Why many data science projects fail? PART IV
Overengineering - complexity kills projects.
Everyone wants to work with latest tech. Data scientists wants to learn neural networks and new frameworks. Companies wants to show how advanced they are.
Start small and make first version work well! Simple regression is good enough - leave advanced models and better performance for second iteration.
No comments yet
Be the first to comment
MORE INSIGHTS YOU MIGHT LIKELearn more by discovering other insights
Why so many data science project fails?
I will do series of posts about this topic.
First problem. Most people just don’t know what “data science” does.
Data science is misunderstood with reporting, business analysis or data engineering. Often all of above is combined into single project and called “data science”. People constantly expects multiple outputs from “data science” project, which leads to failed projects.
My suggestion - define single project objective. And see if data science is actually needed at all - likely it is engineering problem.
Why many data science projects fail? PART III
Collected data usually have serious quality problems.
Data science is by-product of usual business operations. Business operations are often messy. Building new projects on weak foundations is extra hard.
I don’t know shortcuts here. Companies should embrace analytical culture early on in the journey. Strong data foundation will maximise chances to excel in data science.
We seriously lack data science management talent.
There is a shortage of experienced data scientists ready and wanting to become managers. Data scientists also lacks management skills..
On the other hand we have strong product managers with no experience in data science project.
Solution would be to devote much more management resources to data science projects. Consider having dual management structure: data scientist converting to management AND senior project manager.