Introduction
What do microwaves and the use of machine learning for businesses have in common? Well superficially not a lot – but they make a great analogy for the reason that businesses fail at using machine learning. this analogy as well as the reasons for this are outlined really well in the article Why businesses fail at machine learning by Cassie Kozyrkov (the article can be found here).
Summary of Article
In the article, Kozyrkov talks about the 4 main reasons that she believes that businesses fail at adopting machine learning and having it help their business. They are as follows:
- Not having a clear business goal
- Not having the right data
- Not having the right skills
- Not having the right culture
The first reason is the one that Kozyrkov believes causes most businesses to fail – not having the right business goal. Businesses just dive into machine learning because it is a hot topic right now, before evaluating whether or not they actually have a use case for it. Make sure that you have a clear use case and business goal for machine learning before diving into it.
Next, not having the right data – garbage in, garbage out. Make sure that your data is clean and appropriate for the problem at hand. If you don’t know whether your data is good ask for help from someone who is well-versed in this field. But make sure that your data is good in some way or another, otherwise your model won’t be useful, because as they say: garbage in, garbage out.
Another factor is not having the right skills. Machine learning is a complex field, and learning all the skills necessary to start using it for your business can be hard. So, if you don’t have the right skills, outsource it and hire someone that does or learns the skills necessary, but ensure that whoever is setting up, and running your model, is well-versed in this field.
Finally, not having the right culture for machine learning can also be a factor that causes some businesses to fail at using machine learning. Make sure your business has a culture that fosters experimentation and improvement. Machine learning takes time to get right for your specific business, so you need to be patient. The model will need to train on some data, improve, then test, and improve again, and all of this can take time. So, foster a business culture of experimentation and improvement and understand that your insights won’t come right away.
My Take
Overall, I really enjoyed the article. It was interesting and insightful. I also really enjoyed the analogy to a kitchen and restraint throughout the whole article. This analogy was a great one and really helped me understand the concepts better. It was also interesting to see the implementation reason that businesses fail at using machine learning. I never realized that the lack of outsourcing really is a detrimental factor to businesses when using machine learning. After reading the article I understand why, if you are trying to implement something that you don’t know how to use, you are bound to make some mistakes, especially in this complex field. Another factor that really never occurred to me was the lack of a use case when a business went in to use machine learning. I always had a use case when I used it, so I never considered how important it was to have one and how hard it can be if you don’t. this article also presented another important point, of the lack of good data going into the algorithms, so you won’t get good insights coming out – garbage in, garbage out. Since I am invested and interested in this field, I know this and have come across a number of times that bad data, has been the cause of many problems for me, so I know how important it is to clean your data. But, since this aspect of machine learning isn’t glamorized as much, most people underestimate its importance to the whole process, which can cause serious problems. Finally, the lack of an appropriate company culture. Since machine learning is such a hot topic right now, it really feels that as soon as you decide that you are going to use it, there are insights for you the next day. However, this is simply not the case, since these models take a lot of time to be made, trained, tested, tweaked, and then finally used. Again, to someone that works on these models, it makes sense, but I completely understand how someone who might just be caught up in all the fanfare and not be truly informed in this field can get swayed elsewhere. I also liked the way that the article was written, it was extremely thorough and easy to follow which made it an enjoyable read.
Conclusion
All in all, this article was a good one. It was extremely informative and effective the use of analogies helped me understand what it was saying from abstract concepts to actual real-world examples. I highly recommend you read it when you have the chance(the article can be found here).