A CPM Is a Number, Not an Explanation
Digital advertising auctions are the subject here—not CPM as an isolated metric. The question is why a number that looks like a price does not, by itself, tell you what an impression cost or why it went to one advertiser. Understanding that distinction gives a buyer, seller, or measurement analyst a way to read cost reports as outcomes of an allocation-and-pricing system rather than as explanations of that system.
The price is the end of a sequence
An ad auction begins with an opportunity to show an ad. A search, a visit to a website, an app session, or a streaming-TV viewing event may create that opportunity. Sellers put it into an auction, advertisers compete through buying software, and the auctioneer decides two related things: who receives the placement and what the winner pays. The Media Rating Council (MRC) calls these the two core functions of an auction—allocation and pricing. In many digital environments, the sequence from consumer action to served ad takes less than a tenth of a second.
That speed can make the result look simpler than the process. A report may show one CPM, perhaps alongside impressions and spend. But the number is the visible end of several decisions. Was the opportunity eligible for a particular buyer? Which bids were allowed to compete? How were the bids ranked? Did relevance, quality, context, or a business rule affect the result? Was there a reserve price? Which payment rule converted the winning position into a charge? Did pacing keep the campaign out of some auctions? How was the delivered impression counted?
CPM is useful because it gives buyers and sellers a common way to express value around impressions. Yet it can appear at different points in the process. The MRC's glossary distinguishes a nominal bid, the amount an advertiser enters, from an effective bid, the value the auctioneer attributes to that bid after applying relevant factors. A final clearing price is another step: it is the amount determined by the auction's payment rule. An average reported CPM may be a later summary of spend and measured delivery. Those quantities can be related without being interchangeable.
The distinction is not pedantic. Suppose two hypothetical campaigns both display an $8 CPM. For one, $8 is the advertiser's nominal bid. The auctioneer may convert it into a different effective value, rank it against other qualified bids, and charge a price determined by a second-price or modified second-price rule. For the other, $8 is an average cost after a first-price payment, a fee, a discount, and the campaign's actual delivery have been accounted for. The matching number does not show that the campaigns faced the same competitors, used the same rules, or bought impressions of the same value. It only shows that one reported quantity is equal.
That is why the first question about an auction price should not be “Is this CPM high or low?” It should be “Which stage does this CPM describe, and what decisions connect it to the other stages?”
A useful way to keep the stages separate is to ask who owns each statement. The advertiser owns the nominal bid as an expression of what they are willing to offer under particular campaign settings. The auctioneer owns the effective value used inside its decision process. The payment rule determines the clearing price. A report or invoice then records a financial and measurement outcome over some defined set of impressions. Each statement can be accurate in its own context and still be misleading when substituted for another. Saying “we paid $8 CPM” is incomplete until the reader knows whether that means a bid, a platform-reported media cost, a gross or net amount, or an average calculated over delivered impressions.
The denominator matters because allocation changes the population being measured. Imagine a seller receives one thousand impression opportunities but sells only the opportunities that pass a reserve and attract qualified demand. An average CPM calculated over the sold impressions describes a different population from a figure that includes every opportunity presented to the auction. If pacing keeps a campaign out of some auctions, its delivered average describes the opportunities it entered and won, not necessarily the market it could have entered. Neither number is automatically wrong. They answer different questions. The trouble begins when they are compared as if they described the same opportunity set, time window, and supply conditions.
The distinction also changes what counts as a useful improvement. An advertiser may welcome a lower average CPM, but a lower price could accompany fewer relevant impressions or a lower allocation rate. A seller may welcome a higher clearing price, but the increase could come with fewer impressions sold or a changed mix of inventory. A consumer may receive a more relevant ad even when a price comparison alone looks worse, or receive a less relevant ad at a lower price. The MRC's account of auction value puts these participants together for a reason: allocation and pricing create different consequences, and a single metric cannot represent all of them.
At this stage, the point is not to declare one auction format superior. The MRC standards deliberately do not prescribe a single auction design or replace protocols such as OpenRTB. Their premise is more basic: whatever design an auctioneer chooses, participants need enough information to understand how pricing and allocation work and to validate reported outcomes. That leaves room for first-price, second-price, modified, and more complex systems while still demanding a legible relationship between the input, the decision, and the result.
