Why AI speeds up the work but doesn't move the outcome
The difference between work time and total time explains why so many AI projects never reach the business.
A while ago I sat down with the purchasing team of a retail company that wanted to understand how to improve their processes with AI. They were already adopting it and applying it day to day; still, the director wasn't seeing the department's key numbers move.
When we started digging in, the process they had set up worked like this: the analyst logged an invoice in about 30 minutes, a time that AI brought down to 12. A real, concrete improvement.
But the total time, from when the invoice was issued to when the money came in, barely changed: from 45 days it had dropped to 38.
The director's question was simple: how can it be that if I cut the task almost in half, the billing time only dropped 15%?
The point is that of those 45 initial days, the actual work was 30 minutes. Everything else was waiting or idle time: approvals, rework, back-and-forth between teams, things that happen while the case travels through the whole system.
That's what almost no one separates. One thing is work time, the time to execute a task. Another is the total process time. AI, today, is strongly accelerating the first. But if the process spends almost all its time waiting, speeding up the work doesn't move the total.
That's why improving each person, one by one, rarely changes the business result. You have to look at the whole process.
Next time something takes too long, it's worth looking at where it sits waiting before thinking about how to make it faster.
Many organizations live the same paradox: they adopt AI across several areas, each person works faster and, even so, the business indicators don't move. The answer isn't in the team's effort. It's in what you measure.
A field experiment with more than 7,000 workers showed this same pattern. AI was integrated for email, meetings and individual tasks. They gained time, but the way the team worked didn't change. AI improves what each person can change alone, not what needs coordination to change.
From there comes a practical rule for any redesign with AI: the key isn't to speed up every step, but to identify and redesign where the wait piles up. The bottleneck.
Rule of thumb: if processing time drops and lead time (the total process time) doesn't, that's where the bottleneck is.
Before adding a tool to each task, it's worth looking at the whole process and finding the queue. That's where the value lives.