Ask a room full of business leaders who’s rolled out AI, and every hand goes up. Ask who’s seeing it on the bottom line, and most of those hands come back down.
They’re not imagining it. McKinsey’s State of AI research found 88% of organisations now use AI in at least one part of the business, yet only around 6% report a significant impact on their bottom line. The survey has picked up the same pattern three years running: adoption keeps climbing, while value stays stubbornly flat.
The tools work, and people are using them every day. So where’s the value going? In this blog, we’re looking at where AI’s promised returns are getting lost, and what the businesses seeing real results are doing differently.
Where does all the saved time go?
Most AI rollouts follow a familiar script. Licences are bought, training is done, and before long AI is drafting emails, summarising meetings and writing code right across the business. Ask anyone and they’ll tell you it saves them time – and they’re right. But those savings have a habit of disappearing before they ever reach the business.
If ten people each save 5% of their week, your business doesn’t get half a person back. Nobody gathers up those fragments and points them at something that matters, so the time leaks out into more emails and more meetings.
Then there’s rework. AI makes it easy to produce more; more drafts, more slides, more code, and a lot of that needs checking, fixing or redoing. A study by MIT economist Mert Demirer and colleagues tracked more than 500,000 software developers. The most advanced AI coding tools more than tripled the amount of code they wrote, but the amount of software they actually released rose by just 30%. Three times the activity; a fraction of the progress.
Faster tasks, same slow business
Part of the problem is where AI gets pointed. Most rollouts speed up individual tasks, so the report that took a morning now takes ten minutes. Then it sits in someone’s inbox for a week, waiting for sign-off.
That waiting is where most of the time is now going. Asana’s research found knowledge workers spend almost 60% of their day on “work about work”: chasing updates, searching for information and sitting in status meetings. And it scales badly. In a team of 50 people there are 1,225 possible one-to-one connections, and every one is a chance for work to stall. Speed up the tasks alone and you’ve only improved the smaller half of the day.
We’ve seen this idea before. Lean, process re-engineering, ERP, robotic process automation; each promised a huge change, and each made individual activities faster. Very few fixed the handoffs between them. Drop AI into the same old workflow and it’ll join that list.
The 70% nobody budgets for
There’s a reason those handoffs survive every new wave of technology: fixing them is a people problem, and you can’t buy your way out of a people problem. BCG’s 10-20-70 rule says 10% of AI success comes from algorithms, 20% from technology and data, and 70% from people and processes.
Most AI budgets are weighted the other way round. The tools get bought, but the roles around them stay exactly as they were – research suggests the vast majority of organisations haven’t redesigned a single job around what AI can do.
Training tends to follow the same pattern; a course, a certificate and back to the day job. But learning that sits apart from real work rarely sticks, and people won’t experiment if they’re worried about being caught out. The businesses that move fastest are the ones where “I don’t know, let’s find out” is a perfectly good answer.
What the 6% do differently
McKinsey’s research singles out a small group it calls “AI high performers”: the 6% of organisations where AI drives at least 5% of their earnings. One thing stands out about them. They’re nearly three times as likely as everyone else to redesign their workflows around AI, instead of bolting it onto the old process. They’ve stopped asking where AI can help, and started asking how the work should be done.
They pick one workflow and look at it end to end (order to cash, claim to settlement, hire to retire) instead of scattering tools across departments. They measure it honestly, putting the time from start to finish next to the time spent on actual work. When a process takes three weeks and the work takes a day, it’s pretty obvious where to look.
They give that workflow a single owner, accountable for the whole journey, so it isn’t five departments each guarding their own step. And when the work changes, they change the jobs to match, so nobody drifts back to the old way of doing things.
From caterpillar to butterfly
A caterpillar that learns to crawl faster is still a caterpillar.
Give everyone a productivity tool and they’ll all get a little quicker, but the organisation keeps the same shape, with the same handoffs and the same sign-offs.
Rebuild a workflow around what AI makes possible and you get something different altogether. And like the butterfly, there’s no going back.
So here’s a challenge for Monday morning. Pick the one workflow in your business that frustrates you most. Write down how long it takes from start to finish, then how long the actual work takes. The gap between those two numbers is where your AI return is hiding – and finding it is how you keep your hand up next time someone asks.
