I have come across a tricky question. How do I ensure that all the adaptive models get refreshed in similar frequency irrespective of how frequently the offers are presented to the customer.
For example, I have an offer A that has high demand and would be presented to the customer almost daily. Based on the reponses recorded we refresh the model.Similary there is an offer B that gets presented to the customer once in a month and hence this model doesnt get refreshed.
The issue is, adaptive model perfromance is erratic. How to train these models so that they are in tune with other models in terms of performance?
Is there any way to switch the adaptive score based on the frequency of the offer?
The question is very generic and it is guite difficult to give a straight answer. It depends on many things. like for e.g.
1. Are propositon A and B of similar type? Do you expect them to be offered with similar frequency?
2. Are proposition A and B introduced on the market at different moment in times? Did A see many more responses before B was introduced.
3. Do you use directly the model proensitry when prioritizing, or do you use any marketing weight or margin calculation in your prioritization.
Just few ideas:
If both offers A and B were introduced together, then you could use the smooth proensity to ensure that the innitial evidence is taken into account.
If proposition B is introduced at a later stage, again you could use a smoot propensity for proposition B, with a higher starting propensity, to ensure that proposition B does get a chance to be offered and therefore responses.
If B is a 'special' product and has different characteristics then A, then it is not necessary bad that B gets less fewquently offered. You can try to boost proposition B (for e.g. using a marketing weight calculation) in the beginning until you train the model a bit.