Ask HN: What do you do when your client asks for something stupid?

I work for a startup that sells consulting services on the side (software and data science). Sometimes I am asked by clients to do something stupid. E.g. running a clustering algorithm on 2D data where there obviously are no clusters (but hey the new segmentation is the result of a machine learning algorithm and the client likes it), or building performance metrics where we clearly have no relevant volume of data. How do you deal with that ?