Advertising placement: Strategies to prevent over-delivery of performance advertising!

During the process of performance advertising , if the problem of overdelivery occurs, it will not only cause advertisers trouble in allocating budgets and controlling the delivery rhythm, but also make advertisers question the reliability of the platform. Faced with this situation, what preventive measures should product managers take in advance?

In the process of delivering performance advertising, the problem of overdelivery is often encountered. For example, an advertising group with a cost of 200 yuan may actually cost 230 yuan. If there are many such over-invested advertising groups and the amount is not large, it will cause advertisers to have trouble allocating budgets and controlling the rhythm of delivery, and it will also make advertisers question the reliability of the platform.

Faced with this situation, how should product managers solve it?

Let’s first look at why advertising is overdelivered.

Taking CPM bidding ads as an example, before the ad billing has reached the limit and is notified to be offline, roughly speaking, the ad needs to go through five stages: request, retrieval, screening, sorting and return, and exposure. The time difference in the middle will result in the ad being offline commanded but the remaining ads that have been retrieved and sent cannot be terminated, and will still incur charges due to being displayed. The greater the traffic volume and the higher the number of concurrent requests, the more significant the oversubscription phenomenon will be.

Figure: Advertisement placement and billing process using CPM bidding as an example

What are the commonly used strategies when facing over-delivery of advertisements? Let's go through them one by one:

1. Shorten the time interval between the settlement issuing offline instructions and offline advertising

As can be seen from the above, the main reason for overdelivery is the time difference between the instruction to take the ad offline and the actual time when the ad is taken offline. The longer this time difference is, the more advertisements will be issued during this period, and the greater the amount of over-investment will be. To solve this problem, we can use technical means to optimize the processing speed of the program and minimize the time difference.

However, since a series of actions such as retrieving ads occur before the ad offline instruction is issued, even if the technology reduces the time interval between the settlement issuing the offline instruction and the ad offline to "0", it is still impossible to completely and effectively avoid ad overdelivery.

The mainstream practice in the industry is to reduce over-delivery through flow control. Let’s take a closer look

2. Reduce the number of ads retrieved before the offline command is issued, i.e. flow control

The main purpose of traffic control is to slow down the delivery of ads under the account when the account balance is low, and to reduce the number of ads retrieved before the offline instruction is issued. This method can significantly control overinvestment. The general centralized method for flow control is:

1. Probabilistic algorithm strategy - reduce the proportion of retrieval according to the consumption ratio

The probabilistic algorithm is a relatively simple algorithm developed by the advertising strategy product manager. It determines the probability of an advertisement being retrieved based on the ratio of actual consumption to target consumption (limit). As the actual consumption approaches the target consumption, the probability of the ad being retrieved decreases. This reduces the number of ads retrieved before the offline command is issued.

Advantages: The algorithm requirements are relatively simple, the personnel cost is low, it can be implemented by small companies, and it is quite effective. Can greatly reduce overinvestment

Disadvantages: The control process is of large granularity, which can easily cause the advertising budget to be not smooth enough, with fast consumption in the early stage and slow consumption in the later stage. It can only solve the problem of overdelivery, but cannot help advertisers optimize delivery results.

2. Pid algorithm strategy-control the delivery speed

Pid algorithm is the abbreviation of the first letters of proportion, integral, and derivative. The pid algorithm is a very classic and widely used control algorithm. It was first used in automatic control. It can be used in everything from the temperature control of a small component to the design of the flight path and flight speed of a drone.

The principle of the Pid algorithm is to dynamically use proportion, integration, differentiation and winning rate prediction based on real-time delivery feedback data to control the uniformity and stability of the delivery speed. The basis of PID control is proportional control; integral control can eliminate steady-state errors, but may increase overshoot; differential control can speed up the response speed of large inertia systems and reduce overshoot tendencies. The prediction is made based on the mathematical fitting of the flow fluctuation waveform and the delivery progress waveform using "Fourier Transform".

Figure: Schematic diagram of Pid algorithm

It doesn’t matter if you don’t understand the concept. Let’s take a look at the advantages and disadvantages of this algorithm.

