Cross Sell Modelling Data Preparation

predictive_model
data_science
finance

#1

I was preparing Data for Crpss Sell Top Up model for a Personal Loan Lending Use case.
I have to predict each month which customers have high chances of being converted (intend to buy the cross sell product).

I have data with respect to every campaign month.
Suppose in the month of Jan, 100 customers(existing) were campaigned out of which 5 wanted to buy the new product (or bought a new product). In the month of Feb 95 which did not buy in Jan and 15 new are included in the campaign and say 7 of 115 buy the new product.

If I prepare data stacked on 6 months of Campaign one customer might appear more than once.
Now I was thinking of rolling up the data at Customer level across the campaign months I have selected for model(binary classification algorithm) training. Rolling up meaning aggregating the transaction history and etc. This I think needs to be done to consider an unique customer scenario.

I have already read this blog: https://www.analyticsvidhya.com/blog/2015/08/learn-cross-selling-upselling/
Please help me understand do I have an ambiguity in the data preparation I will be doing.

Thanks in advance.


#2

I myself have figured it out. Since month -on- month same customer will be appearing most of the time.
Therefore it would be better to build the model on a single month of recent campaign data.
The month of campaign to be chosen for training and test would be decided in accordance to business supported by some analysis w.r.t the top up conversion rate , proportion of customer repeating etc (customers who are once converted for top up will not be campaigned for the next few months).
In addition to that 6 months of recent transaction history can be considered to create transaction history related variables along with demographic variables.


#3

Thanks for taking the time out and replying to the community!