In-depth analysis - Internet advertising bidding model

The development of bidding models in Internet advertising has undergone many changes. From the initial CPM to the CPC bidding model, with Facebook's oCPX (such as oCPM/oCPC) becoming popular online, a new generation of bidding models such as double bidding, activation and payment, Facebook's AEO and VO have emerged one after another.

This series of articles will introduce and use a more general "four points, three rates, two controls and one reinforcement" analysis framework to analyze various bidding models. I believe that after reading this series of articles, you will have a more comprehensive understanding of the most popular bidding models.

This series of articles is divided into three parts. The first part will briefly introduce the specific business logic of basic bidding models such as CPM/CPC/CPA , and then introduce the "four points and three rates" in the "four points, three rates, two controls and one strengthening" and introduce oCPX. The middle part will introduce the " two controls and one enhancement" in the "four points, three rates, two controls and one enhancement" and analyze smart bidding models such as oCPX, activation and payment, AEO/VO, etc. The next article will mainly analyze the differences between the interests of all parties in affiliate advertising and RTB advertising and those in large media delivery platforms.

This series of articles will also explore the following issues:

(1) From CPM to CPC, how has the bargaining relationship between media platforms and advertisers changed?

(2) Under what circumstances can the CPA bidding model be used?

(3) Why do advertisers sometimes want to reduce click-through rates?

(4) In the oCPX bidding model, should advertisers send back real behavioral data?

(5) Why do many domestic oCPX bidding models have over-cost compensation mechanisms?

(6) Why do domestic media platforms require that the number of accumulated behaviors (such as activations) of advertisements reach a certain number before they start paying for the over-cost compensation of the oCPX bidding model?

(7) Why should a certain piece of owned traffic be converted from oCPC mode to oCPM?

(8) What kind of bidding method is double bidding?

(9) Why do we need smart bidding products?

(10) Why does a certain owned traffic only have oCPM model while another Shanjia still has the oCPC model as an option?

(11) How many profit models does DSP have? What changes have occurred in the interests of the three parties under each model?

· The article is coming ·

Large media advertising platforms, such as Facebook and Google, usually establish their own advertising platforms to monetize their own traffic. At this time, the media is also a delivery platform, referred to as the " media platform " in this article.

1. Basic knowledge

An Internet advertisement usually goes through the following processes. Taking the information flow as an example, when a user opens an APP, he or she may sometimes browse an advertisement (ad display, abbreviated as m). The user may click "View details" below the advertisement (ad click, abbreviated as c). On the details page, the user may continue to click "Download Now" (ad download, abbreviated as i), then download the game through the app store or Android app store and open it (activate, abbreviated as a). After playing for a while, the user may spend money to buy props (pay, abbreviated as p), and there may be more in-depth user behaviors in the future.

CPM advertising (Cost Per Mille) charges according to the number of times the ad is displayed. Because the cost of each display is very small, everyone agrees to multiply it by 1000, which is the cost of one thousand ad displays.

If the CPM bid of an ad from advertiser A is RMB 10 per thousand times, and the CPM bid of an ad from advertiser B is RMB 12 per thousand times, then the media platform will sort the ads based on the CPM bids and display the ad that ranks first (i.e., the ad from advertiser B) (for the sake of clarity, assume that only one ad is shown in this ad slot, and the following discussion is also based on this setting).

Note: The “m” in CPM actually means thousand times. However, the “c” (click) and “a” (behavior) in the later developed CPC, CPA and other models are all types of behavior, so people later got used to using “m” to represent impressions.

CPC advertising (Cost Per Click) charges based on the number of times the ad is clicked, because the cost of a single click is usually large enough and does not need to be multiplied by 1000.

If the CPC bid of an ad from advertiser A is 0.3 yuan, and the CPC bid of an ad from advertiser B is 0.4 yuan, the media platform will not directly sell the ad from advertiser B with a higher CPC bid, because the click-through rates of the two ads may not be the same.

