Detailed explanation of OCPM/OCPC deep conversion dual bidding capabilities

Regarding OCPX, namely OCPC/ OCPM /OCPA, these advertising optimization modes are collectively referred to as oCPX advertising below.

Like most public account authors, I like to talk about my personal opinions and ideas before getting into the main topic, just like serving some side dishes before a main meal. But this time I decided to get straight to the point and get straight to the point.

1. What problem does double bidding mainly solve? What are the benefits for advertisers and advertising platforms?

We know that the ability of OCPX advertising is mainly to solve the problem of advertising conversion, such as the activation of applications and games, the filling of forms and leads, etc., but these conversions are shallow conversions for advertisers and cannot reach deeper conversions. For example, game advertisers hope to buy more paying users, application advertisers hope to have a better retention rate of activated users, and forms and leads hope that the filled forms and leads are effective conversion leads.

The closer the conversion is to the back end, the more guaranteed the effectiveness of the advertiser's advertising will be. Therefore, advertisers have sufficient demand to require that the effect of advertising be more inclined towards deeper conversion.

For advertising platforms, there are two main reasons that prompt them to continuously optimize deep-level conversions to protect the interests of advertisers.

First of all, the growth rate of my country's economy is not as high as in previous years. Advertisers have limited budgets, but there are still many companies with traffic, such as Tencent, ByteDance, Baidu, Kuaishou, etc., which all have very large advertising traffic (ie advertising inventory). The fierce competition for advertisers' budgets requires advertising platforms to continuously optimize to optimize advertising effects and ensure the interests of advertisers.

Secondly, obtaining these in-depth conversion data is conducive to improving the capabilities of advertising models. The deeper the advertiser's data, the more valuable it is. For example, a paying user of a game is worth more than 500 yuan, and in the automotive industry, for example, a BMW lead is worth more than 2,700 yuan (the automotive industry vertical APP Autohome has an MAU of about 65 million and an annual revenue of more than 9 billion, mainly in sales leads). Originally, advertisers were reluctant to send this data back to the advertising platform, but because advertising requires deeper conversions to ensure the effect, it must be sent back.

Once an advertising platform has such data, it will have a very good treasure trove of user data for its advertising model or for exploring other businesses. For example, a media that did not originally make games suddenly wants to make games. The paid user data of different game types sent back by advertisers is simply an extremely accurate user group. As long as the quality of the game is good enough, it can be targeted at this group of users.

In this way, both parties benefit, and bidding for deep-level conversions can be easier to promote.

2. What is double bidding? Why double bid?

Double bidding refers to bidding with two targets at the time of delivery, such as making a bid for ad activation and another bid for ad payment at the same time. This situation is the double bidding mode.

For general advertising bidding, advertisers bid according to a single goal, such as CPC, CPM or CPA. In theory, bids for deep-level conversions can also be single bids, and advertisers are also happy to see such a model, such as direct bidding for paid users, such as 500 yuan for a paying user.

It is not possible to directly bid for a single goal for deep-level conversions. There are two big problems with the model of bidding directly on a deep, single goal.

First, deep conversion data is very sparse (i.e., small in quantity). During the training phase of a new ad, it usually takes about 10 conversions to train a relatively reliable conversion data estimation model (some platforms will have fewer, but generally 6-10 conversions). For example, in general video information flow, the CTR of game ads is 1%, and the activation CVR is around 10%. A 10% activation-to-payment ratio is already pretty good. In other words, training 10 paid conversions requires 100,000 exposures (100,000 * 1% *10% *10% = 10). Based on the estimated ECPM of 60-80 for short video games, the traffic cost here is 6,000-8,000. Considering that some large platforms launch thousands of new ads every day, this will be a very high traffic cost for model training and learning. This is something that advertising platforms cannot afford.

Secondly, if charging is directly based on deep-level conversion goals and no charging if there is no conversion, advertisers have a lot of room for cheating. Cheating mainly occurs during the training period of new ads, and the method of cheating is to reduce the number of return ad conversions. Because once the training period of a new ad has passed, the advertising platform has accumulated enough conversion data to determine the status of the returned data in real time, thereby calculating the DCVR (deep conversion CVR) value to reduce the advertiser's bid ECPM.

During the training period of the new advertising model, the advertising system is unable to make this adjustment because it does not know whether the conversion data is sparse because there is insufficient conversion data or because the advertiser does not send back enough deep conversion data. Therefore, it will keep running until the ad exposure exceeds a certain set threshold. Some advertisers target this scenario by sending fewer conversions back on new ads. At this time, the system is unable to make an accurate judgment, resulting in a lot of traffic being pulled away.

So, in response to this situation, many advertising platforms have launched a dual bidding model, that is, one bid for shallow targets and one bid for deep conversion targets. For example, one bid for user activation and one bid for user payment. In the early stage, the model and bidding are run according to shallow targets, and deep conversion targets are gradually accumulated. When there is enough data for deep conversion goals, the advertising system will implement certain bidding strategies based on the advertiser's demands (it should be noted that this does not directly switch to a single deep bidding mode).

3. What is the difference between deep conversion OCPX double bidding and ordinary oCPX advertising?

First, in terms of the conversion data volume, deep conversion OCPX data is more sparse, and deep conversion data is transmitted back more slowly, some even every other day (such as second retention). The premise for the advertising algorithm to be quickly tuned is that the data flow back is fast enough. There are many differences between the deep conversion model and the ordinary OCPX advertising model due to the amount of data and the speed of return flow.

