How 758 consultants delivered work rated 30% higher with AI

Do employees who use AI deliver better results?

A consulting firm had to decide whether to put AI in the hands of all its people. That it feels faster was clear early on. Whether better work comes out of it, nobody could say. So it was not debated but measured.

GPT-4 from OpenAI for everyone

Employees were given GPT-4 from OpenAI. With it they could gather ideas, write drafts, rephrase, summarise, structure, look for counterarguments and critique texts. Human graders read the results without knowing who had worked with AI and who had not.

Rated over 30% higher, finished over 25% faster.

• over 30%higher-rated work – graded by people who did not know who had used AI
• over 25%faster to finish
• over 12%more tasks completed

The gap did not come from the AI doing the work, but from it being in the room from the first minute. The comparison ran against the control group in the same firm, not against an industry average.

The study also shows where the AI's capability frontier sits: on one task deliberately placed beyond it, the answers were less often correct – and more convincingly phrased than the control group's. The error came well packaged. Which is why the audit settles first which tasks belong inside.