What We Get Wrong About eCPM

What's your intuitive feeling when you look at the eCPM metric? To me, it meant something like this:

The value of ad space

The reason I thought this way is because eCPM is measured in currency ($), making it easy to associate with value—just like the price of a product. It also didn't help that it's expressed as a price per unit of impression (Ad Revenue = eCPM * Impressions). If the number of impressions is fixed, eCPM dictates the revenue.

Approaching this from a strictly economic perspective, price serves as the standard for determining the value of goods and services in a market economy. So to me, it felt exactly like the market-evaluated value of the ad space.

But is eCPM really the true market value of ad space?

As mentioned earlier, can we truly consider the eCPM we see as the actual market value of the ad placement? Does eCPM purely reflect the advertiser's willingness to pay for that space alone?

The composition of eCPM

We often think of eCPM as the compensation for ad impressions. The formula for ad revenue, Revenue = impression * eCPM/1000, calculates profit by multiplying indiscriminately aggregated impressions by eCPM at the individual impression level. Just looking at the formula, it seems to be based on the fundamental premise that all impressions are equal.

However, we need to examine whether all impressions are actually identical. What does this mean?

For example, if an ad placement requests ads at regular time intervals, the number of ads users see in a session becomes dependent on their retention time. Users who stay longer will see more ads, while those who stay for a shorter time will see fewer. Therefore, the number of times an ad is shown is determined by the user's session length.

We started with the basic assumption that "all ad impressions are uniform." But is there really no difference between the first ad seen by 유저 1 and the 20th ad seen by 유저 3...? What is the significance of order in impressions where the context of sequence is involved?

eCPM

But what about the eCPM we actually look at? Usually, we check the eCPM that corresponds to an average, which completely fails to reflect the diminishing marginal utility. For instance, A and C take on a very similar shape on the curve above and are actually the same type of ad space, but if you look at their aggregated single-number eCPM, it looks like this:

Although there are slight differences in the graph, despite having similar overall value formations and declining trends, placement A's eCPM is about 30% higher. This phenomenon occurs due to the properties of the screen where the placement belongs. In the example above, the differences between the screens containing placements A and C were as follows:

• Placement A: A placement with short retention time <> Because the retention time is short, all users see the ad exactly once.

• Placement C: A placement with long retention time <> Because the retention time is long, all users see the ad 5 times.

If the formation of eCPM based on the increasing number of accumulated ads shown to users in a session is identical for both placements A and C like this:

The average eCPM can be calculated as follows:

• Placement A: 1000 x 1.5 / (1x1000) = $1.5

• Placement C: 1000 x (1.5+1.3+1.1+1.0+0.9) / (5x1000) = $1.16

Value...?

Coming back to this... in terms of the market-evaluated value of ad space mentioned at the beginning, are placements A and C with different eCPMs actually different in value? It doesn't seem that way. Both placements were evaluated with the exact same eCPM price under identical ad impression conditions. However, due to the nature of the page where the placement belongs, the difference in aggregated eCPM values occurred solely because of the high or low number of ad requests.

When looking bluntly at the eCPM average, values that looked quite different were actually being evaluated identically under the same conditions. In fact, eCPM expressed as a single number seems insufficient to be viewed solely as the market value of the ad space. The variance in average eCPM was caused by factors that impact eCPM, such as the number of ad requests driven by retention time.

It might be more accurate to define it like this:

• eCPM = Value of the ad space + Properties of the page or service containing the space

I believe the value of an ad space should be based on evaluation under identical conditions. From that perspective, looking at eCPM as just a big number makes it hard to see it as true value itself, given the omitted context and unequal conditions.

So, what I want to say is

People often tend to emphasize only eCPM when talking about ad monetization. It also feels like we overreact to the ups and downs of eCPM by using it as the sole health-check barometer for a given ad placement.

But ad monetization is not represented by eCPM alone. A high eCPM doesn't mean you're monetizing well, and a low one doesn't mean you're doing it poorly. eCPM is just one element for monetization, not an end in itself. Ultimately, we must pursue a direction that maximizes revenue, and we need to remember that eCPM is a metric that can—and should—change based on strategic decisions during that process.

Then what approach should we take to improve our service's ad revenue? The ad market takes the form of a marketplace that operates strictly on the principles of supply and demand. Publishers are the suppliers providing impression opportunities, while advertisers are the demanders paying the price to post ads on those placements.

From a publisher's perspective, ad revenue is a dynamic determined by how many ad impression opportunities are provided, evaluated alongside the demand and willingness to pay from the advertiser side. Therefore, chasing a single metric doesn't guarantee improved revenue. For a simple example, to raise eCPM, you just need to reduce the supply of impression opportunities. However, this doesn't guarantee that revenue will grow. Revenue is determined by the magnitude of the delta between the increased eCPM <> decreased Impressions.

In the end, it's a matter of balance. Figuring out how to find the balance that optimizes revenue should be our approach. In the following post, I'll talk about our efforts to strike that balance in ad revenue.