Buying a number with lost revenue

When a finance director asks what last month’s advertising produced, the dashboard in the meeting is rarely a sufficient answer: it records clicks, attributed orders, and a chain of events that look causal because they sit in sequence. The specialist who has to answer that question honestly is often the person who proposes to switch the ads off in whole cities and then wait for the sales line to move, or fail to move, in the dark. That proposal sounds like sabotage. Anyone whose job is to protect this month’s revenue hears it that way. It is also, in the narrow set of markets where a holdout can be drawn, the way the work of proving usefulness actually begins. The job is not to make a louder campaign, but to impose a period of silence on a market that was being pressed, and to treat whatever happens next as the figure worth taking upstairs.

A whole region, and nothing to track

Incremental testing of this kind does not follow a shopper from an impression to a purchase. It turns one region dark, leaves a comparable region running, and reads the gap between them. The reading is only as good as the comparison: the two regions have to behave alike before the test, nothing else can hit one of them during it, and the gap carries an uncertainty that has to be stated with it. The tracking system is not the instrument. The specialist who works with geographic holdouts is choosing a cruder unit on purpose: a city, a cluster of cities, a region whose boundary can be defended in a meeting and whose sales can be compared with a sibling market left running. People will cross that boundary, and that leak belongs to the design rather than arriving as a surprise. What the method refuses is the comfort of a report that shows a conversion and invites everyone to stop asking whether the advertising produced it. There is no need to watch an individual, because the region is either receiving the campaign or it is not. That switch is what the test controls; everything else about the two regions has to be argued, not assumed.

The weeks the market stays dark

The wait is the part of the work that looks, from outside the analytics team, like negligence. A 2011 paper by Vaver and Koehler still sets the duration at 4 to 8 weeks of artificial darkness on the market, and that window is what the specialist is asking a company to sit through. Those figures come from 2011, which makes them more than two years old, and the duration has not been replaced in the routine that still uses it. Those weeks are not empty: the specialist is watching a sales line that was supposed to be protected and is instead being used as a control, while calls from the darkened regions arrive as complaints. The finance director sees a dip. The next question is whether the test can be cut short because the answer already looks expensive. Holding the blackout for the full window is the craft, because ending it when the first bad week arrives returns the team to the dashboard they already had, with a hole in the revenue and no number that can survive a second meeting. Choosing the region is the first argument inside the company, because a city that is too small will not produce a readable gap and a city that is too important will not be released by the people who depend on its sales. The specialist is bargaining for a darkness that is large enough to measure and small enough to be tolerated, and that bargain is the job as much as the arithmetic that follows.

A scheduled drop in a darkened region can be measured against the region left running. An unplanned drop has nothing to be measured against.

What 2025 and 2026 put on paper

The craft now has dates on the door. IAB issued documents on September 9, 2025, and on November 3, 2025, and those texts treat incremental measurement as something that can be named rather than left as a private argument between a specialist and a finance director. An announcement dated May 2026 and a piece published on September 6, 2026, at ppc.land together give the date November 11, 2025. They also report an entry threshold of 5,000 for the programme, a figure whose unit and period the published note does not spell out, which is why a specialist quoting it has to say what it counts before anyone treats it as a price. A longer history of that measurement problem sits in an earlier essay on this site, which stays with the puzzle of wasted spend, while the working weeks of the person who now has to impose the test, and the period in which the job consists of not advertising, remain the subject here.

The argument upstairs

The finance director is not being asked to enjoy a loss. The specialist’s argument is narrower: a dip that was scheduled, bounded by geography, and left to run is a measurement, whereas a dip that arrives as a surprise is only a bad month. That argument is hard to win because it asks a company to treat forgone revenue as the price of a number. The stated threshold of 5,000 sets the scale of the conversation before the first week of darkness, and the wait of those weeks keeps the cost running after the meeting in which the test was approved. The specialist who can hold that line is not proving that advertising always works. The dark region may show that the campaign was intercepting demand that would have arrived anyway, and that result is still the point of the work. It is also the result that is hardest to carry back upstairs, because it looks like a confession that the spend was unnecessary.

The method can cost the company sales it would otherwise make in the darkened region, and it saves the advertising spend that region would have consumed. Whether the net result is a loss depends on how much of that demand the advertising was creating rather than collecting. It works only where a region can be isolated, a campaign switched off cleanly, and buyers are tied to geography. Outside those conditions the same silence proves nothing, and a dashboard on another channel does not inherit the result. The specialist running the test is paid to keep those boundaries clear before the quiet begins.