How far is ASM from AI?

According to many predictions, Apple's search ads will enter the domestic market around June this year. Faced with this big cake, those who previously engaged in ASO , SEM , and some advertising agencies have launched ASM services. In addition to traditional manual delivery, there are also some ASM services that take advantage of the AI ​​trend and claim to be " artificial intelligence ", which have attracted everyone's attention.

As a researcher who has experienced a series of trends such as cloud computing, big data , and artificial intelligence, and is actually still doing one thing, I have a deeper understanding of AI. Here I would like to share with you some of my views on AI+ASM. When ASM really enters the domestic market, you can also use it as a reference when choosing services.

Agent? Service provider?

Recently, some domestic advertising companies have launched Apple search advertising services, and many of them claim to be Apple's "agents." But those who are familiar with the international advertising business know that there are very few "agents" abroad who collect "commissions" from advertising platforms like those in China. Although foreign advertising service companies are also called agencies, most of them are "service providers" that charge advertisers fees.

A typical example abroad is the SEM ecosystem centered around Google advertising. Since Google opened its advertising API as early as around 2005 , "technology-based" companies and " marketing- based" companies have been developing in parallel in the international market. Marketing companies, such as large advertising groups like WPP , are similar to domestic "KA agencies". They mainly use manual services and combine some third-party technical services to provide SEM services to advertisers. Technology-oriented ones, such as small and medium-sized SEM companies like Kenshoo and Wordstream , and Adobe Cloud Marketing, which had revenue of over $1 billion in 2016, can manage advertisements on multiple bidding platforms such as Google. These companies mainly provide technical services, such as keyword expansion recommendations and automatic optimization of advertising plans.

Although Google followed local customs after entering the Chinese market, it also started a rebate agency system. But based on Apple's usual practice, it is basically difficult to set up "agents" specifically for the domestic market.

Therefore, it is estimated that it is unlikely that the domestic ASM market will see a situation like Baidu SEM, where large star agencies dominate the market. The most likely scenario is that large "marketing-oriented" advertising companies will continue to provide ASM services to existing clients, but a group of new "technology-oriented" companies will surely emerge to compete for the market. ASM's entry into the domestic market will bring more than just a traffic channel ; it is likely to bring about substantial changes to the domestic advertising service business model . ASM services and CP operations should have more options.

AI and ASM

ASM's business model determines that a group of "technology-oriented" service companies will definitely emerge, and it is not surprising that they adopt popular AI technology.

As for what is AI? The concept of AI (artificial intelligence) was actually proposed as early as 1956. There are various definitions, but generally speaking, anything that can replace human mental labor can be called AI. AlphaGo is AI, but we make an ASM delivery rule: "If the ad impressions of a keyword remain zero for three consecutive days, then the bid will increase by $0.1." This is a typical "expert system" rule, which is also AI.

Figure 1 The development of artificial intelligence has experienced many ups and downs

As for the hottest "deep learning" in the AI ​​field, a typical example is Baidu's Phoenix Nest advertising system. Previously, a large number of programmers worked overtime every night, mostly to make "thousandths" improvements to the click-through rate of advertisements. After Dr. Yu Kai came, he introduced deep learning, and the effect was a "percentile" improvement. AI technology has improved significantly, but it is hardly revolutionary.

AI has become popular again recently. In addition to the demonstration effect brought by products such as AlphaGo, what is more important is the reduction of technology application threshold and hardware costs, allowing AI to be implemented in more fields.

When it comes to products like ASM, whether it is suitable to use AI and how to use AI ultimately depends on product requirements. ASM has several specific problems that need to be solved regardless of manual or AI-automated delivery:

1. Difficulty in word choice. Although the total number of keywords in ASM is much less than that in SEM, it is also in the millions. You need to choose based on multiple dimensions such as popularity. It is not a good idea to choose too many options, as this may cause the quality of your account to decline.

2 Keyword management is difficult. An account may have dozens or thousands of keywords, and each word has its own characteristics, such as "train tickets", which is popular during the holiday season. Managing so many keywords is like operating so many stocks at the same time.

3. Bidding is difficult. There are multiple users bidding for each word, and the competition strategy is complex.

4 Unified optimization is difficult. How to optimize multiple keywords uniformly? Which words can be budgeted less, which words can be budgeted more, and so on.

5. It is difficult to define user portraits . Keywords and user portraits jointly determine the click-through rate. However, data is difficult to obtain, and only by combining different words with appropriate user portraits can good results be achieved.

6Cold start problem. ASM lacks historical deployment experience and often requires a lot of attempts before there will be consumption.

In response to these problems, there are also some corresponding solutions in the AI ​​field:

1. Learn from expert experience. We mainly learn rules from SEM experts. For example, we found some experts who were good at advertising on Baidu Fengchao and Google Play to do ASM delivery. We gave them some guidance and suggestions, and then summarized them into some delivery rules for machines to use.

2 AI autonomous learning. That is, the machine continuously and automatically improves itself from historical data.

3. Big data support. All kinds of ASO data can be used in ASM delivery.

4. Transfer learning. Mainly the use of various SEM algorithms. For problems like cold start, bidding game, and unified optimization, there are some available algorithms in the SEM field that can all be applied to ASM.

The premise of AI application is that it must be supported by big data and, secondly, have some technical accumulation. ASM basically meets these requirements. Our APPBK team has been working on ASO tools before, so we have a good data foundation and have previously worked in search advertising research and development in large companies, so we naturally chose the ASM direction. In addition, some companies that previously worked on SEM technology are also paying attention to this field.

Service Selection

For CP and operation staff, the choice of ASM service mainly depends on demand and cost performance. Here are some simple suggestions:

Small and medium-sized APPs: Choose some SAAS services and auxiliary delivery tools and deliver them independently.

Large APP: Choose high-quality agent services. You can also build your own system or purchase a customized PAAS system. The service fee abroad is generally 10%-15% of the advertising cost.

Agents: Build their own delivery systems or cooperate with technology companies.

Mobile application product promotion service: APP promotion service Qinggua Media advertising

The author of this article @王亮 is compiled and published by (Qinggua Media). Please indicate the author information and source when reprinting! Site Map