How to understand advertising attribution performance?

There is a word in advertising that many people don't understand, that is "advertising attribution". Advertising can understand it, but what does attribution mean?

The origin of attribution theory: Heider first wrote about attribution theory in his book "The Psychology of Interpersonal Relationships". The causal explanation and inference of the behavior process of others or oneself through control, and advertising attribution is who ultimately owns the advertising results! When it comes to this, there may still be some people who don't quite understand it. Let's use an example to illustrate it, and it's easy to understand!

When it comes to advertising attribution, there is a classic example: in a football game, player A receives an assist from player B and scores a goal. So how should the credit for this score be distributed? At this time, A said, of course, it should be all my credit. B said, can you shoot easily without my assists. At this time, the cafeteria aunt also jumped out and said, I usually cook nutritious meals for you to ensure your high-intensity training, so I also have a share of the credit.


Advertising attribution is a very important part of every advertising platform, especially effect-oriented ads. Advertisers place ads on the advertising platform, if they cannot see the conversion effect brought by any ads. Then the advertiser's ad delivery is unsustainable. It's like a boy chasing a girl, sending flowers and gifts, and inviting a meal to a movie, but the girl doesn't respond. In this case, most boys will start to back off.

Entering the topic, advertising can be roughly divided into brand advertising and performance advertising. The purpose of brand advertising is mainly to pursue a large amount of exposure. However, performance advertising will pay more attention to ROI, which is the return on investment ratio. How many conversions an advertiser has brought about when an advertisement is placed (the definition of conversion can include placing an order, adding a shopping cart, collecting products, and even likes and comments in social advertisements), and the advertiser will evaluate it according to the conversions brought. What is the effect of this ad placement, and make corresponding decisions. So for advertising platforms, how do they know which advertisement brought an order or a like? To sum it up in one sentence: find the most relevant ad impressions or ad clicks for each conversion.

At present, major advertising platforms, such as Facebook, Google, Ali, Jingdong, etc., mainly use the Last Click model (Facebook can allow display and click attribution, and Google also provides advertisers with rich attribution models, such as First Click , Last Click and other attribution models), so the following only takes the Last Click model as an example.

The scene is as follows, Xiao Ming is going to buy a basketball on Taobao, so he searches and browses a lot of basketball products on Taobao, and finally settles on a certain basketball and places an order to pay. At the same time when Xiao Ming placed the order, the attribution system of Taobao Express also obtained this order record and the relevant information of the user who placed the order. Generally, there are the following steps:

1. Order information and advertisement click information

The order information generally includes the user account name (replaced by PIN below), the order SKU, and the order time.

The advertisement click information generally refers to the click log, and the click log contains a lot of information, such as the device ID of the clicked advertisement, the clicked advertisement time, the advertisement delivery information corresponding to the clicked advertisement, and so on.

In addition, there will be the following key-value pairs in the Taobao Express database. The key is generally the user's PIN, and the value is the unique identifier of each device type that this PIN can be associated with, such as IDFA for IOS devices and IMEI for Android devices. , cookie information on your browser, etc.

2. Information association

Taobao will find out whether the PIN has clicked advertisements within a fixed time window (such as the last 15 days) according to the PIN of this order.

Suppose Xiaoming's order information is: {PIN: xiaoming001, order SKU: 12345, order time: 2017/7/7 14:59}

and the key-value pairs of Xiaoming's various device IDs stored in the through train are: {key = xiaoming001,value = (IDFA,111),(IMEI,222),(cookie,333)}.

Next, the system will refer to the key-value pair for the PIN placed in the order to find the advertisement clicks of the PIN under all device types.

3. Relevance determination Orders in

Taobao Express are divided into direct orders and indirect orders. Direct orders refer to seeing an advertisement of a certain SKU, clicking on it, and finally purchasing this SKU. Then the system will judge it as a direct order. If you see and click on the advertisement of a certain SKU, but do not buy the SKU in the end, but buy other SKUs under the advertiser's store, then the order will be judged as an indirect order.

Suppose the following two ad clicks are found in step 2:

1. click_id = 1, sku = 12345, and the click time is 2017/7/7 14:57

2. click_id = 1, sku = 23456, and the click time is 2017/ 7/7 14:58 (sku23456 and sku12345 belong to the same store)

The first click is the direct order mentioned above, and the second click is the indirect order mentioned above. However, due to the arbitration logic of orders, an order can only be attributed to the most relevant click. Obviously, the priority of direct orders will be higher than indirect orders, so the system will attribute Xiaoming's order to the first click.

The above examples are based on advertisements on Taobao, so there is no need for complex cross-platform conversion tracking and monitoring.

After reading Last Click's attribution logic, go back to the football example at the beginning. According to Last Click's decision logic, this goal should be 100% attributable to Player A. But obviously, Athlete B also made a very important contribution to this goal. Therefore, the Last Click model has its shortcomings. But if the Last Click model is not used, what percentage of the credit we should attribute to Athlete B is obviously not easy to judge. Therefore, most advertising platforms choose the Last Click model.
This article comes from the Jiajiao Advertising Alliance! Original link: http://www.jiazhua.com/wz/1125.html