Your AI Is Working. Your Productivity Metrics Aren't.

Half of all workers now use AI weekly. Most businesses have spent money on it. And almost nobody can prove it's working. Three major studies — from the National Bureau of Economic Research, ADP Research, and Atlassian — all reach the same conclusion: AI productivity gains aren't showing up in the numbers. Not because AI doesn't work. Because businesses are still measuring productivity the way they did before AI existed.

The NBER study is the most striking. Surveying roughly 6,000 executives across the US, UK, Germany, and Australia, researchers found that nine in ten firms reported no measurable impact on productivity or employment from AI over the preceding three years. The average firm uses AI just 1.5 hours per week. At that dose, the surprise would be if it did move the needle.

ADP's People at Work 2026 report surveyed 39,000 workers across 36 markets. In Australia, 44 per cent of workers now use AI multiple times per week. Among daily users, something counterintuitive emerges: they are four times more likely than non-users to say they feel less productive than they could be.

Read past the headline and the picture shifts. Those same daily AI users report half the stress of non-users — 11 per cent experiencing negative stress versus 23 per cent. They report higher engagement, better team cohesion, and greater job security. They feel less productive, but by every other measure, they're doing better.

The explanation is straightforward. AI eliminates the small, repetitive tasks that generate a steady stream of completion signals — the email sent, the invoice processed, the schedule updated. Without those micro-achievements, workers feel less busy. But "less busy" and "less productive" are not the same thing. The routine work is gone. The higher-value work that remains doesn't produce the same sense of ticking boxes.

Atlassian's State of Teams 2026 report, surveying 12,000 knowledge workers and 170 Fortune 1000 executives, quantifies the gap. Eighty-nine per cent of executives say AI has increased the speed of work. Six per cent can point to specific, measurable ROI. We wrote recently about this measurement gap in professional services specifically — the Atlassian data shows it runs across every industry.

For a trades business, the old metrics — jobs quoted, calls answered, hours logged — still get tracked. But if AI is handling quoting, call intake, and scheduling, those activity counts drop while revenue per truck per week climbs. Without outcome-level tracking, the owner sees activity falling and assumes something is broken. For an accounting firm, the pattern is identical: documents processed per hour drops because AI handles the routine ones silently, while realisation rate and client retention improve in ways nobody is recording.

Atlassian estimates the coordination overhead from fragmented AI deployment — what they call the "fragmentation tax" — costs Fortune 500 companies roughly $161 billion annually. For an SME, the dollars are smaller but the ratio is the same. Four disconnected AI tools with no outcome tracking is like installing a turbo and judging performance by how loud the engine sounds.

The 14 per cent of teams in Atlassian's study that reported clear AI ROI had one thing in common: they changed their metrics before they changed their tools. They measured outcomes — revenue generated, errors avoided, response times — rather than activity. For trades businesses, that means tracking revenue per field worker per week, first-time fix rate, and average days from quote to signed job. For professional services, it means realisation rate, write-off percentage, and revenue per client per hour invested. This is fundamentally an operations throughput problem — you need to know what's coming out, not just what's going in.

Pick three outcome metrics that matter to your bottom line. Track them for 30 days. Then deploy or expand your AI tools and track the same three for another 30 days. If you can't see the difference in those numbers, the problem is either the wrong AI tool or the wrong workflow — and you'll know which. The businesses in the 6 per cent didn't get there by buying more software. They got there by deciding what "productive" actually means.

Key takeaways

Not sure which metrics to track?

A 30-minute call can identify the three outcome measures that matter most for your business — and whether your current AI tools are moving them. Book a call to get started.