Why 90% of AI projects fail
Many AI projects never reach the Live phase. In fact 90% of AI projects are terminated way before that, just after/during POC.
But why?
Here is the scenario: the company executives agreed to allocate a budget for an AI project that uses the company’s data to predict the stock inventory shortages (or buying patterns, or next fraud etc.) for an expected ROI of 30% yearly (let’s say). Discussing with the development team, they have agreed to certain metrics like accuracy of the AI solution , its response time (how much time/cpu power it takes to make a prediction).
All agreed, now the AI development team is set to work.
Naturally there are going to arise unexpected issues: the data is actually more scattered than it seemed (it would cost more to integrate it), the POC runs as intended but scaling on a real 1B items inventory leads to 1 day response time instead of 10 seconds, real dataset needs significant cleaning even on production etc.
The AI project leader needs to go back to the stakeholders to say the metrics values are not achieved and the initial ROI expected has been eroded. Result? The project is terminated.
I give an alternative end.
The team leader as soon as a new issue arises, he re-defines a new value for the AI solution.
Example.
Recommended by LinkedIn
Ok, we cannot deliver any more AI predictions on all inventory items with a response time within 1h without incurring significant CPU costs.
But the dataset now has been cleaned and we know that selected category items X and Y can be predicted easily (read cheaply) AND the company buys many of them (let’s say X+Y is 30% of all inventory).
Based on this understanding (not available from the beginning), Redesign the solution with new metrics and deliver a new ROI. In this case the solution should cost less (less items) but still deliver value.
The difference between a terminated and a successful AI project is the ability to re-define value as the development issues arise.
In short:
‘Show me the money’ is the way. Applied to AI project I should say
‘Show me the (new) money’ is the way
Integration Strategist for Business Operations, Project Performance, and Digital Transformation
1yHi Andrea, I know this is short notice, but there's a gathering of construction industry enthusiasts to discuss AI projects for construction quality. It's a brainstorming session to kick off a larger event for PMI/CQM in September. Would you like to join and offer some insights? Here is more information at http://cqm.us/CQi.
ERP, Data, & AI Leader | Strategy Consulting & Delivery | Driving Digital Transformation & Business Impact
1yAndrea Isoni Agreed. so real. Isn't part of the AI and analytics, the discovery of real insights (expected vs. unexpected) rather than what we think we know
Leadership and Keynote Speaker and member of the Data Science Research Centre at University of Derby
1yThis is a really interesting and valuable approach.