95% failure

  • 22nd Aug 2025
    •  • 
  • 4 min read

It’s time to weigh in on the paper everyone’s been talking about: the State of AI in Business 2025 report that came out last week from MIT’s Project NANDA. Y’know the one that said “95% of organizations are getting zero return” on GenAI. Yeah, that one.

First, for those of us in the organizational transformation space for a while, this comes as no surprise. In his 2019 book, Why Digital Transformations Fail, author Tony Saldana had many of the same findings as in this new MIT paper, with roughly the same odds. “90% of success in digital transformation is determined by change management or culture…only 10% by the technology.”

The central issue of any transformation project is that once it gets framed as a technology project (which almost always happens), alignment is lost and it’s just a matter of time before it fizzles out.

In other words, when we focus on the what of transformation (the tech) rather than the who of transformation (the people), it falls flat.

One would think the business world had repeated this experiment often enough to learn from the results, but apparently not. And I have some theories as to why.

  1. The Myth of Leadership. This is a sacred cow that can be perilous to one’s career to take on, but as I’ve written about before “leadership” isn’t really a thing in the business sense. It has much more in common with words like “rich” or “secure” or even “love” than it does with processes, deliverables, or decisions. But those who self-identify as leaders, or even aspiring leaders, tend to fixate on their own alleged power and determination over the power of culture and habit.
  2. Fear of Lifecycle Design. Everything changes, everything dies. Not exactly controversial, but topics we perishable humans prefer to sidestep. However, our avoidance blinds us to design opportunities. When we design for every element of the business (and indeed the people in it!) to change and eventually die, we design differently and as I often argue, much better! Only when we see each stage can we prepare to do every stage well.
  3. Outdated Org Structure. As I wrote about in a recent edition of my weekly newsletter, it’s not effective to play “pin the AI on the org chart.” Modern organizational structures are actually pretty old, they came in with the assembly line era we’re still using them in the knowledge work era. Command and control hierarchies and matrices still kinda mostly worked…until AI. Now as knowledge becomes an abundant and quantifiable (tokenized) resource, business workflows and structures are breaking the rest of the way. Businesses can’t see this very well because it happens between the lines in their org charts.
  4. Shiny Object Syndrome. The old part isn’t new. Since people are talking so much about Artificial Intelligence as the next thing, it’s easy for the Human Intelligence part to feel less conversationally sexy. We tend to fixate on what is new and novel, often accompanied by fear of the unknown. It’s a primitive and entirely predictable reaction that tends to wear off with time, but it hasn’t worn off yet.
  5. This Time is Different. As I delight in pointing out frequently via LinkedIn and other places, all the business and relational problems we have with AI today are really the same kinds of problems we have with our coworkers and other fellow humans. Which is great news! Because we already have the interpersonal skills and devices to handle that. We may not be used to applying those old human lessons with “synthetic people spirits,” but they do work. Many people have argued with me on this, and all ended up agreeing in the end and it’s helped them do AI better. I know it doesn’t feel like it should be true, but at least so far…it’s not that different.

If you liked these thoughts, I bet you’d love my Signals & Subtractions newsletter. Here I get a bit more expansive, but over there I keep it light and tight: one signal, one strategic human prompt, one subtraction opportunity, and a new analogy every week that you can use to talk about AI with people who are still figuring out what it is and how to put it to good use. 300-500 words and a key graphic or two for your inbox, to help you navigate your workweek.