Prediction tree

How to read it

The color red represents a high probability of purchase (the more intense the red is the higher the probability) The blue color represents a high probability of no purchase (the darker the blue is, the less likely users are to buy)

The first point on the left represents 100% of the population, a first split is made on the variable total_order_amount.

On the one hand, those who have bought in total (over their entire purchase history) for less than 25€, have very little chance of buying (very dark blue); on the other hand, those who have bought for more than 25€, they have a little more chance of buying (light blue) Among those who have recently been in the funnel have a good chance of buying (light red circle), and among them, those whose first visit date is less than 28 days, have a higher probability of buying even more (very dark red circle).

And for those whose first visit is more than 28 days, we see that those who have seen less than 22 pages recently are unlikely to buy, except for those who have a total page view history of less than 55 and who have recently seen more than 4 pages.

Unlike those who have a total page view history of more than 55 and who have recently seen less than 8 pages, who will not buy.

Etc., etc. by following all nodes.

Lessons you learn from it

1. The most predictive variable is the total amount of purchases (total_order_amount), I have to take it into account when I create my segments

2. The recent presence in the purchasing tunnel significantly changes the deal (not surprisingly, hence the interest in making the relaunch abandon baskets)

3. The volume of pages viewed is an indicator of the probability of purchase and depends on the date of first visit (it is deduced that there is a kind of ratio that allows to deduce an intentionist, according to his date of first visit, his total number of pages viewed ever and his number of pages viewed recently) So I have to create as many segments as there are red nodes on the far right of the screen to be able to find all the visitors who are intentionalists (or dig this ratio story to create a new score variable to facilitate the rest)

4. Among the other predictive variables to keep in mind are the recent_view_product and recent_view_category, I see that the higher they are, the less chance there is of buying, probably because they are users who shop around without really knowing what they want, unlike those who see few products and categories and who are more likely to buy quickly