What is - Data-driven attribution

What is - Data-driven attribution - and how it works

Let's imagine that you need to track which keywords played the biggest role in conversion or which channels worked best in an advertising campaign. This will allow you to understand Data-driven attribution - the most advanced model for today.

Why is she unique

Brands have a lot of opportunities to interact with customers. It is important to know which are effective and which are useless. Compared to traditional models, data-driven attribution can increase Google Ads search campaign conversions by an average of 5% while maintaining acquisition costs.

Let’s compare this model with the most popular one – last-click attribution. In contrast, Data-Driven will help:

work effectively with channels that are considered less effective in the last click model - display, video, mobile traffic;
carefully and carefully assess the contribution of remarketing, which consistently shows a high result in the model for the latter.

Now let's create our model as an example

We are a household appliances store. The Google Ads Search Attribution report shows a user's 10-step journey to purchase. There are all kinds of keywords: general non-brand search (“household appliances store”), product names (“refrigerator”, “buy refrigerator”), product plus brand (“Morozko refrigerator”), store name. Users went through this chain and bought something.

For simplicity, let's imagine that 100 users went through this path, and 25 of them bought the product (converted). The conversion rate will be 25%.

Step 1. Calculate the conversion rate (CVR)

For now, we will not assign different values to individual clicks but will calculate using a linear model that all words have an equal contribution to the final conversion rate. That is as if all the keywords had the same effect on the fact that 25% of users bought refrigerators.

Step 2. We take the same chain without one keyword and compare CVR - evaluate the contribution of this word

The chain is the same, but without the word "refrigerator to the country". In the sequence of clicks, it was in the fifth position. It turns out that the conversion rate of such a path is already 10%.

That is, the absence of a keyword from the previous chain reduced the conversion rate from 25 to 10%. The contribution of this word is the difference in the conversion rates of the two chains.

Step 3. Determine the weight of this keyword and adjust the value distribution

In total, we see:

Chain without the word "refrigerator to the cottage" - 10%.
A chain with this word is 25%.

That is, the word increases the conversion rate of the path by 2.5 times (by 150%). We increase the value of this word but leave the rest unchanged.

Step 4. Tens of thousands of times repeat this procedure for each chain and each keyword

Data-driven attribution uses machine learning, so we can compare the chain many times. If a keyword increases its conversion rate, we increase its weight, no matter what position it appears in. At the same time, we do not reduce the weight of those words that do not positively affect the conversion rate.

Step 5. We get the final result: the conversion weight of each word

We sum up the values of all words and divide the individual weight of the keyword by a common denominator - the sum of the weights of all keywords.

When the simulation is finished, we get a weighted ranking for each keyword: the percentage of weight that it takes for itself.

The weight is redistributed among all participants in the conversion chains. Such a model is not subjective, because it takes into account all the data about interactions with users. And most importantly, all calculations are done automatically.