup?
Annually as a baseline, and after any material change - a site redesign, a demand partner change, an ad server migration, or losing someone who held significant undocumented knowledge. Most of what an audit finds is drift rather than error: settings that were correct when set and stopped being correct as inventory, demand, and traffic changed around them. Drift doesn't stop after you correct it once.
What should we do first with the audit findings?
Sort them into three buckets rather than working down a single ranked list. Fix this fortnight: small, contained, no cross-team dependency, with an owner and a date. Fix this quarter: real work needing coordination and a named owner. Accept and revisit: findings not worth fixing given your size, stack, or roadmap. That third bucket is the one teams skip, and it's what makes the other two credible.