Analysis of the operational process of Tik Tok advertising promotion model!

How are advertisements for products searched on JD.com displayed on Douyin? Could it be that personal data has been leaked? In fact, this is just a simple operation of programmatic advertising. In this article, the author will analyze the operating process of this advertising model.

A few days ago, I felt that brushing my teeth was too strenuous and I couldn’t brush them cleanly, so I opened JD.com, searched for electric toothbrushes, and then completed a series of purchase and payment operations in one go.

The weather in Shenzhen has been weird recently. It rains every day. The wipers, which I haven't changed for a year, make a hissing noise but still can't clean the windshield cleanly. Well, you guessed it right. Then I opened JD.com again and bought a lot of things, only to find that I spent two hundred and fifty yuan.

After buying it, I happily watched Douyin. Well, the girl is so pretty...

Suddenly, such advertisements appeared. Before, all I saw were crayfish, games, various apps, etc. These two are really impressive, I have never seen them before. It appeared after I bought something. It was the first time that it appeared in my Tik Tok girl.

Is the privacy leaked? Was my data sold by JD.com?

In fact, this is just the simplest operation of programmatic advertising.

Let's first simulate the process with a simple conversation:

Taking this ad as an example, the entire process has the following dialogue:

Me: Open TikTok

Douyin: (This idiot is here again) holding a loudspeaker and starting to shout. I have an ad space for rent. The person viewing the ad is a fool. His ID is xxxxx. He is 18 years old and very handsome...

The content of the call was transmitted, and A standing nearby heard it, B heard it, C heard it, and JD also heard it. Each of them should look through their notes to see if there is any record of this idiot.

A: I think this guy is average, I’ll give him 20 cents.

B: I’ll pay 30 cents.

C: I don’t know him. Goodbye.

JD.com: Isn’t this the fool who just searched for electric toothbrush here? I'll give him 1 dollar to promote his advertisement.

TikTok: OK, you offered the highest bid, so the position is given to you, and you can take the content.

JD.com: OK, I’ll find an electric toothbrush advertisement for you.

Me: Damn, I just bought a toothbrush and now I’m promoting this thing!

This is roughly the process, of course the above process is not the only path.

Key technologies

To understand the above process, let us first introduce the key technologies used in this delivery:

Programmatic advertising : refers to the use of technical means to conduct advertising transactions and management.

Advertisers can programmatically purchase media resources and use algorithms and technologies to automatically achieve precise target audience targeting, delivering ads only to the right people. The media can programmatically sell media resources across media and terminals (computers, mobile phones, tablets, Internet TVs, etc.), and use technology to stratify advertising traffic and perform differentiated pricing (for example, prices in first-tier cities are higher than those in second- and third-tier cities, and prices during prime time are higher than those in other time periods).

SSP : Supply-side platform, provides traffic. Manage media advertising space, optimize advertising delivery, and help traffic parties increase revenue.

DSP : Demand-side platform, which provides advertising delivery services to advertisers and agencies to accurately reach targeted users.

ADX : A real-time advertising trading platform. The advertising bidding process is completed within ADX.

DMP : Data management platform, which integrates scattered data from multiple parties into a unified technology platform, and standardizes and segments these data, allowing users to push these segmentation results to the platform in the existing interactive marketing environment.

RTB : Real-time bidding is a bidding technology that uses third-party technology to evaluate and bid for each user display behavior on millions of websites or mobile devices. Unlike buying a large number of delivery frequencies, real-time bidding avoids reaching invalid audiences and makes purchases for meaningful users.

E-commerce advertising is generally measured by effectiveness, which is the safest way for the demand side. That is to say, CPA, CPS, and CPM are usually chosen, but the specific choice depends on Douyin’s confidence in its push algorithm and the quality of the pushed users.

Several charging methods for current advertisements are as follows:

1. CPA (Cost Per Action) Cost per Action: CPA is a type of advertising that is priced based on the actual effectiveness of advertising - that is, it is charged based on valid questionnaires or registrations responded to, without limiting the amount of advertising delivered.
2. CPS (Cost Per Sales): The advertising amount is calculated based on the actual number of products sold.
3. CPM (Cost Per Mille) Cost per thousand: CPM is a display payment advertising, paid by the number of impressions
4. CPT (Cost Per Time) Cost per time period: CPT is a type of advertising that is charged by time, and ads are displayed in fixed ad slots within a fixed time.
5. CPC (Cost Per Click) Cost per click: CPC is a pay-per-click advertising, charging according to the number of times the ad is clicked.

