A brief introduction to the evolution of digital advertising

The advertising industry has a long history. With the advancement of technology, there are more and more opportunities for media and users to come into contact, so advertisers also have more opportunities to reach users. In the traditional advertising industry, there has been a saying that "I know that half of my advertising budget is wasted, but I don't know where it is wasted." This statement has been refuted with the development of digital advertising .

Since the first online advertisement was born on GNN in the 1990s, digital advertising has undergone many changes. This article talks about the changes in digital advertising product technology.

Participants in digital advertising

First, we will briefly introduce the relevant roles of the digital advertising ecosystem, which mainly include the following four categories:

1. Advertisers are the demanders of advertising, such as Apple and Volkswagen, who are brand owners that have the demand for advertising to market their products.
2. Media platform: It is the supplier and seller of traffic. Many Internet media are also providers of advertising platforms, so they are collectively referred to as media platforms, such as Tencent, Alibaba, ByteDance, etc.
3. Advertising audience: The general public who watch/consume advertisements are the potential consumers that advertisers want to reach.
4. Advertising agency: a service provider that provides creative design and advertising placement for advertisers, or a media agency that aggregates media traffic and redistributes it, such as the well-known 4A company. The evolution of digital advertising is also about innovating digital advertising transaction methods and upgrading digital advertising product forms around various participants.

01 Contract Advertising

1.1 Exclusive advertising priced in CPT

At the beginning of the Internet, portal websites such as Yahoo have already gathered a considerable amount of traffic. In pursuit of commercial realization, media platforms initially thought of borrowing the way newspapers and magazines sell page columns, splitting the many modules of the website into creative display banners, and selling contract advertising using the CPT (Cost Per Time) pricing method based on display time. The main technology of advertising products at this time is the scheduling system for controlling contract delivery. It does not require dynamic targeting of different customer groups and has relatively little reliance on technology.

Advertisements priced in CPT are exclusive ads. When a banner position is sold to advertiser A at a certain moment, advertiser B, a business competitor, will inevitably lose the exposure opportunity at that moment. Therefore, CPT contract display ads have a certain premium ability and crowding-out effect.

1.2 Targeted advertising based on CPM

However, from the perspective of media monetization, among the browsing users at the same time, there may be mountaineering enthusiasts and house-bound mothers. If cosmetics advertisements are displayed at this time, it will not be attractive to the former, and this part of the traffic will be wasted.

Naturally, media platforms have developed ways to display targeted advertisements to different users . This type of targeted display advertising, which is now commonplace, was originally sold to brand advertisers in the form of contracts, with an agreed pricing method based on the number of impressions, namely cost per thousand impressions (CPM, Cost per Mille) .

Audience-targeted advertising turns static window-style ads into dynamic ads based on "location + crowd", and the advertising system processes and produces rich crowd tags for advertisers to choose from. Crowd tags usually include three categories: attribute tags, behavior tags, and custom tags.

Attribute labels include demographic labels such as geographic location, age, gender, and device. Media platforms provide this type of coarse-grained labels after collecting and processing the information of registered users. For missing values ​​in some fields, data mining and completion can be performed using tools such as ALS. In addition, contextual scenarios, such as website channels (technology, fashion, finance) and article content types, can also be processed into relevant tags through TF-IDF to extract keywords and text topic models.

Behavioral labels are tags that predict users' preferences based on their historical behaviors, such as game enthusiasts, maternal and child shopping preferences, etc. The key to labeling is to select a set of online user behaviors as features and map them to specific directional labels. This actually turns into a classification problem, such as the early use of LR modeling to the gradual evolution of deep learning algorithms.

Custom tags are crowd selection that is deeply processed based on the customized needs of advertisers, such as redirection and new customer recommendations (look-alike). Currently, most of them are connected with the DMP data management platform of the advertising system.

02 Bidding Advertising

2.1 From 'Planned Economy' to 'Market Economy'

The earliest bidding advertising originated from Google. As the main monetization method of search engines, keyword advertisements are embedded in search results through auctions, which can achieve precise traffic control and higher commercial returns.

