• Petter Hultin Gustafsson

Why your data science efforts haven't given you instrumental value

Experiments and prototyping in notebooks is great and will give you valuable insights, but it won’t have any instrumental value.



Instrumental vs intrinsic value


These concepts have largely been used in moral or practical philosophy where values such as friendship or wisdom is believed to be valuable for its own sake. The value of gold or stocks however are instrumental in the way that they can be used as a means to an end with no value for its own sake. Imagine you are lost in a forest, in this scenario wisdom would be more valuable than a bag full of gold.


So how can we try to translate these concepts to business and data science?


Generally a lot of data projects give insights — you experiment and prototype for new possibilities. Over 50% of them fail and only 13% make it to a stage where they can change your business. That being said, even if it didn’t translate to anything it could still give you intrinsic value in the sense that you got some nice insights, you got a better overview of your business models and a cool keynote presentation where people in the room learned something. These things are intrinsic in the way that the data scientists end result didn’t really change anything in a tangible way. It was valuable for its own sake but could never really be used as a means to an end.


However, having your data scientists generate instrumental value for your business is when the data goes in to production, it’s when your data has impact on your profit, revenue and real time interaction for your customers. Your data science projects have instrumental value when your data can be deployed and be used as a means to an end.


How MLOps is changing that


The concept of MLOps grew from the slow and ineffective way of businesses to organise their data workflow. A lot of data scientists are not familiar with the infrastructure they are operating in, and even less, can build a model that can be easily deployed, and better yet, maintained. Most machine learning projects have a terrible ROI because businesses don’t know how to structure their data teams and more importantly, let the individuals continue to develop and grow, as the industry progresses. In the data industry, things are changing at an incredible speed, and with that, the expectations on skillsets. MLOps is a necessity, just like DevOps, and will allow more and more individuals and businesses to truly deliver value from data to their customers without first going bankrupt.


MLOps is available as a plug and play solution to your AWS account.


Get started with a 30-day free trial.


Sign up here to get a handbook that’s relevant for you.