Failure to launch: the hard realities of AI

A fascinating  recent report by MIT Sloan in collaboration with BCG has confirmed what many industry insiders have known for years: only a small minority of companies manage to make their initial AI projects succeed.

Based on global survey of more than 3,000 managers and scholars in 29 industries, the report’s authors discovered that only 11% of organisations saw significant financial benefits from their AI programmes.

However, for those of us working at the sharp end of predictive analytics, machine learning and data science for the last three decades, this is not news. To be fair, this view doesn’t represent  a major tech industry dirty secret, it’s just that in the blizzard of hype around AI and machine learning, few people want to tell stories or write articles about how often these initiatives fail to make an impact.

At Smart Vision Europe, we think that’s a shame because failure teaches valuable lessons. Especially for those who want to avoid it. Indeed we’ve gone so far as to deliver more than 35 public seminars on the theme of Predictive Analytics in Real World and this year, we even issued a free downloadable book called The Insider’s Guide to Predictive Analytics with the intention of addressing precisely this issue: how to not screw it all up.

In the last twenty years we’ve progressed from a landscape where AI programmes were the preserve of tech-savvy mavericks or niche experimental projects to a world where they are more common than not. In fact, the MIT Sloan report discovered that nearly 60% of the respondents indicated their own company had an AI strategy. What makes compelling reading, is that so many of the authors’ discoveries chime completely with what industry insiders have been preaching for years. This is particularly true with regard to the primary drivers of success in the 11% of companies that managed to make AI impactful.

The authors’ identified a willingness to “intentionally change processes, broadly and deeply, to facilitate organizational learning with AI” as a key prerequisite of AI driving significant benefits. Understanding from the outset, that in order to properly deploy these new capabilities, we need to plan and prepare for change, is not just a platitude, with AI, it’s an imperative. So much so that at Smart Vision, we ask our clients ‘What will you do differently?” and “Who does this affect?” long before we begin discussing algorithms.

The report also frames success in terms of three different kinds of learnings: organisational learning, human learning and machine learning. The authors point out the that these initiatives can give rise to a valuable feedback loop between the AI application providing results and humans providing necessary corrections and adjustments – “this cycle of mutual learning makes humans and machines smarter”. You can see this very notion echoed in the cyclical nature of long-established analytical methodologies such as CRISP-DM.

Finally, the report makes clear that successful AI programmes are not merely technical challenges that can be solved by “having the right data, technology, and talent, organized around a corporate strategy” rather they require “large-scale organizational shifts in mindsets”. Amen to that.

Download your free copy of our Understanding Significance Testing white paper
Subscribe to our email newsletter today to receive updates on the latest news, tutorials and events, and get your free copy of our latest white paper.
We respect your privacy. Your information is safe and will never be shared.
Don't miss out. Subscribe today.
WordPress Popup Plugin
Scroll to Top