At the same time, the display ads of major portals that were still sold under contract also encountered many problems.

First, contract advertising has an agreed quantity, while advertisers have increasingly sophisticated audience-targeted delivery needs, which poses a great challenge to media platforms in making more segmented traffic distribution within the agreed quantity restrictions.

Secondly, contract advertising has a competitive advantage for large brand owners with sufficient budgets, while small and medium-sized businesses also have strong marketing demands;

In addition, contract advertising lacks transparency for advertisers, and the optimization of effects is completely dependent on the media platform.

Driven by search bidding ads and the above-mentioned problems, display ads have begun to break away from the 'planned economy' model of contract pricing and develop towards the 'market economy' model of free bidding.

2.2 More pricing methods

In the search bidding model, advertising content is mixed with search results content, and sometimes users may not even realize that they are clicking on an ad. Search ads are more naturally combined with actual services, which is also the natural difference between them and window-style display ads. Therefore, the billing model no longer adopts display payment, but adopts the CPC (Cost Per Click) pricing method.

The growth of the bidding advertising market has also put advertisers in a situation of multi-party game. For advertisers, what they care about is the conversion payment obtained from consumers after the advertisement is placed, that is, the return on investment (ROI). Under the CPC pricing model, the media platform is only responsible for the number of clicks it brings, while the estimation of the step from click to conversion is borne by the advertiser. This leads to two paradoxes:

First, advertisers are too optimistic and over-estimate, which results in higher costs. Although media platform revenues increase, advertisers’ ROI decreases, which is not conducive to overall market growth.

Second, advertisers underestimate the forecast and lower the bid to reduce consumption. As a result, advertisers cannot obtain enough expected traffic and the revenue of media platforms also decreases. Therefore, advertisers tend to pay closer to conversion, and media platforms are also motivated to expand the market, which leads to the CPS (Cost Per Sale) and CPA (Cost Per Action) pricing models.

So has CPS/CPA become the best pricing method? Not entirely true. The reasons will not be discussed here and will be analyzed below.

2.3 eCPM, a key indicator for evaluating advertising effectiveness

After paying by click or conversion, does it mean that as long as the advertiser bids high enough, they will definitely be able to obtain the media’s advertising resources? No, even if the bid is very high, but the user has no interest in the ad and does not click on it, the media platform still cannot collect any fees. Therefore, the quality of an advertisement needs to be quantified, and eCPM is a key measurement indicator.

eCPM (expected Cost Per Mille) is the expected revenue per thousand impressions , which is equal to total ad revenue/total number of ad impressions*1000.

We know that an advertisement usually goes through three actions from exposure to conversion and payment for the advertiser, namely exposure M, click C, and payment (conversion) S (A). The probability of a user clicking on an ad after being exposed to it is the click-through rate P (M→C) , and the probability of a user making a payment (conversion) after entering the advertiser's landing page is the conversion rate P (C→S) .

Under the CPM pricing method, eCPM=CPM. For example, if the CPM bid of advertiser A is 5 yuan per thousand times, and the CPM bid of advertiser B is 3 yuan per thousand times, the media will directly display the ad with the higher bid.

Under the CPC pricing method, eCPM=click rate*ad unit price*1000=P(M→C)*CPC*1000 (multiplied by 1000 because CPM is calculated based on thousand impressions). For example, if advertiser A's CPC bid is 1 yuan, and the click-through rate of ad A is 0.02; advertiser B's CPC bid is 0.8 yuan, and the click-through rate of ad B is 0.05, then:

• eCPM_A=0.02*1*1000=20 yuan
• eCPM_B=0.05*0.8*1000=40 yuan

After sorting by eCPM, the ad that is finally displayed is from advertiser B with a lower unit price.

Under the CPS pricing method, that is, by estimating the conversion rate one more step under CPC pricing, it is easy to get eCPM=click rate*conversion rate*ad unit price*1000=P(M→C)*P(C→S)*CPS*1000.