Advantages: After the introduction of the Pid algorithm, the control over the advertising consumption speed is more delicate, and there will be no sudden and drastic changes in the display opportunities at the critical point. It is very helpful for advertising overdelivery. It can extend the advertising delivery time, help consume the advertising budget evenly according to demand within the delivery time, and reduce the situation where the advertising budget is consumed in an extreme time due to excessive traffic.

Disadvantages: Requires a relatively strong algorithm team and big data team. The inability to dynamically adjust the delivery speed according to traffic changes may cause high-priced advertisers to exit the bidding market prematurely, which is not conducive to maximizing the overall revenue of the platform.

1. Pacing algorithm strategy - control budget smooth consumption

Simply put, the Pacing algorithm estimates budget allocation based on historical traffic share so that the advertisement covers as many effective users as possible. The specific approach is to introduce advertising ranking parameters into the algorithm, participate in the delivery rate PTR, and adjust the consumption speed through rt intelligent pace so that consumption can be consumed year-on-year according to changes in traffic. The goal is to make the consumption distribution curve as close as possible to the flow distribution.

Problem with traditional bidding algorithms: Not optimal when budget is limited

A simple second bid will cause high ecpm customers to be quickly consumed and withdraw from the bidding, which is not good for the ad space and the media. It attempts to maximize revenue at each opportunity independently of other auction opportunities.

① For ads that perform well, the budget is consumed too quickly. Exit the market early:

I. With the withdrawal of advertising, market competition will decrease. In the second-bid auction mechanism, reduced competition will lead to lower revenue;

II. Delivery with broad targeting requirements may seize a large amount of targeted traffic from a highly competitive area, resulting in the inability to consume delivery with strict targeting requirements. As a result, income decreases;

III. Ads will get a skewed performance among their target audience segments. That is, the audience is unevenly distributed;

② This approach delays the availability of the latest consumption information, resulting in overinvestment and subsequent loss of profit. That is, overdelivery caused by the time difference between the request and the actual consumption;

Advantages of the budget pacing algorithm:

The main idea of ​​the algorithm is to allocate displays based on overall traffic and transaction conditions. The goal is to distribute impressions for each flight evenly throughout the day.

①For each campaign, we can get a forecast of the impression traffic pattern visible during the day, based on which we can determine a budget allocation plan. This allocation plan is determined based on the proportion of the delivered budget and the predicted reliable traffic.

②At runtime, we closely monitor the ad delivery volume of each campaign. If the delivery volume is consumed faster than the allocation, we will lower their display probability and not allow them to participate in certain auctions.

Algorithm practice:

3. Adjust pricing strategy and bring forward pricing timing

Through the above analysis, we can see that the main reason for overdelivery is that the notification of ad offline is later than the ad delivery. If the amount of advertising that needs to be delivered can be predicted before the budget is almost reached, can the problem of overdelivery be completely solved?

The answer is, yes. As long as the withheld fee reaches the budget, the status will be displayed as close to the budget. No new ads will be issued for ads that are close to the budget, and the budget consumption will be completed by relying on the remaining amount issued previously. At this time, the withheld fees are only estimated consumption and have not been actually paid.

Advantages : It can solve the problem of over-delivery at the root;

Disadvantages: It is necessary to prepare an estimate of the advertising exposure rate. Deducting fees in advance before the exposure may cause losses to the advertiser. This is an immature system, so caution is advised.

4. Platform system background control

The strategies mentioned above all use technical means to achieve control over advertising over-delivery. In addition to technical means, we can also reduce this situation through some backend controls, such as:

1. The minimum budget cannot be lower than a certain value, such as 50. Because the smaller the budget, the smaller the regulatory role the algorithm can play;
2. Advertisers are advised not to modify bids and budgets frequently. Not only will it disrupt the algorithm training process, but it will also create a blind spot for algorithm adjustment during bidding and budgeting, leading to over-investment.
3. When an advertiser reduces their budget, they set a limit. For example, if the adjustment is to take effect immediately, the lower limit of the budget adjustment needs to be greater than a certain value of the consumed amount. Or simply reduce the amount and take effect the next day;
4. No charge for overinvestment, etc.

The above is some understanding and practice of the strategy to prevent over-delivery in performance advertising. If there are any shortcomings, we welcome everyone to discuss them together.

Source: Huajiao Society (ID: i_huajiao)