Part of the communication challenges between data scientists and the business result from thinking one methodology is going to solve two problems. Illustrative example: The biz asks for a highly predictive churn model (this could be extended to many different use cases, but we're keeping it simple here). In reality, the biz wants to be able to:
1. Accurately identify customers with a high risk of churn so that they can implement some type of corrective measures.
2. They also want recommendations (based on data) that will inform what corrective measures could potentially have the biggest impact on reducing churn. To give the biz what they're expecting, it's possible that you'll need to build two separate models. (one that is highly predictive, the other that is easily interpretable). Bonus, once you've already collected the data, it's not that much incremental effort to build multiple models.
Agree or Disagree? And if you agree, are you already approaching things this way?

❇️ @AI_Python_EN