At this time, the media platform usually uses a click-through rate prediction model to estimate the click-through rate (CTR) of each advertisement, that is, the probability p(m->c) from m to c, and then uses

eCPM=CTR CPC 1000

(It needs to be multiplied by 1000 because CPM is the price for 1000 impressions)

Calculate the eCPM (estimated CPM) of the two ads separately.

Assuming that advertiser A's click-through rate CTR_A is estimated to be 0.03 and advertiser B's ad CTR_B is 0.02, then we can calculate

eCPM_A = 0.03 * 0.3 * 1000 = 9 yuan

eCPM_B = 0.02 * 0.4 * 1000 = 8 yuan

Then sort by eCPM and display the ad of advertiser A with the highest ad cost.

It is worth noting that although the ranking is based on eCPM, the charging is not based on eCPM, but on CPC. That is to say, if this ad is clicked by a user, the media platform will charge the advertiser 0.3 yuan. If the ad is not clicked by users, the media platform will not charge the advertiser.

Behavior price conversion formula:

In the process of m->c->i->a->p, the price of each behavior A is equal to the price of the subsequent behavior B multiplied by the ratio of behavior A to behavior B. For example

CPM=p(m->c) *CPC * 1000

(The only thing you need to do is multiply the CPM by 1000)

CPC=p(c->a)*CPA

This is one of the most commonly used formulas in advertising algorithms. It can be used to convert the prices of different behaviors. It will be used frequently later, so please be familiar with it.

2. The “Four Points” in “Four Points, Three Rates, Two Controls, and One Strengthening”

1. Bidding point:

In fact, no matter whether it is CPM, CPC or subsequent CPA, OCPM, oCPC, double bidding and other bidding modes, most media platforms will ultimately convert the bid into eCPM based on the conversion formula, and then sort the eCPM and select the advertisement with the highest eCPM to display. In this way, the benefits of each media exposure opportunity can be measured directly, and the media’s own interests can be maximized. In this article, the “bidding points” we discuss are all at m (impressions).

2. Billing point:

That is, the media platform ultimately charges advertisers based on the number of ads.

For example, in CPM, charges are based on the number of impressions, so the billing point is m. In CPC advertising, although the eCPM ranking is used to determine which ad to display, the ads that are displayed are charged based on the number of clicks, and the billing point is c.

3. Bidding point:

That is, the price that the advertiser fills in the media platform's delivery background.

In CPM and CPC advertising, the bid point and billing point are in the same behavior. The bid point of CPM is also m, which is the price per thousand impressions, while the bid point of CPC advertising is c, which is the price per click. In oCPX, the bidding point and the billing point are separated.

4. Assessment points:

In advertising, what advertisers really want to bid for may not be impressions or clicks, or even downloads and payments, but LTV (Life Time Value), that is, all the value that the user brings to the advertiser during the life cycle of using the advertiser's product.

In a perfect world, advertisers could bid, for example, $0.80 for every $1 of LTV. Then the advertiser’s ROI is 1/0.8=1.25>1, and the advertiser can just sit back and count the money. Unfortunately, LTV is usually difficult to calculate and quantify. For example, even if users do not pay for props, they may still play the value of accompanying players. Therefore, when advertisers are placing ads, they usually look for a more quantifiable indicator that is further up the chain to evaluate the work of the delivery optimizer.

For example, when evaluating the cost of a paying user or the seven-day ROI, we call this point the evaluation point. In the following analysis, we assume that the assessment point is payment (abbreviated as p). In practice, it can be another point, but it does not affect the conclusion of our discussion.

As shown in the figure, in the CPM bidding mode, the bidding point/billing point/bid point is m, and the assessment point is assumed to be p. For the CPC bidding model, the bidding point is m, the billing point/bidding point is c, and the assessment point is assumed to be p.

Corresponding to the four points are three ratios, p(bidding point->billing point), p(billing point->bid point), and p(bid point->assessment point).

The big difference between CPM/CPC and all the bidding models discussed later in this article lies in the different positions of these four key points. As long as we clearly analyze the impact of these four key points and the three ratios between them on the interests of the three parties, we can extend it to all bidding models.

Next, we gradually introduce each ratio and analyze its impact on the interests of the three parties.

