Today most e-mail marketers have started using predictive modeling to better target customers for particular products/services.. It is because predictive modeling in marketing activities can deliver data-driven predictions that you can use to influence the future.
In e-mail marketing, the cost incurred for every thousand potential customers who view the advertisement (CPM) is $7-8 or less. Many marketers believe that it is not worth building a predictive marketing model as the savings from a predictive model is less or negligible considering it takes time and money to hire predictive modelers for model building. Rather, predictive modeling should be used in Direct Mail where the CPM is $500- 600 or more.
1. The cost of acquiring a permission based subscriber e-mail address is high (between $10 and $40).
Traditional View of Marketing
Mail to all the customers. Someone is going to buy.Modern View of Marketing
Mail to only those customers who are interested and more likely to buy your product.
Spam affects Email Marketing Campaign!
Spam is commercial email, junk mail or bulk mail that has not been requested by the recipient. A relevant email to a top customer can get lost in the spam if you mail him too frequent and he can also label you a "spammer".
Permission-based emails is a latest weapon to avoid being labeled a spammer!Permission-based emails are requested, anticipated, personal, relevant and verified by user.
Reasons to use Predictive Modeling for Email Marketing
2. A relevant email to a valued customer can get lost in the spam if you mail too frequent. Many users unsubscribe or delete e-mails without reading them and can also label you a spammer and unsubscribe for future email updates.
3. Unsubscribe rates are often 2-3% or more per month. If the value of each subscriber is $15- $25 and the number of subscribers are more than 500k, the annual loss from unsubscribers can reach up to millions of dollars.
4. It is not required to build predictive model every time you send an email campaign. You can build a model once and perform scoring from the next time. Note - It is important to check the stability of the model once in a quarter by comparing the current scoring with the predicted probability from training data set.