Propensity of customer to buy a new product?

Question

I am tasked with deciding whether a new product should be developed. I have data of customer transactions of previous products and wish to build a model which determines whether customers will have a higher propensity to buy this new product.

My approaches in mind:

1. Look at the average daily sales for each product, then find the closest product (using Euclidean distance) to the new product based on a set of covariates. If the closest product isn't the highest-grossing, then decide not to launch.

2. Build a classification model based on the transactional data and the covariates of each product. For each transaction a customer makes, augment the dataset by assigning 0 to the products that wasn't bought by the customer. The model thus works out the predicted probability that each customer buys each product at a given point in time. Then use as input a series of times, and the covariates of the new product to work out the average propensity score for the new product.

How plausible is such an approach? Is there a more statistical/data science approach to answering this question?

Answer 1

The first approach is simple and may be useful, but it is sensitive to your metric choice (e.g. you have chosen Euclidean) and seasonality. It's better to first validate this approach for your previously released products.

For the second approach, I think it's again necessary to validate and see how accurate you are with existing products.

For both, You need to respect the time/temporal order in validation procedure.