In addition, users (Internet users) are actually also a participating role, but each media considers user demands (user experience) to different degrees and in different ways, so in this article, we only discuss the game between the media, platforms, and advertisers.

3. “Three Rates” p (Price Point -> Assessment Point)

Let’s take a look at how advertisers set bid points under CPM and CPC?

CPM bidding model:

Suppose we are promoting a game, and through calculation we find that the cost of acquiring a paying user is less than 80 yuan, we can make money. In other words, advertisers (or delivery optimizers) have an implicit bid for paying users in their minds (i.e. implicit CPP = 80 yuan), which will be referred to as paid bid in the future. Then the advertiser will estimate the probability of p(m->p) based on the previous delivery data. Assuming that the estimated click-through rate CTR p(m->c) is about 0.03, p(c->i) is about 0.1, p(i->a) is about 0.4, and p(a->p) is about 0.1, then the entire p(m->p) is 0.03 0.1 0.4*0.1=0.00012, so there is

cpm = p(m->p)* implicit bid*1000

cpm = 0.00012 * 80 * 1000 = 9.6 yuan

Therefore, the advertiser's CPM bid is set at 9.6 yuan.

Some readers who have placed advertisements may say, "I never calculated this way when I placed advertisements." Yes, in practice, advertisers do not always do this explicitly. They usually set an empirical CPM first, and then look at the paying user cost in the report. If the paying cost is higher than 80, they lower the price; if it is lower than 80, they raise the price. The actual effect achieved in the end is the same as described above:

That is, the advertiser's bid will implicitly include a p (bidding point->assessment point) estimated by a fixed value, which is p (m->p) in CPM, so that the bid of the assessment point (paid bid) is converted into the bid of the bidding point (i.e. CPM bid).

CPC bidding mode:

Similarly, the advertiser's CPC bid implicitly includes a p (bid point -> test point) estimated with a fixed value, which is p (c->p) in CPC, so that the bid of the test point (paid bid) is converted into the bid of the bid point (i.e. CPC bid).

We found that in CPM and CPC (actually, it also applies to other bidding modes), the ratio p (bidding point -> assessment point) is designed, that is, the ratio from the bidding point (m for CPM, c for CPC) to the assessment point (p). Next, let’s take a look at the true value of this ratio and the accuracy of the estimate, and their impact on the interests of both parties in the game (i.e., media platforms and advertisers). Put the conclusion first, then analyze them one by one.

p (bid point -> assessment point) The impact of the true value of the ratio on media and advertisers

By increasing the true value of the ratio p (bid point -> assessment point), advertisers seeking volume can obtain a higher bid point price at the same paying user cost.

For example, the CPM price in the CPM model (because cpm = p(m->p)* paid bid*1000). This will help the media platform win more display opportunities that would otherwise not be won and obtain more volume, so its revenue will also increase. For advertisers with limited budgets and no volume requirements, by keeping the bid point unchanged, they can gain more paying users and improve ROI.

Therefore, when it comes to improving the actual value of p (bid point -> assessment point), media platforms and advertisers have exactly the same interests. For example, under the CPM model, media platforms will improve p (bid point -> assessment point) by optimizing the position of ad slots and thereby increasing click-through rates. Because of the alignment of interests, advertisers can trust the advice and tools provided by media platforms in this regard.

p (bid point -> assessment point) The impact of the accuracy of ratio estimates on media and advertisers

For any user, advertisers explicitly or implicitly use a fixed value (although it will be adjusted, it is fixed for different traffic at the same time) to estimate p (bid point -> assessment point), thereby converting the paid bid into a bid point bid. Obviously, this ratio is different for each user. Therefore, this estimate may sometimes be too high, causing the bid point to be too high, resulting in a decrease in advertiser ROI, and increased revenue for the media platform; sometimes it may be too low, causing the CPM value of the bidding point to be too low, resulting in unfavorable bidding and failure to obtain sufficient volume, resulting in a decrease in revenue for the media platform.