I assume that this ad was purchased by JD.com using the RTB bidding rules.

Process Disassembly

Let’s break down the products I searched for on JD.com and see how the ads were displayed on TikTok.

1. data

Let’s start with the data:

As a user of Douyin and JD.com, I have corresponding user portraits on Douyin and JD.com respectively.

Of course, the user portraits of different platforms will inevitably collect different specific information and have different focuses. For example: Tik Tok generates a user portrait based on the users I follow, the video content I browse, the completeness of the videos I browse, and some basic information I fill in when registering.

As an e-commerce platform, JD.com must be more concerned about the portraits generated by my purchasing habits, favorite products, searched products, categories of interest, and so on.

But in general, the generation of portraits can generally be summarized as the following formula:

Portrait = user basic information + user interests + user behavior + related information

User information : including registration information, geographic location, age, gender, etc.

User interests : followed stores, frequently browsed categories, and favorited items.

User behavior : behavior records within the traffic (here should be the JD APP client), such as search actions, purchases, browsing time, search frequency, etc.

Related information : derived information obtained from some label attribution analysis.

The records generated by my searches on JD.com are user behavior and interest information, which are stored in the database in real time as a very important part of the user portrait. Action is taken only when there is a need. Unlike information collected passively, it is more urgent and intense. In the portrait system, the weight of this label over a period of time is undoubtedly very high.

1. Delivery

Where there is advertising content, there must be delivery.

There are two situations: one is that JD’s own delivery platform (i.e. DSP) is connected to Toutiao’s ADX; the other is that JD’s marketers deliver directly on Toutiao’s marketing platform.

Because this ad is associated with the data generated by my real-time search, we exclude the second case. JD.com should be connected to Toutiao's ADX, so that it can extract my tags from JD.com's database in real time as matching criteria.

1. User Engagement

After the preliminary work is done, it is time for users to participate.

When we browse videos on TikTok, it will display advertisements at intervals. We previously assumed that the advertisements were RTB bidding advertisements. When I see this ad space, Douyin sends messages to N platforms at the same time through ADX, informing them that there is an ad space available for sale and inviting each DSP to participate in the bidding.

When sending a message, it carries user information, which has a unique identification number (assuming it is a mobile phone number). When JD.com received the message, it matched the unique identification number sent by Douyin with the identification number in its own DMP database and found that the user existed and had recently searched for a certain product.

Pay a high price to buy the ad space and eventually win the right to display the ad in that ad space. Then, the product information advertising content that needs to be delivered is returned to Douyin.

What is written here is relatively simple. In fact, there are many rules in the internal logic. Of course, these are all completed at the millisecond level.

After that, the user sees the advertisement content and curses: XX!

The above is an introduction to the entire process. However, after reading the whole article, you should find that: In fact, I not only searched for this product, I actually completed the purchase.

The two products I purchased are items that are used very infrequently in daily life. I have already bought them and generally will not buy them again in the short term, unless JD.com wants me to return the products and then make a choice, which is of course impossible. According to the RTB bidding logic, JD.com is the one that loses in this wave.

From the previous introduction, we know that the decision basis for JD.com’s push is my search behavior in the APP. Active search means a strong intention to buy. At this time, the proportion of labels used for behavior will be higher than other portrait labels. So, when JD received the bidding request for the Douyin ad space, it pushed the ad to me.

What's the problem?

I guess it may be due to the huge number of users and data, and the search records are still in the cache and are not associated with the purchase records. Or maybe JD.com simply didn’t do any frequency marking on the categories of its products, and didn’t make any estimates on the user’s next purchase expectations.

Based on the above, I suggest: classify the consumption frequency of goods, integrate and update the portraits in a timely manner, and output reliable decision-making opinions.

Related reading:

1. Tik Tok promotion and monetization operation tutorial!

2.2019 Tik Tok promotion and operation strategy!

3. The four major matrices of Douyin promotion and operation!

4. What are the necessary factors for Douyin promotion?

5. Sharing of practical tips on Douyin promotion and operation!

6. Douyin promotion method: 6 types of content that Douyin limits traffic!

Author: Qinggu

Source: Qinggu