(00:05):
This is Five-Minute Friday on Identifying Commercial Machine Learning Problems.
(00:27):
Five-Minute Friday this week is the first in a three-part series on strategies for getting business value out of machine learning. This week, I’ll be covering my first strategy, which is being confident that there’s a commercial problem to solve before starting data collection or ML model development.
(00:45):
This might sound like an obvious tip, but in practice ML practitioners often go about this the wrong way around. We often get excited about some new machine learning technique and then we seek out a nail to hit with our shiny new hammer. Go ahead and do that for fun, but if you’re serious about finding a compelling commercial problem to solve with ML then you’re better focusing on identifying a problem to solve first. Then, after the problem is identified, start thinking about what data modeling approach might be effective for solving it.
(01:15):
Ok, so then how do we go about identifying a commercial problem to solve? First, you can reflect yourself. Is there some process or platform you use in your personal or professional life that could be automated? An opportunity might not present itself immediately, but sit with this question for days or even weeks. You could even set a reminder in your phone to tell you to write this question out on a piece of paper each morning; again the question’s roughly: “Could something here by automated with data?” Reflect, reflect, reflect. And by writing that out every morning, you’ll hopefully queue the question more often and then sitting on the train, in the shower, or bored in a meeting, an idea could pop into your head.
(01:56):
Personally, most of my ideas for improving some process or platform with data or automation don’t come from me. They come from others, such as the users of a software platform I’ve built. As the chief data scientist at a machine learning company, my primary remit is to manage the data scientists on my team as they carry out research and development and then to ensure that the models we devise make it effectively into production. However, I wear lots of other hats. One of those is holding occasional meetings with power users to get their thoughts on what’s missing in the platform or what could be better. Since we have a data- and ML-driven platform, many of these ideas from power users can be thought about or prototyped by our data science team.
(02:35):
In addition to existing power users, during sales pitches prospective clients can provide a wealth of insight into functionality that could be missing in your platform or product as it is today. Prospective clients will often deliberately contrast you with your competitors, which can generate direct or indirect ideas as to novel ML-based functionality that your product could potentially have.
(02:56):
All right, so that’s this week’s strategy for getting business value out of ML. Focus on identifying a commercial problem to be solved as opposed to looking for a nail to hammer with some shiny new ML-technique hammer you’ve recently learned about. To identify these commercial problems, you can reflect yourself, talk to power users, or talk to prospective clients.
(03:17):
For Five-Minute Friday next week, we’ll have part two of this three-part series on strategies for getting business value out of ML, in which we’ll discuss data collection as a stepping stone to devising a model to solve the commercial problem that has been now identified. So stay tuned for that!
(03:33):
In the meantime, keep on rockin’ it out there, folks, and I’m looking forward to enjoying another round of the SuperDataScience podcast with you very soon.