Posted: 23 Apr 2020 11:07 EDT Last activity: 3 Nov 2020 0:11 EST
Switching from Randomization(Champion Challenger) to Adaptive Model
For the below requirement we are looking for some inputs. Appreciate any response
Phase 1: Build a Decision Strategy having multiple offers and use a champion challenger component for randomization. Setup/Run a multi-channel campaign and refer the decision strategy having champion challenger component for a few months.
Phase 2: Build a new Adaptive Model. Update the Decision Strategy created in Phase 1 to refer to the Adaptive Model(by replacing the Champion Challenger component).
Note: Email Offers - Pega would be sending offer data to an external system at the end of campaign run and the external system is going to trigger the offer emails to the customers. Later the external system is expected to share the Customer Response data with Pega(Clicked,Accepted etc).
1. For the offers generated during Phase 1, is it possible to load the customer response data back in Pega? since there is no Adaptive Model in place.
2. Once we switch to Adaptive Model, can the model take it to account all the customer responses received for offers generated during Phase 1?
Between Phase 1 and Phase 2, the offers would remain same.
If you want to use the data and responses you should include an adaptive model in phase 1 *before* whatever decisioning you apply. So you do run the decisions through adaptive so it can learn, but you would not use its propensities. Of course the feedback mechanism needs to be set up as well in order for the models to learn. Only in such a config will your models be able to learn from phase 1 data. Well - you could also collect all customer data + responses in a data lake and bulk train the adaptive models later, but this is typically quite a hassle to get straight.
I would not use a champion challenger in phase 1 - you would need to explicitly configure all propositions/offers as inputs and manually set all the percentages. Why not just feed them into a prioritization component and order by @random() ?
Thanks Otto for your response. Please find below my comments
1. So your recommendation is to use Adaptive Model as a place holder in strategy during Phase 1 and bulk load the customer responses through model management landing page(since in phase 1 feedback loop won't be in place). With this setup, the model would keep learning during Phase 1 and when business is ready to utilize the model's learning, then update the strategy to select the offers based on Model's Propensities(in Phase 2)
2. Regarding your question on Champion Challenger - During Phase 1, business wants equal distribution of offers (We will be having less than 10 offers to start with). Hence we are planning to use Champion Challenger and assign a fixed % to each of the offers.
Posted: 2 months ago
Updated: 2 months ago
Posted: 2 Nov 2020 10:20 EST Updated: 3 Nov 2020 0:11 EST