From a data scientist on Medium: “It’s easy to not understand your customer’s needs.”


So a guy walks into a bar and starts talking about unsupervised learning...

Ok, not quite. Well actually, where I live in Cambridge, MA, that's not really so improbable 😉

I found this article on Medium interesting, written by a data scientist talking in part about why data science projects may not be working, and one of them was "solving the wrong problem." He had come up with a new tool/method to solve an old problem, and it didn't go over well with the users, and so he concluded:

...It’s easy to not understand your customer’s needs. Pick your favorite large company: you can find a project they spent hundreds of millions of dollars on only to find out nobody wanted. Flops happens and they are totally normal. Flops will happen to you and it’s okay! You can’t avoid them, so accept them and let them happy early and often. The more quickly you can pivot from a flop, the less of a problem they’ll be...."

Dr. Jonathan Nolis, Data Scientist

I'm not writing this insight today to pick on Jonathan, after all, it looks like maybe he's based in AZ (my birth state) and hey: we're a friendly bunch! However, that first sentence (my bolding) isn't really accurate, and this is almost certainly part of what contributes to the 85% failure rates on data/analytics projects. No clear idea what the users need/want and/or vague business objectives. That said, as stated in my comment on Medium to Jonathan, I would agree that this not the primary responsibility of the data scientist.

If you're a biz stakeholder, product manager, or analytics leader, then you should be giving your product development team much clearer objectives. It is not difficult to understand customers needs; you just need to regularly go out and talk to them. While there is definitely skill involved in conducting great research and extracting business objectives from stakeholders, you don't need heavy training to get started. You may not discover every latent need or problem to solve, but it's definitely better than not talking to them at all, or taking wild guesses. After all, at some point, too many "flops" start to add up financially and otherwise.

I understand that with certain types of machine learning, there is inherent ambiguity in what might come out the tailpipe. However, you can spend a little more time, and probably a lot less money, doing a little research before committing the resources to implementation of something that may have zero value to anyone.

Here's to a few fewer flops, even if we can't stop them all!