An article linked in this blog discusses why 87% of AI projects never make it into production

Deborah Leff and Chris Capos discussed the reasons for failure. Deborah was CTO for data science and AI at IBM, and Chris was SVP of data and analytics at Gap. Here is the link to the original article. My take on the article is following:

Reasons for high failure rate:

• Many don't have the proper leadership support to create the conditions for success.

• Data is a double-edged sword. ML products can only be built if there is enough data. But bad quality data can ruin a lot of hard work.

• Data lives in different formats, structured and unstructured; they are kept in different places with various security and privacy requirements, meaning that projects slow to a crawl right at the start because the data needs to be collected and cleaned.

• Launching an ML product requires an extraordinary level of collaboration which many companies are not used to in this way.

• Lack of monitoring of the model after development is another reason for failure. After handing in the ML model, if there is no dedicated team to monitor it, the model deteriorates its performance.

• Failure in figuring out how we educate the business leaders across the organization. It is crucial that business leaders understand these concepts. AI will not replace managers, but managers who use AI will replace those who don't

Screenshot

the original article appeared in Venture beat

To ensure success:

• The key to success is keeping it simple. It is not about the sophistication of a model; instead, creating a better experience for customers

• Pick a small project to get started

• Choose a pain point to solve where we can show demonstrable progress

• Ensure having the right team, cross-functionally, to solve the problem

• Leverage third parties and folks like IBM and others to help accelerate the journey at the beginning.