If the p (bid point -> assessment point) estimated by the advertiser using a fixed value is too high or too low, the advertiser's own interests will be damaged. Therefore, the advertiser has the motivation to adjust this estimated value by adjusting the bid to improve accuracy.

4. “Three Rates” p (bidding point->billing point)

For CPC, the difference from CPM is that the billing point and bidding point are both c, while the bidding point is always m. Therefore, there will be a rate of p(bidding point->billing point). Next, we will discuss the true value of this ratio and the accuracy of the estimate, and how it affects the interests of both parties in the game (i.e., media platforms and advertisers). Put the conclusion first, then analyze them one by one.

The impact of the true value of the p (bidding point -> billing point) ratio on media and advertisers

By increasing the actual value from the bidding point (m) to the billing point (for example, c in the CPC bidding model), advertisers seeking volume can get a higher CPM at the same payment point. (For example, in the CPC bidding model, CPM=CTR CPC 1000), winning more display opportunities that were originally not won, and getting more volume. Because CPM increased, the media also earned more revenue.

However, this may not be the case for advertisers who are not looking for volume. In this article, advertisers who do not pursue quantity refer to those who, after obtaining a certain quantity, no longer seek to continue to expand their purchasing volume.

At this time, some friends may ask, as long as the actual paying user cost is lower than the acceptable maximum paying user cost estimated based on the user LTV, advertisers should have as much as possible, so that the total profit will be higher. Why are there advertisers who don’t pursue quantity?

There are two typical types of advertisers whose demand for volume stops increasing after reaching a certain level:

One type is small agents or CP themselves. The money invested through advertising cannot be recovered in a short period of time. It requires a cycle. For example, heavy games may take several months, while casual games may take several weeks. For small agencies with insufficient financial resources or CPs that place their own advertisements, even if the ROI is greater than 1, their budget is limited.

The second category is advertisers with limited service capabilities. When the purchase volume increases, our service capabilities cannot keep up. However, in the Internet industry, this situation is usually rare because the marginal cost of increasing service capabilities is very low.

▶ Why do advertisers sometimes want to reduce click-through rates?

So for advertisers who don’t seek volume, why is it not beneficial for them to increase the actual value of p (bidding point -> billing point)? One reason that is easy to think of is that advertising is charged according to the number of behaviors at the billing point. For example, the billing point is at click c, and the advertiser will not be charged if there is no click.

Whether an advertising platform brings 2 clicks out of 100 impressions or 2 clicks out of 10 impressions, for advertisers, it is the same amount of money to bring the same number of clicks, and the increase in click-through rate does not directly benefit the advertisers. And in some cases, the lower the p (bidding point -> billing point) ratio is, the better it is for advertisers who are not looking for volume. This will become clearer if we look at a case from a non-advertising industry - telecommunications fraud.

Analogy question: Why are some scam text messages so poorly written?

A certain telecommunications fraud company first casts a wide net and sends fraudulent text messages in bulk. If anyone takes the bait, they will reply with a text message or call, and then use a whole set of rhetoric from the specialists to commit fraud. The cost of sending mass text messages is very low and can basically be ignored, which is similar to the display m in the CPC model; while further fraud through communication with specialists will require the specialists' time and energy, which is similar to the click c in the CPC model and requires costs. However, the fraud company has limited manpower, and the number of people who need specialist "service" must be limited, otherwise the number of calls cannot be met. So in this analogy, the scam company is the advertiser who doesn’t seek volume.

If the fraudulent text message is too deceptive and has a high "click rate", some people who are not easily fooled may not react for a while but still call the specialist for consultation. With the specialist's limited manpower, for example, answering 1,000 calls a day, only a relatively small number of users can be "converted" in the end. On the other hand, if the text message is as crude as in the picture, and the "click rate" is very low, then only very gullible people will call the specialist. Then, even if the specialist receives 1,000 calls, there is a greater chance of "converting" users.

Back to the Internet advertising industry, similarly, in some cases, p(bidding point->billing point) is negatively correlated with p(billing point->assessment point). This is why for advertisers who do not pursue volume, increasing p (bidding point -> billing point) may not necessarily be beneficial to them and may even be harmful. As a media platform, because the marginal service cost of the Internet is very low and advertisers seeking volume generally account for the majority, the impact is generally not very large.

