BoA just published a report calling AI bigger than electricity and the internet combined. But AI is currently lifting productivity by just 0.1% per year.

Ethan Mollick laid this out at the New York Public Library last week, speaking to several hundred corporate leaders.

Meanwhile, Goldman Sachs found no meaningful relationship between AI and productivity at the economy-wide level – while reporting 30% gains in the two sectors where AI has concentrated: Customer Support and Software. The gains are real. They’re just not scaling.

Mollick’s explanation for why: KPIs, he argued, are “the biggest enemy” of AI transformation. They force organisations into incremental improvement when the real gains come from replacing processes entirely.

“KPIs aren’t designed to measure what doesn’t exist yet. Or what does, but shouldn’t.”

They measure how well you’re doing the thing you already do.

So when AI makes it possible to do something fundamentally different – to replace a workflow rather than speed it up – the measurement system can’t see it. And what the measurement system can’t see, the organisation won’t fund, won’t resource and won’t reward.

This connects directly to what Microsoft’s data showed a few weeks ago – that 45% of AI users say it feels safer to focus on current goals than to redesign work around AI.

It’s a rational response to how performance is measured. The KPIs are doing exactly what they were designed to do. The problem is that they were designed for a world that’s no longer the one you’re operating in.

Having said that, KPIs exist for good reasons. Accountability, alignment, resource allocation – these don’t disappear because AI arrived.

The answer isn’t to abandon measurement. It’s to recognise that the measurement system itself is now a design problem.

If you only measure what AI improves within your existing processes, you’ll capture the 0.1%. If you can figure out how to measure and reward the replacement of processes – the work that didn’t exist before – you start closing the gap toward what’s actually possible.

Mollick also noted something I think is underappreciated: AI companies are now building their own consulting arms. That’s an admission that the models can’t solve the deployment problem. The technology works. The organisations don’t – yet.

Sources: Nick Lichtenberg, “‘Nobody knows anything’ and ‘this time is different’: the phrases that define – and haunt – the AI economy,” Fortune (25 May 2026). Bank of America and Goldman Sachs data as cited in the article.