in Engineering

An Inside-Out Perspective on the CPM Pricing Model for RTB

By William Browne, Senior Modeling Scientist

Part of Quantcast’s mission in delivering well-targeted advertising is to reconcile the different pricing models our advertisers use. A pricing model is a construct that:

  • Delivers maximum value for our customer
  • Accommodates the fact that customers have more than one definition of value

This post gives an in-depth view of one of the most common pricing models used by Quantcast’s advertisers and some of its seemingly paradoxical qualities.

What is the CPM Pricing Model?

When an advertiser wants a CPM deal from Quantcast, they will structure it in terms of:

  • Number of advertisements
  • Budget per advertisement
  • Performance goal

For example, a client might ask us to show 50M advertisements at $3 CPM and a $10 Cost Per Acquisition (CPA). This means that for every thousand impressions they’ll pay us $3. Additionally they’ll typically stipulate a performance goal stated in terms of CPA. Plainly put, for their $150K they’ll want to see a return in actual product purchases driven by our advertising of at least 15K conversions over the term of the deal.

Why Do Advertisers Want CPM Pricing in the First Place?

The idea of buying a certain number of impressions seemed strange to me the first time I heard about it. Why not simply pay for conversions, as in a CPA pricing structure? This is what you’d think an advertiser would actually want. However, it turns out that CPA is not the only goal an advertiser might want. There’s the idea of utility, which I have borrowed from economics. Utility is a concept that allows you to express total value as a function of multiple inputs.  

The viewpoint that our clients only value CPA is a little narrow. They may value CPA primarily but also have an impressions or budget constraint to meet. They may even care about constraints without an explicit focus on number of conversions more than the conversion constraint itself. Effectively, they mix goals: give me conversions and show a certain number of impressions. This extends to a variety of different measurables (e.g. viewability, click-through rate, time spent on homepage, etc.). I’m going to highlight a few scenarios a client may wish for, how the idea of utility applies to them, and how the CPM deal helps maximize client utility.

Quantcast is very good at delivering the correct number of advertisements. So, our clients that buy CPM deals are nearly guaranteed that their budget will be spent within the time frame they desire to spend it. This is very helpful when their media budgets are rigid. Simply put, we have a control mechanism in place that leverages our targeting algorithms and enforces the budget constraint given by our client.

A CPA pricing model complicates the predictability of delivering a budget. There’s a heavy seasonal component to conversions so the number that any client receives can vary wildly month over month. Despite this, in aggregate it’s possible to get a decent estimate of how many online conversions a particular client might receive. However, when you add in attribution among multiple advertising partners, price and inventory fluctuations in the RTB landscape, and a number of other variables, estimating the number of attributable conversions becomes very difficult. This creates a scenario where the budget (in conversions) we’ll use is sufficiently uncertain over the time spans we typically run campaigns that the client prefers to deal with something more certain. In other words, telling a client a priori how much budget we’ll use is much more transparent and easy to understand in the constraints of the CPM model.

Another advantage of CPM pricing concerns the question of conversion attribution.  The industry standard is that the last advertising partner who tags the converter with an advertisement gets attributed the conversion.  There’s pervasive awareness throughout the industry that this system of “last-touch” describes only a fraction of the factors that lead to a sale.

Shouldn’t relevant advertising drive visits to the home-page? Shouldn’t it help consumers differentiate products from high SKU retailers? Shouldn’t it drive user engagement? Quantcast and many of our clients believe that relevant advertising does this and more. The problem is that it’s difficult, though not impossible, to validate these metrics through third party auditing agencies because so many of them are based on a last-touch attribution model. The CPM cost model allows Quantcast to more liberally go after promising consumers, without being constrained by a pure focus on attributed conversions.

If the client’s utility includes spending a particular budget or more inclusive definitions of performance, then the CPM pricing model serves them well. The CPM pricing model is a way to access some less tangible aspects of online advertising campaigns.  

What is Surprising About this Pricing Model?

For reasons mentioned in the previous section, the CPM model is good for campaigns that want to drive conversion performance as well as behavior correlated with conversion performance. However, it leads to an interesting and challenging problem because the advertiser pays a fixed price whereas Quantcast pays a dynamic price for each impression. I will explain how this works from the perspective of the Demand Side Platform (DSP), a system that allows the client to handle multiple exchange and platform interfaces through one service, and the client.

Imagine you’re running a campaign over several media channels. To get the most conversions for the advertiser, Quantcast will always buy the cheapest possible next conversion. In aggregate, this is the way to provide our customers the most value in terms of conversions. The law of large numbers applies in the aggregate regime. So along high volume coordinates, those where we buy large volumes of inventory, we expect our internal CPAs to be roughly equal to one another. One example of a high volume coordinate is an entire advertising exchange. Supposing that we’re buying bid inventory from exchange A and exchange B. We would expect our CPA on A and B to be roughly equal. When you fragment your viewpoint into smaller units like individual sites, the CPA can deviate from expectation.

The inventory on RTB exchanges is highly fragmented and of variable quality. Because we purchase inventory bid by bid, we pay a range of prices depending on a multitude of features corresponding with the bidding opportunity. What we’ve gotten really good at over the years is identifying “deals.”  Our targeting algorithms pick out and price bids based on how likely it is that the bid will lead to a conversion event. We do this by modeling all relevant consumers on the Internet and their likelihood to buy our advertiser’s products.

Due to the fragmentation of the media space, sometimes we buy multiple impressions for a low cost leading to a conversion, and sometimes we buy relatively few higher-cost impressions leading to a conversion. We’ve spent the same in both cases and managed an equivalent and near optimal CPA. As a quirk of the CPM pricing structure, our clients spend more procuring one conversion than the other despite an optimal allocation of their advertising dollars.

I’ll illustrate why buying on both site A and site B is better than buying on site A alone.  First let’s think of what happens when we buy more inventory from a site. Assuming that both impressions and conversions experience diminishing returns, the more we buy of either the less we get per dollar spent. This is an interaction between conversion rate as we buy more impressions, which decreases, and the cost per impression which increases. This is doubly bad for the aggregate conversion rate. For the same budget, we get a lower conversion rate and buy less impressions when we restrict our inventory. Together, these two factors lead to a lower total number of conversions per dollar spent.

Imagine an example where cost increases as we buy more inventory from a site and we make the optimistic assumption that conversion rate stays constant. What happens when we buy (optimally) from both sites? Quantcast buys 1,000 impressions on site A at $1 CPM to generate 10 conversions and 500 impressions on site B at $2 CPM to generate another 10 conversions for a total of 20 conversions. The client will have paid 2x the amount for a site A conversion as they did for the site B conversion simply as an artifact of this pricing model! The above may be a little misleading from the outside. Suppose to improve overall external CPA we instead spend all of our money on site B with the higher conversion rate. Since publishers have limited inventory on their site, on average, impressions would become more expensive. So with all of the money we spent on the two sites, let’s say we’d procure only 900 impressions here. This would get 19 conversions. This is less than if we had distributed our spend across site A and B. Hence by trying to improve the external CPA we lose total conversions.


The CPM pricing model offers a number of practical advantages in delivering an ad campaign. Budgets are well defined and relatively easy to control. The downside is that the pricing can seem counterintuitive to the client when viewed from the perspective of the advertiser CPA. However, Quantcast’s targeting is configured to deliver the optimal number of conversions and, as such, advertiser CPA discrepancies are symptomatic of a strength of the system. What may seem like a CPA imbalance as viewed by the client is actually a symptom of our systems optimizing overall conversions.