The impact of the accuracy of p (bidding point -> billing point) ratio estimates on media and advertisers

Unlike CPM, the estimate from m to c, that is, p (bidding point -> billing point), is estimated by the media platform through machine learning. The estimated value of each user and each advertisement in different contexts is usually different.

If the estimated value of a certain advertisement is too high, then the CPM of the advertisement at the bidding point will be relatively high (because CPM=p(bidding point->billing point) * p(billing point->bid point) * bid at the bidding point), thus occupying the display opportunities of other advertisements. However, the actual billing amount will be less, and the revenue of the advertising platform will be reduced. For advertisers, if the number of billings is lower, the billing will be proportionally lower, and the final ROI will not change significantly.

If the estimated value of a certain advertisement is too low, then the CPM of the advertisement at the bidding point will be relatively low, and the advertisements that are not the best in the bidding queue will be placed higher up, resulting in lower revenue for the advertising platform. For advertisers, the amount they get is less.

So we see that the more accurate the media platform's estimate of p (bidding point -> billing point), the higher the revenue. For advertisers, the main impact is the amount of advertising revenue they can earn, which is relatively small.

▶ From CPM to CPC, what changes have taken place in the bargaining relationship between media platforms and advertisers?

The impact of CPM to CPC on advertisers:

If the assessment point is at or after c, from CPM to CPC, the interests of advertisers are greatly guaranteed. It turns out that when advertisers use fixed values ​​to estimate p(m->c), it will be very inaccurate, and whether it is too high or too low, it will have a negative impact on the advertiser's ROI or volume.

In CPC, p(m->c) belongs to the section from bidding point to billing point. This value (click-through rate) is estimated by the media platform using a machine learning model that understands user characteristics and labels (i.e. c). It is much more accurate than the estimate by the advertiser using a fixed value, and if the estimate is too high, the advertiser's ROI will not be significantly affected. Therefore, there are almost only advantages and no disadvantages for advertisers to use CPC instead of CPM.

For some brand advertisers who only evaluate m and not c, the evaluation point is m, and the above analysis is not applicable. Such advertisers will not use and have no need to use CPC.

The impact of CPM to CPC on media platforms:

If the advertiser's evaluation point is c or after c, the change from CPM to CPC will bring more benefits than disadvantages to the media.

In CPM, p(m->c) is a fixed value estimated by the advertiser for all ads. Assuming that the advertiser's estimate is very accurate, the average of all ads is used as the fixed value. Then nearly half of the advertisements will be overestimated and half will be underestimated. After a period of time, advertisers will find that many bids with low estimates fail, while bids with high estimates are more likely to succeed. Therefore, when looking at the report, the average cost of the ads delivered is significantly higher than the bid.

At this time, in order to achieve the target cost, the fixed value used for estimation must be lowered (that is, the bid point is lowered), resulting in most estimates being on the low side, which means that the media platform's revenue is mainly reduced.

The actual delivery data of the CPC model shows that although the media platform bears the consequences of inaccurate p(m->c) estimation (because whether it is too high or too low, the media platform will pay for it), but because the estimation is too accurate, the loss is smaller than the loss caused by the decline in media platform revenue due to the advertiser's obviously low estimation in CPM. So overall, the media platform’s revenue also increases under the CPC model.

In addition, regarding the actual value of the ratio, under the CPC model, advertisers with limited budgets have different interests and media platforms in terms of whether to increase the m->c ratio. Fortunately, the impact of this part is relatively small. In actual delivery, CPC still has more advantages than disadvantages for media platforms.

▶ When can the CPA model be used?

The success of CPC makes people think about a question: whether it is possible to follow the same method and move the billing point and bidding point to a, which is the CPA model.

If the data of behavior a is fully controlled and collected by the media platform, then there is no problem and the CPA model is valid. For example, if the media platform is Taobao and advertisers promote their products on Taobao, then Taobao can fully control the behavior of a, so it is possible to do CPA (or even CPS).

