When a Price Changes, Which Part of the System Changed?

A changed CPM is an observation, not a diagnosis. The practical problem in digital advertising auctions is to determine whether the change came from measurement and delivery, from the set of bids allowed to compete, from ranking and payment rules, or from a buyer's response to the market. A disciplined diagnosis gives each explanation a record to test and leaves the conclusion provisional when the necessary record is missing.

Start with the measurement before the mechanism

The first mistake in investigating a cost change is to treat two numbers as comparable before checking what each one measures. A CPM may refer to a nominal bid, an effective value, a clearing price, an invoice amount, or an average over delivered impressions. Even when two reports use the same label, they may differ in whether they include fees or discounts, which impressions form the denominator, what time window they cover, and whether the value is gross or net.

The first check is therefore not “what changed in the auction?” It is “did the measured quantity and its population stay the same?” Compare the unit, denominator, delivery window, supply scope, and billing treatment. If one report averages spend over impressions that were measured as delivered and another summarizes auction opportunities or won impressions, a change between them may be a measurement or population change rather than a change in auction economics. If a campaign was paced differently, it may have entered a different mix of opportunities even when the nominal CPM remained constant.

This check matters especially when allocation changes. An auction can receive many opportunities and sell only those that meet eligibility and reserve conditions. A campaign can then report a higher or lower average over its selected wins while the broader opportunity set has not changed in the same way. The average is not necessarily wrong; it is answering a narrower question. The investigation becomes unsound when the narrower average is used as evidence about every opportunity presented to the auction.

The next check is timing. The MRC's transparency standards call for auction results to be available on a next-day or 24–48-hour basis, while allowing another reporting window when supported. A same-day dashboard and a later reconciled report may therefore describe different states of the data. A small apparent movement can disappear when late results, corrected delivery records, or billing adjustments arrive. Before explaining a price change, record when the data were extracted and whether the reporting process was complete.

Sampling creates a similar problem. The standards allow sampling to make reporting feasible, but require it to be generally disclosed and designed to represent typical activity, taking account of sample size, time of day, day of week, seasonality, and other conditions that affect auction results. A sample that overrepresents a high-competition period can make the average look like a market-wide shift. The question is not simply whether a report is sampled. It is whether the sampling design supports the comparison being made.