Secondly, the deep conversion OCPX delivery method must have dual-target bidding, that is, the transition from the first stage to the second stage must be retained in the short term. The advertising system is currently unable to completely offer only deep target prices and ignore shallow conversion target bids.

The oCPX double-bidding advertising model for deep conversions is very similar to when oCPX’s advertising capabilities were first launched, and both have two stages. When many advertising platforms first start using OCPX, they also go through two stages. In the first stage, they bid based on CPM or CPC. When the advertising estimation model has accumulated enough data, they switch to the OCPX model for bidding. However, with the advancement of technology, the first stage bidding was gradually cancelled.

Now, most of the advertising platforms we see on the market have actually cancelled the first-stage bidding for OCPX ads and directly bid on the shallow OCPX targets.

Deep conversion oCPX is also divided into two stages. In the early stage, shallow targets are bid according to OCPX, while deep conversion data is accumulated. Once there is enough in-depth conversion data, determine the corresponding bidding strategy for the ad based on the advertiser's needs. But this is not like ordinary shallow OCPX which directly converts from CPM or CPC to ECPM of OCPX deep conversion. Deep conversion OCPX has its own unique bidding strategies. The root cause is advertiser demand.

For example, for app advertisers, their dual goals are activation cost and retention rate. The advertiser's goal is to meet both the activation cost and the secondary retention cost. For game advertisers, they only need to pay the cost of deep conversion into paying users.

4. Deep conversion OCPX double bidding, what is the calculation formula of ECPM in bidding ranking? What are the ECPM value strategies based on advertisers’ needs?

Currently, most of the advertising systems on the market rank ads by ECPM. Whether it is shallow conversion or deep conversion, it will eventually be converted into ECPM value for comparison and ranking. Give priority to high-value ads.

The formula for calculating ECPM for shallow conversion OCPX is:

ECPM 1= α* CPA 1* pCTR * pCVR, where α is the dynamic pricing factor, CPA1 is the advertiser's shallow target bid, pCTR is the estimated click-through rate of the ad, and pCVR is the estimated conversion rate from ad click to shallow target conversion.

The general formula for calculating ECPM for deep conversion OCPX is:

ECPM2 = β* CPA2* pCTR * pCVR 2, where β is the dynamic pricing factor, CPA2 is the deep target bid for the ad, pCTR is the estimated click-through rate for the ad, and pCVR2 is the estimated conversion rate from ad click to deep target conversion.

In the early stage of advertising, shallow conversion ECPM1 will be used for bidding ranking. After the deep conversion data is accumulated, deep conversion ECPM2 will be calculated and bid with ECPM1 according to the following rules.

If the core demand of the advertiser is to meet the deep conversion goal, then when the deep conversion data is accumulated enough, the bidding ranking of the advertisement will adopt ECPM2 for bidding. At this time, the main cost assessed is the deep conversion goal, and the shallow goal is very likely to exceed the cost.

As shown in the figure below, general game delivery is divided into activation cost and payment cost, and the delivery is targeted at the cost of paying users. The main target users are the small ellipse on the right, that is, users whose activation cost does not meet the standard but whose payment cost meets the standard. We can understand that the conversion rate of this part of users is not so high, but once converted, they have a particularly strong willingness to pay. Advertisers have no reason to resist acquiring these users.

If the advertiser's core demand is to meet both goals, for example, the app needs to meet both the activation target cost and the retention target cost, then after the deep conversion data is accumulated, the ECPM of the advertising bidding ranking will be min (ECPM1, ECPM2), which is the two intersecting parts of the above figure.

Sometimes, due to the serious delay in data return, such as deep conversion is the retention rate or the lead to the store, since the advertising model cannot obtain the real-time data model, the ECPM of the deep conversion target will be calculated using the weight value, that is,

ECPM = α * CPA1 * PCTR * PCVR * λ (pDVR / target_DVR)

Among them, α is the dynamic pricing factor, CPA1 is the advertiser's shallow target bid, pCTR is the estimated click-through rate of the ad, pCVR is the estimated conversion rate from ad clicks to shallow target conversions, λ is the dynamic adjustment factor for deep conversions, pDVR is the estimated conversion rate from shallow conversions to deep conversions, and target_DVR is the target value for the conversion rate from shallow targets to deep targets. For example, this formula is used to optimize secondary retention and effective leads. According to the weight of the ratio of the estimated deep target conversion rate to the target conversion rate, high-conversion users can have higher competitiveness in the second stage and continue to approach the target deep conversion rate.

However, in actual applications, you will find various problems. Either the volume is not enough, or the deep conversion of the advertisement exceeds the cost. During this period, you need to continuously analyze the data, maintain active communication with the strategy algorithm team, and adjust the advertising strategy to ensure that the advertiser can obtain stable volume and effect.

In the demand for deep conversion, there are other optimization capabilities that can be explored in depth. For example, the current deep conversion in the gaming industry mainly buys the payment rate, not the amount that the user can pay. Paying 10 yuan and paying 1,000 yuan are the same deep conversion bids, so there is room for optimization here.

However, every improvement of the advertising model involves the exchange of core data. We will have the opportunity to talk about the in-depth analysis of this later.

Author: PMCoder Road

Source: PMCoder Road (pmcoder)