Resolved: Just a consideration about the marketing part
Good morning. I just want to ask this thing: you said that we should spend more money with people who are more likely to buy again a book. But if a person is really likely to come back, why should we spend money with that person? Let's suppose for example that we don't want to advertise people with less of 50% of probability to come back. If customer 001 has a 51% of probability to come back, we could still forget about him, but if a person has a 99% of probability to come back, shouldn't we forget about him, too? Why should I spend money advertising a person that almost surely will come back even without the advertise? So I was wondering if maybe we should spend our money just with people in a middle way, for example between 70% and 90%, while we could forget about the other people. Thanks
Hi,
Marketing sometimes can be hard and really subjective :D
In principle, you are correct in your logic - if a person is almost guaranteed to come back, we don't need to spend extra money on them. However, unfortunately, the real world is a bit more messy than that.
First of all, our model predictions are just that - predictions. They won't be correct all the time and someone with 95% may turn out to miss out on the next product. In that sense, it may be worth it to spend $1 of marketing on that person in order to guarantee that he buys the $50 product (all of these prices are just examples). Sure, in most cases that person would buy the product even without any ads, but by investing a little, you may further "guarantee" them as a customer.
Another thing to consider is the training data. We may have data with people's behavior, but there is nothing in there on whether the person was targeted with ads or not. Depending on the specific situation, it may be the case that most of the users in the dataset were targeted. And so, the results are valid only for people that saw the ad. In other words, if we don't spend marketing money on them, their probability of coming back would be smaller than what the model predicts. A person may have 95% of coming back precisely because they were targeted.
Thirdly, if we don't spend marketing on them, maybe a competitor would.
All of the above discussions may or may not apply in certain scenarios. It really depends on the specifics of the situation. Your suggestion would also be valid in many situations.
Hope this helps!
Best,
Nikola, 365 Team
Really exhaustive answer, thank you!
