A recent HBR piece on AI transformation got me thinking – not because I disagreed with it, but because it stopped one step short of where the real problem lives. The frame was something like: most AI initiatives fail in the “last mile,” the gap between promising pilots and production value. That’s true. But it didn’t answer the question I keep running into: why does the last mile feel so familiar every single time?
I think I know why. Every AI ROI conversation I’ve been part of starts the same way: a spreadsheet. Hours saved. FTEs redeployed. Licensing costs amortized. The numbers go in the cells, the cells tell a story, and the story usually ends with a business case that gets approved and then shelved six months later when results don’t materialize.
The spreadsheet isn’t wrong. It’s just incomplete.
What it captures is what AI replaces. What it almost never captures is what AI creates. Three units of value that never show up in the model: handoffs avoided, decisions accelerated, and judgment applied faster. Those don’t have a column. So the model treats them like they don’t exist.
Think about what it means when an analyst who used to spend all day Monday building a report now gets that report at 8am on Monday. The value isn’t in the hours saved – it’s in what they do with Monday morning. Do they go deeper on an insight they never had time to pursue? Do they catch a trend before it becomes a problem? Do they finally have the context to make a call they’ve been deferring? That’s the return. It doesn’t fit in a cell.
But measuring it is actually the easier half of the problem. The harder part is what happens to the organization around it.
Most companies have spent years – sometimes decades – optimizing around the gaps that AI just closed. Your processes assume human bottlenecks. Your roles are built around them. Remove the bottleneck and the surrounding infrastructure still points to where it used to be. That’s not a technology problem. That’s an organizational immune system problem — and it will quietly reject the transformation while the dashboard shows green.
In regulated environments, the immune system is even more entrenched. The bottlenecks you’re removing often exist because someone decided, at some point, that a human had to be in the loop. Some of those decisions are real: regulatory, liability, audit trail. Some of them are archaeological – a process designed in 2009 for a reason nobody remembers. AI forces you to find out which is which. That’s uncomfortable work. In insurance and healthcare, it’s also unavoidable – because the cost of getting it wrong isn’t always just a sprint retro, sometimes it’s an exam finding.
The question I use to get my team oriented: what do you own now that you couldn’t have owned before? Not “what can the AI do” – but what does that free you to decide or to now drive? That’s the question that separates teams that absorb AI from teams that merely tolerate it.
The spreadsheet will tell you the AI saved 200 hours a month. It won’t tell you whether anyone redesigned the 200 hours of work that depended on those hours being slow. That’s the gap. And closing it isn’t a technology decision — it’s a leadership one.