Speed at one station is easy to see and easy to celebrate. Someone finishes a report in twenty minutes instead of two hours, and it feels like an unambiguous win, because it is one — locally.
Whether that speed turns into something delivered, decided, or shipped is a longer and much less visible chain to trace. It runs through approvals, handoffs, other people's calendars, and other people's priorities — none of which sped up just because one step did.
Throughput, in the sense the book uses it, is the rate at which a system generates completed value — money through the door, decisions made, product delivered. Not the rate at which any one resource moves.
AI is extremely good at increasing local speed. It is not automatically good at increasing throughput, because throughput depends on the whole chain, and most AI tools are only ever touching one link of it.
I've started asking a different question before celebrating a speed win: did this make the system faster, or did it just make one person's part of the system faster? Those aren't the same question, and they don't always have the same answer.
Current Hypothesis
Local speed and system throughput are frequently reported as the same metric and are almost never the same thing.
Not a conclusion — a working idea, revisited as evidence comes in.