Advertising & agencies

By catching under-delivery while a campaign is still running, not after the period closes. A makegood is almost always a shortfall that was visible in the pacing data days before it became a settlement — the signal was there; the system to act on it was not. Move detection forward, placing checks where exposure is greatest, and a large share of makegoods stop happening because the campaign is corrected while it can still deliver against its guarantee. What remains is smaller and budgeted for. This is the logic of the framework that protected more than USD 20 million in client billings.

Staged QA places verification deliberately at the points in the delivery chain where errors are most likely and still correctable, weighted by financial exposure so the heaviest scrutiny lands on the largest billings at risk. Checking everything equally sounds rigorous but fails in practice — the volume is too high, so checks get skipped under pressure and coverage becomes a fiction. Staged QA is designed to run at full volume without collapsing: more scrutiny on a high-value, guarantee-heavy campaign, less on a small low-risk one. It is the right checking in the few places that move money, not more checking everywhere.

By standardising what is inspected and how it is scored, so coverage is explicit, comparable across teams and markets, and demonstrable on request. "We check everything" does not survive a serious client audit, and increasingly clients will not accept it. A defensible model lets you show a real, scored baseline rather than a reassurance — which shifts the conversation from confidence to evidence. In a market where clients audit their agencies, provable coverage becomes a competitive asset: you can demonstrate quality rather than describe effort, and a prospective client sees proof instead of a promise.

Usually because each team carries its own tacit sense of what good means and what counts as a defect — not because some are careless. Without a shared, explicit standard, you cannot compare one market to another, and you cannot tell whether a low score reflects worse work or stricter marking. The fix is to define quality precisely enough that it means the same thing everywhere it is measured. Only then is the score honest and the real pattern visible: which teams need support, which markets carry the most exposure, and where the systemic failure modes sit beneath the local variation.

Not by asking people to be more careful. At 95% across more than 2,000 campaigns and 450 clients, the residual defects are not random — they cluster into a handful of systemic failure modes. The work was to segment those remaining defects, find the recurring root of each, and design that cause out of the delivery process rather than inspecting harder for it. The last few points of quality are the hardest precisely because they require fixing the system, not the symptom. Once the systemic causes are removed, the score moves and, crucially, holds, instead of drifting back when attention shifts.

Yes — it is designed to make them stronger, not to replace them. The quality layer sits across delivery, feeding every catch back so recurring causes are removed at source, which over time lowers the checking burden on the teams themselves. The aim is a quality function the agency owns, with a clear standard, provable coverage and a cadence that holds the gain — not a permanent dependency on me. Much of the value is in transferring the framework so your own people run it after I have gone, with the systemic defects already designed out of their workflow.

Sometimes it is the deals, and I will say so. If campaigns are sold with guarantees the available inventory cannot realistically deliver, no amount of downstream checking will fix it — the shortfall is engineered at the point of sale, and the honest correction is upstream in how deals are structured and what is promised. Building an elaborate quality machine on top of a sales or inventory problem just adds cost without closing the gap. Staged QA earns its place where delivery runs at real scale, guarantees are frequent, and shortfalls are leaking margin invisibly — not where the gap is created before delivery even begins.

A verification platform measures delivery; it does not decide what happens because of what it measures. Without staged checks placed where money leaks, clear ownership of who acts when pacing slips, and a loop that feeds catches back into delivery, a platform produces accurate data that problems still flow straight past. The tooling is useful, but the makegood leak is a governance and process problem far more often than a tooling one. This work installs the operating system around the data — where to check, who acts, and how recurring causes get designed out — which is what actually protects the margin.