If the data of behavior a is collected by the advertiser and then sent back to the media platform by the advertiser itself, there will be problems if the billing point is moved to a. For example, if an advertiser bids 80 yuan for action a, then the advertiser has an incentive not to send back the data of action a, or to withhold part of the data of action a to the media platform, thereby allowing the media platform to collect less money from the advertiser.

So we can get a rule:

Payment points usually cannot exceed the behavior points that are fully controlled by the media platform and cannot reach the behavior points collected by advertisers.

5. “Three Rates” p (billing point -> bidding point)

According to the above analysis, if behavior a is collected by the advertiser, then we cannot directly bid and charge based on CPA. The estimate of the p(c->a) segment can only be estimated by the advertiser using a fixed value. According to the above analysis, this is disadvantageous to both advertisers and media platforms.

The characteristic data (user characteristics and advertising characteristics) estimated by the ratio p(c->a) are in the hands of the media platform, but the label is in the hands of the advertiser. Therefore, if p(c->a) is to be estimated, one party must give in and give the data to the other party.

Usually, media platforms have better data analysis capabilities than individual advertisers. Advertisers give label data to media platforms (the following will analyze why advertisers are motivated to do so), and the media platforms make estimates based on p(c->a). At this time, FB's innovative oCPX appeared, which separated the bidding point and the billing point, with the billing point at m (or c) and the bidding point at a.

We can understand oCPX in this way: oCPX = separation of billing point and bidding point + intelligent bidding control under continuous bidding

That is, advertisers actively provide behavioral data to the media platform, and the media platform provides advertisers with free bid conversion services to convert bid point prices into billing point prices (by estimating p(c->a)). In addition, the media platform provides intelligent bidding control services under continuous bidding. The smart bidding control service under continuous bidding will be discussed in detail later.

The impact of the real value of the p (billing point -> bidding point) ratio on media and advertisers

The higher p (billing point -> bidding point), the higher the CPM bid converted from the advertiser's bidding point to the billing point bid and then to the bidding point will be, and the advertising platform's revenue will increase. For advertisers seeking volume, the higher this ratio, the higher the converted CPM bid will be, and the easier it will be to get volume. For advertisers who are not looking for volume, by keeping the pay-per-point bid unchanged, more bid point behaviors can be obtained and ROI can be improved.

We can see that the interests of media platforms and advertisers are consistent in increasing the true value of the p (billing point -> bidding point) ratio.

The impact of the accuracy of p (billing point -> bidding point) ratio estimates on media and advertisers

If the estimated value predicted by the media platform using the model is too high, the advertiser's bid at the billing point will be too high, resulting in a decrease in ROI. If it is too low, the CPM converted to the bidding point will be too low and no volume will be obtained. For media platforms, a higher estimated value will translate into a higher billing point bid and bidding point CPM bid, resulting in increased revenue. If the estimate is too low, the CPM converted from the bid point will be lower, the competition in the advertising queue will not be fierce enough, and the media platform’s revenue will decline.

Here we will find a problem. When the p (billing point -> bidding point) estimated by the media platform is too high, the media platform itself benefits, but it is disadvantageous to advertisers. Media platforms have the motivation to overestimate this ratio in the short term. How can this problem be solved? We will explain this in detail in the next article.

So far, the impact of the three ratios of four key points in the large media platform on both sides of the game has been fully analyzed. There are various basic and smart bidding modes, and the big difference lies in the different positions of these four key points. One of the keys to understanding these bidding patterns is to clearly analyze the impact of these four key points and the three ratios between them on the interests of the three parties.

This article introduces the "four points and three rates" in the "four points and three rates, two controls and one strengthening", as well as basic bidding modes such as CPM/CPC/CPA. In the next article of this series, we will continue to introduce the "two controls and one enhancement" in the "four points, three rates, two controls and one enhancement", as well as analyze various smart bidding modes such as oCPX, double bidding, activation and payment, AEO, and VO. Hope you don't miss it.

Author: Shentanshe

Source: Shentanshe