Top Takeaways from ODSC session: ‘Why ML Projects Fail and How to Avoid Them’

July 29, 2021

AI and ML are disrupting every industry and companies are investing heavily in hopes of gaining any kind of competitive edge. Yet, about half of their ML projects, 47% to be precise, never make it to production i.e. they never pass the proof of concept stage or trial stage.

At ODSC Europe 2021, Natasha Montagu, Solution Architect at iMerit provided real-life lessons to help avoid the most common machine learning pitfalls. Natasha has the unique experience of working on both the client and vendor sides of hundreds of projects across multiple industries, assisting companies in unlocking the power of their data and transforming it into cost-effective, scalable models.


Here are 5 key takeaways from the session:

  • Data deficiency is among the most common reasons behind ML project failure. To combat this, it is critical to ensure that the data is both relevant and accurate, in addition to ensuring availability.
  • Data is a historic snapshot and not necessarily an accurate representation of the data yet to come and any significant change in the data can cause the model to give inaccurate results. To ensure high-quality data, an effective data collection strategy should be in place.
  • To minimize human obstacles causing an ML project to fail, it is essential for all stakeholders of the project to collaborate, communicate, and coordinate effectively. 
  • When a project involves several subject matter experts, holding one person accountable for the labeling decisions ensures consistency while also allowing for verification of the decision’s reasoning and pivoting if required. 
  • Failure can often be the best teaching. One shouldn’t be afraid of failing and instead reflect on the learnings and plan effectively to rise back stronger. 

To watch the on-demand webinar, click here.

Why ML Projects Fail and How to Avoid Them