Using eCPM as the ranking basis is a major difference between advertising products and search recommendation products. Since the estimated eCPM is ultimately multiplied by the advertising unit price, ad click-through rate prediction is more suitable for modeling as a regression problem rather than a ranking problem.

03 Programmatic Advertising

3.1 Real-time Bidding

The emergence of the bidding model has provided opportunities for many small and medium-sized Internet media to realize their traffic, and gradually gave rise to the AdNetwork (Advertising Network, online advertising alliance) business model. Well-known ones include Google's AdSense and Baidu Advertising Alliance.

However, AdNetwork's traffic sales are a 'black box' for advertisers. Advertisers cannot customize audience labels, and data feedback is not timely. The demand for customer customization and effect optimization has given rise to RTB (Real Time Bidding) , which in turn opened up a new business model for programmatic advertising.

The core process of RTB is: when a user visits the media, the ad space information and user ID (cookie information or deviceID, etc.) are used to generate a price inquiry request and send it to each advertising demander. The demanders determine whether they are the target audience and make bidding decisions. After multiple parties bid in real time, the advertising creative with the highest eCPM is finally displayed to the user.

The open bidding process of RTB requires the support of a programmatic trading platform, which is the advertising trading platform ADX (Ad Exchange) .

3.2 Advertising Exchange Market

ADX is like the stock exchange in the material world. On one hand, it connects the stocks (advertising spaces) owned by the media for sale, and on the other hand, it connects the demands of investors (advertisers) where the highest bidder wins. Those valuable advertising spaces are like blue chips and are sought after by many buyers. At the same time, the remaining traffic of small and medium-sized media can also find suitable buyers to generate transactions.

Some well-known ADXs on the market include the advertising platforms of BAT, ByteDance’s Massive Engine, Google’s AdX, etc.

Just as there are many stock exchanges in the world, there is more than one ADX as mentioned above. The docking specifications and operating procedures of different ADXs are not the same. It would be a cumbersome task to ask all the high-end advertisers to connect one by one. Therefore, the demand-side platform DSP (Demand Side Platform) dedicated to serving advertisers came into being.

DSP acts on behalf of advertisers to unify the connections between various ADXs and integrate some common needs for advertising, providing advertisers with a more user-friendly operation and management interface. Advertisers set the price of which type of customer and how much they are willing to pay to acquire them. DSP is responsible for executing ad bidding, customer selection, and ad delivery to obtain the maximum return at the lowest possible cost.

So how do you strive to get the maximum return at the lowest cost, or what are the core strategies of DSP for solving advertising optimization? The answer is eCPM. We know that eCPM = click-through rate * bid * 1000. Advertisers certainly hope that the click-through rate under this formula is as high as possible and the bid cost is as low as possible. For example, if you find that there is abundant traffic between 20:00 and 22:00 in a day, the market bidding competition is not fierce, and the audience's ad click response is relatively high, you can choose this targeted period to obtain traffic with a lower bid; for example, if you find that the audience who click on the ad are mainly 20-25 year old male users in second-tier cities on QQ space, you can further optimize the media channel + region + age.

In the process of advertising optimization, the accuracy and segmentation of targeted audience data have become top priorities. However, DSP systems do not necessarily have powerful data acquisition and processing service capabilities, so professional solution providers, namely data management platforms (DMPs) , have emerged in the market.

Simply put, data management platforms usually have three types of product capabilities: data collection, labeling, and data application.

Data collection: The acquisition and collection of metadata is the basis of DMP. From the perspective of data asset ownership, DMP can obtain relevant data from first-party advertisers, second-party media platforms, and third parties. Data collection methods include embedding JS code/SDK in the advertiser's own webpage or application, or obtaining data through manual uploading by the advertiser.

Labeling: Secondary processing of metadata to generate audience-specific labels. For example, when expanding the Look Alike population, you can use seed users who have converted from ads as historical data to train the model, and then input the characteristic data of new customers into the model to determine whether they are potential target audiences, thereby generating new customer recommendation tags for similar audiences.

