It often goes wrong when an algorithm is set up too simply, for example when it only looks at preferred products. Then you will see that the algorithm recommends slippers again after purchasing a certain type of slippers. However, customers do not buy two pairs of slippers in a row very quickly. Let the algorithm look mainly at the purchase that occurs more often after the previous purchase of a specific product. Train the algorithm on that. In this way you predict as accurately as possible which offer increases the chance of a new purchase.
Of course, the possibilities are endless. For example, you can also think of predicting the needs in mobile telephony, green energy or magazines in subscription form.
More than ready
Once you have completed these steps, you are ready to run the first campaigns. We now know who we want to approach and who we would rather not. We approach these customers with a personal offer or via a personal action. Always keep a control group separate, so that you can measure the effect. After the campaigns have been completed, you look at what it has yielded. What is the result compared to the control group? If you execute correctly, you will see that there are huge differences in performance between the campaign and control group.
Saves modeling
Unfortunately, it doesn't always go perfectly. Sometimes you predict incorrectly or a customer is not as easy to influence as you thought. Or the competitor runs an unforeseen campaign that is not included in the probability calculation. That's why I added a final model to this strategy model: the saves model.
Unfortunately, a number of customers have laos telegram data churned. We will make one last attempt to retain them. For these customers, we have set up the saves campaigns using saves/winback modelling. Within saves or winback campaigns, customers who have recently switched are approached one last time to win them back as customers.
Many companies have set up telemarketing campaigns for the saves/winback campaigns, because this generally yields the highest conversion. Telemarketing campaigns are very expensive, so achieving a good result is of great importance. Precisely because these campaigns are expensive, we only want to approach the valuable customers. For this, we use our previous results from the CLV model. In addition, we want to predict the chance that someone will be saved if we approach him or her. Here too, we learn from patterns in data that can be seen with customers who have been saved in the past and customers who have not been saved in a campaign. With the predictions, we know who has the most chance of accepting a saves offer. We only approach the promising people. This allows you to retain the same number of customers or even more customers in the saves campaigns with the same budget. And you don't have to call 10,000 customers, but 4,000 valuable customers who you know have a good chance of returning.