A new BCG study blames bad management. The gains are real at the task level and leak away at the system level.
A new BCG study says it’s figured out why AI isn’t in the productivity numbers. The culprit? Bad management — leaders “really struggling to articulate what the vision and strategy is.”
Part of that is fair. The data’s real: 42% of the 12,000 workers surveyed say they save a workday a week, and two-thirds got no guidance on what to do with the time. The fear point is real too — people hide their AI use instead of sharing it. Clarity matters.
But watch the move. A consulting firm studied tech’s messiest economic puzzle and concluded the fix is clearer communication — which happens to be what consulting firms sell. It also skips its own reporting: compute can cost more than the people it replaces, Microsoft is dropping AI coding licenses, Uber burned its annual AI budget in four months. You can’t town-hall your way out of negative unit economics.
So what is the source? The gains are real at the task level and leak away at the system level — because we bolt AI onto workflows built for humans. A faster step in a process you never redesigned doesn’t move the whole. Solow’s computers in 1987, electric motors before that: productivity only jumped once firms rebuilt work around the tool and the tech got cheap.
A clear vision is the start line, not the fix. The fix is structural: rebuild a few processes end to end around AI, remove steps instead of just speeding them up. Deploy where the economics already pay. Measure outcomes, not tokens.
Real and worth-it-yet are different things. That’s the actual paradox.