Data application: Output customized tags to DSP/ADX for advertising or data analysis and report generation of relevant tag population portraits. Of course, there are products for service demanders, and naturally there are also products for service suppliers, namely the advertising traffic supply side platform SSP (Supply Side Platform). Many Internet giants also operate ADX and aggregate a large number of media advertising spaces. SSP will not be further elaborated here.

Programmatic advertising trading market relationship diagram (picture from "Programmatic Advertising Practice")

04 Native and Mobile Advertising

Since the advent of mobile Internet, native advertising, which mixes advertisements with content, has gradually become the mainstream. This is because the mobile Internet has more touchpoints with users, and integrating advertising into a unified content display strategy is also more user-friendly.

4.1 Smart Bidding

After the development of information flow advertising, the bidding transaction method also began to undergo new changes. As mentioned above, the CPS/CPA pricing method is not the optimal solution at present. This is because the result data with conversion results as the core of optimization is controlled by the advertiser. If the advertiser intentionally does not send the data back to the media platform, there will be a risk of cheating by the advertiser.

Facebook was the first to propose an innovative solution, that is, the media platform still uses advertising sales or conversions as the optimization goal for advertising display and traffic distribution, while advertisers pay in CPC/CPM. This is the optimized smart bidding oCPX (optimizedCPC/CPM) that separates the advertising deduction point from the advertiser's bidding point .

For example, if an advertiser uses product sales as an optimization goal, the advertising system will use historical data and the result data sent back by the advertiser as the basis for further adjusting the automatic bidding of advertisements. For users with a high willingness to purchase products, the bid will be increased to increase exposure and display; for users with a low willingness to purchase products, the bid will be lowered to reduce the waste of advertising exposure.

At present, the oCPX method has gradually become the standard for major media advertising platforms.

4.2 Limited openness of the ‘digital garden’

Mobile internet access enables applications to obtain more user data, while privacy protection is also gaining increasing attention. Some giants have introduced relevant policies on privacy and security. In April 2018, Google's advertising system announced that it would no longer provide log-level data feedback with user ID; in September 2020, Apple's iOS14 system pushed in China announced that it would not support the IDFA Opt-in solution (opt-in), and would no longer return the device ID (device ID) to the App by default.

In the mobile era, without device ID, ADX and DSP have lost the identifier of data mapping, and the RTB market that was once as public and open as a bazaar seems to be starting to become unworkable. At the same time, based on the privacy and security of data assets, media platforms also tend to "privatize" the originally public traffic resources. They do not want to provide such fine-grained information with user identity identification (such as user ID and related advertising data) to the outside world, and tend to build a limited open digital advertising platform.

Without data feedback, how can advertisers customize audience tags to guide their delivery?

The media platform's DMP remains a powerful tool and continues to support advertisers in selecting tags for advertising. However, DMP mainly contains data from media platforms and lacks data from advertisers, such as download conversions, retention, and trial class applications in the back-end link. Media platforms support advertisers uploading data, but advertisers are also concerned about the risks of privacy data security.

How to achieve data exchange without involving privacy leakage? 'Digital neutrality' is one solution. Simply put, the media platform provides an independent data storage space. Advertisers upload first-party data to this privatized space, and the media platform also uploads its own data. Data in all fields are processed using encryption technology. The media platform provides data processing and analysis capabilities to connect the data of both parties, thereby further conducting data mining, modeling, and application.

All media platforms have begun to provide this set of solutions, such as Google's Ads Data Hub and Tencent's Data Joint Zone.

05 Final words

Digital advertising originated from offline media, and its programmatic computing and audience targeting features also bring about advertising experiences that are different for each individual. Digital advertising is a multi-party market, and some game theory knowledge involved is also very interesting. This article only briefly touches on the many concepts of digital advertising. If you want to learn more, I recommend reading "Computational Advertising: Markets and Technologies for Internet Commercial Monetization".

Author: Strategy

Source: StrategyGadget