Most AI implementations are tool deployments dressed up as transformation.
Not a surprise as for a long time, even the AI industry was suggesting these are tools not replacement humans.
Stanford’s research puts a number on the difference though: high performers are nearly three times more likely to redesign the workflow before deploying the tool. Most companies just drop the AI on top of what already exists and wonder why the results don’t come.
The research comes from Stanford’s Digital Economy Lab - 51 enterprise AI deployments across 9 industries, published this month. The headline stat is striking:
88% of organisations are using AI in at least one function. Only a third have started to scale. Two thirds are still running experiments.
Accenture has a name for this: the Proof of Concept Factory. The pilots never die. They just multiply.
What separates the companies that scale isn’t the technology - that was consistently described as the easiest part. Once you knew what was possible.
One professional services firm failed at AI twice in recruiting before working this out. The first attempt collapsed because the team assumed AI would sort out the broken workflow underneath it. When they tried again, they mapped the entire process first. That second attempt took one month and delivered 83% efficiency gains.
There’s a line from the research that captures it well: “AI amplifies whatever process it is applied to. If the process is broken, AI makes it worse faster.”
Not too little AI. Too little process work before the AI arrives.
One thing worth reading
The full report is 116 pages but the Key Findings on page 11 are worth ten minutes on their own.
The Enterprise AI Playbook, Stanford Digital Economy Lab
The finding on agentic AI is worth particular attention - implementations where AI handles 80%+ autonomously delivered 71% median productivity gains. Most companies in the study hadn’t gone near this model yet.
One thing to try
Before your next AI initiative gets approved, ask one question about the problem it’s solving: is this a painkiller or a vitamin?
The recruiting team in the Stanford research wasn’t mildly inconvenienced. They were drowning. That level of pain meant adoption was immediate - no change management required. They didn’t need convincing. They needed rescuing.
Vitamin problems - things that would be nice to improve - are where pilots quietly die. The team uses the tool when they remember to, finds it useful enough, and gradually drifts back to the old way. The pilot never formally fails. It just fades.
The quickest diagnostic: would the people using this tool notice if it disappeared tomorrow? If the honest answer is “probably not for a few weeks” - it’s a vitamin. Find the painkiller first.
No posts
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.