IBM research suggests only 16% of AI projects scale enterprise-wide. The other 84% are multiplying in silos, sprouting up wherever there's a budget, a willing manager, and a bright idea... But these "mushrooms" aren't always scaling to deliver corporate strategy.
Without broad buy in and a true mandate for change from the top, even great projects fail to scale.
And the trouble is, AI makes it trivially easy to start new things. But companies get seduced by AI which can leave you with 1000s of unfinished projects.
A live example: Johnson & Johnson
After a year of broad experimentation, J&J realised its “thousand flowers” approach had generated nearly 900 AI use cases — many redundant or underdelivering. Only 10–15% were driving 80% of the value.
J&J’s CIO, Jim Swanson, explained that the company pivoted by removing a centralised governance board and distributing oversight to the teams best placed to judge value — commercial, supply chain, and research. Redundant projects were shut down, and resources were focused on high-value use cases such as:
· Drug discovery – optimizing lab processes for efficiency
· Supply chain risk – identifying and mitigating potential raw material shortages
· Internal chatbot – handling 10 million employee queries annually
· Rep copilot – coaching sales reps on engaging healthcare professionals, now expanding from medicine to MedTech
As written up in the WSJ:
"The company is tracking progress in three buckets: first, the ability to successfully deploy and implement use cases; second, how widely they are adopted; and third, the extent to which they deliver on business outcomes.
Swanson said the broad experimentation phase was necessary to learn about the technology and what it was good at. “You had to take an iterative approach to say, ‘Where are these technologies useful and where are they not?’"
I really recommend the full article in WSJ, here.
How to Adopt AI Change in Your Organisation
There are three stages of AI change:
- Cultural adoption
- Refocus on ROI
- Deliver multi-year use cases
1. Cultural adoption
The world is not full of tech optimists scrolling LinkedIn and believing every word. One CEO I spoke to recently said they’d surveyed their staff and found that 25% were actively opposed to more AI use—not just ambivalent, but opposed.
That’s why there’s an inevitable first stage: encouraging what I call the AI mushrooms. This means making “AI-first” a corporate mantra, asking people openly, “Where did you last use AI in your job?” and giving them the right tools—safe spaces to experiment, proper training, and certainly not expecting them to test things on their non-secure, personal devices.
The difference with LLMs compared to previous waves of tech is that they’re actually usable by non-experts. Anyone can try them.
At this stage, you’ll see your AI champions emerge. These are the people who naturally love tech—the early adopters who always buy the latest gadget. (To be clear, this isn’t me. I’m a follower in my own world, not a leader.)
These champions need a culture where their experiments are celebrated and shared. When ChatGPT first launched in 2023, I was constantly asking my team: “Where did you use it this week? Show me how to think about this.” I needed their help to see what was possible as honestly it didn't work very well for me and I was not impressed.
Not everything will work—and that’s fine. My first attempt at using LLMs to help me write emails and blogs (like this one) was f-ing useless. But that experience was necessary. I learned how to prompt better, I learned its limitations, and I stuck with it. Today, it is… more useful (I edited out Chat GPT’s unabashed self-praise here when it attempted to draft this paragraph 😆).
This is the cultural stage. Nothing scales yet.
2. Refocus on ROI
The next stage is where business units and P&L leaders need to step in. They have to spot what’s happening, and harness AI around the big problems—the ones worth fixing. This is the J&J story.
And here’s the trap: LLMs look convincing. So there’s an expectation that they just “work.” But my experience is that the first layer is all surface. To get them delivering real business value takes iteration, investment, and focus—the same as any major development project.
Eighty percent reliability is not enough. Getting to 95% takes work, and you can’t do that for everything. So you need to be selective. Choose the big bets.
This is also where the “build vs. buy” question becomes real. It might be the right time to switch from in-house experiments to bought-in solutions. (I wrote about this last week here: https://www.linkedin.com/pulse/buy-vs-build-why-top-us-firms-making-mistake-building-rosie-bailey-2b5je)
It’s also the stage where many projects will die. That’s tough, culturally and managerially. But killing projects is part of the process, as long as it’s done with a learning mindset and entrepreneurial spirit.
3. Deliver multi-year use cases
And then comes the hardest mindset shift: multi-year use cases.
Because LLMs are fast. They give answers in seconds. Small wins feel quick. And it’s easy to forget that real business change takes time.
The biggest ROI is in re-engineering your tech stack, rethinking how you store and manage data, or changing the processes that underpin your business. And yes—that involves people switching focus, changing responsibilities and maybe even redundancies.
This is where people become both your biggest asset and your biggest risk. How do you bring them along when, honestly, they might be scared for their jobs? I am not sure I know the answer here TBH, we need to lead with empathy and listen hard.
The truth is this industrial/tech revolution is playing out much faster than the last industrial one, which took a generation to replace the factories. This time, people need to be prepared to reskill themselves and evolve in real time.
Leaders need to be prepared to guide them through it.
Find out more from Nibble's experience negotiating 100,000 times a month here.
Interested in Nibble?