What the number leaves out
An auction's allocation decision gives the price its economic setting. An advertiser does not pay for an abstract unit called “CPM”; they pay, directly or indirectly, for access to particular impression opportunities. A consumer, device, page, program, audience, or context can change how much an impression is worth to different advertisers. The seller is not only asking which bid is largest. The seller is deciding which demand should be matched with which opportunity, under a set of rules.
The MRC describes the benefits of a fair and transparent auction in terms of all three participants. Consumers may receive a more relevant ad. Advertisers may reach a consumer who is more likely to be interested at a price discovered in real time. Sellers may clear inventory at a fair price. Each benefit depends on both functions. A low reported average may accompany a weak match or a lower allocation rate; an unsold opportunity has no clearing price to average. A high price may reflect strong competition, a valuable audience, a floor, or an adjustment in the ranking system. Without the allocation context, the CPM cannot tell us which interpretation is appropriate.
The relationship between rules and bidding strategy makes the problem concrete. A strategy suited to a first-price auction is not necessarily suited to a second-price auction. If an advertiser believes they are bidding into one kind of auction while the auctioneer is actually using another, the bid may be rational under the wrong model and the resulting performance may be misread. A change in CPM could then be a response to a strategy that was built on a false description of the mechanism, not evidence that the underlying audience suddenly became more or less valuable.
The same issue appears on the seller side. A seller may use a reserve price—a minimum below which an impression will not be sold—and that reserve may be fixed or dynamically adjusted. A bid can exist without producing a sale. Alternatively, a reserve can lift the clearing price above what a simple comparison of bids would suggest. A fee or discount can change the relation between the amount used in the auction and the amount that appears in a billing or performance report. Pacing can change which auctions a campaign enters or the bid levels it uses. These are not interchangeable explanations, and a single CPM does not identify them.
The number also has a measurement boundary. Spend, impressions, clicks, viewability, invalid traffic, audience, and brand-safety measures may come from an auctioneer or a third-party provider. The MRC's standards call for reporting that lets participants understand what they paid and how that price was determined, ideally at line-item granularity and in a machine-readable form. Where full reporting is too costly, sampling and aggregated insights may be used, but the sampling approach and its limitations matter. A precise average can still support a weak conclusion if the denominator, sample, time window, or matching identifiers are unclear.
The question behind the number
The title's answer is therefore not that CPM is useless. CPM is a useful unit and a useful summary once its inputs and context have been reconciled. The problem is treating the summary as a causal account. A cost number tells you what was reported; it does not automatically tell you what competed, what won, how the payment was calculated, or whether the measured delivery corresponds to the opportunities the buyer thought they purchased.
This series follows that distinction backward through the auction. It examines how nominal bids become effective values and final charges; how qualification, ranking, supply paths, and reporting shape the contest; and how a participant can investigate a cost change without confusing a measurement problem with a strategy problem. It then asks a deeper design question raised by recent auction research: when targeting information improves relevance but also narrows competition, can a mechanism use that information without forcing advertisers to compress their preferences into one scalar bid?
That question matters because the auction itself can change the meaning of every reported CPM. The MRC's transparency framework does not require every platform to use one auction type or publish proprietary code. It asks for enough disclosure, reporting, change history, error information, and auditability for participants to understand the rules and outcomes. The research paper, in a different evidentiary category, proposes a mechanism called the Information-Bundling Position Auction and evaluates it in a model and counterfactual retail-media simulation. Those are not the same claim. One is a framework for making auction behavior legible; the other is an argument about what a different design might achieve.
Keeping those claims separate is part of understanding cost. A standard can tell us what evidence an auctioneer should make available. A theorem can tell us what follows under stated assumptions. A simulation can show what happened when estimated data and a proposed mechanism were combined. None of them turns an unexplained CPM into a complete explanation by itself. The durable skill is to follow the number back to the decisions that produced it—and to say clearly when the available evidence stops.