A 75% reduction in writing time. A 66% reduction in planning time. And still, the people who actually knew the subject produced better work.
That's the core tension in a new study from Harvard Business School, where researchers asked 78 employees at IG Group, a derivatives trading firm, to write investment analysis articles. The participants fell into three groups: 12 web analysts who do this kind of work daily, 26 marketing specialists with adjacent knowledge, and 40 technology specialists with no domain background.
The Speed Gains Are Real
AI's impact on efficiency was dramatic across the board. Conceptualization time - the planning and research phase - dropped from 63 minutes to 23 minutes. Actual writing time fell from 87 minutes to 22 minutes. These aren't marginal improvements. That's the kind of time savings that changes how teams think about staffing and project timelines.
But speed only tells half the story.
Quality Still Follows Expertise
When independent evaluators scored the articles on a 5-point scale, the web analysts - the domain insiders - averaged 3.96. Marketing specialists, who had related but not direct experience, scored 3.92. The technology specialists, even with full AI assistance, managed only 3.42. That's a 13% quality gap that AI couldn't bridge.
The researchers, led by HBS Associate Professor Iavor Bojinov, identified why: AI handles what they call "codifiable" tasks well. Brainstorming, organizing information, generating first drafts - these are pattern-matching exercises where large language models (LLMs, the technology behind tools like ChatGPT and Claude) genuinely excel. But interpreting what the data actually means for a specific market, knowing which details matter, understanding the nuance that separates a competent article from a good one - that still requires what the researchers call "lived experience."
As the team put it: "AI makes you feel like you can do anything. But can you do it as well as people whose job it is?"
What This Means for How You Use AI
This study confirms something many daily AI users have already noticed intuitively: AI is a multiplier, not a replacement. It makes your existing 7/10 skill into faster 7/10 output. It does not make your 4/10 skill into 7/10 output.
The practical takeaway is to use AI most aggressively in your areas of strength, where you can spot when it's wrong and steer it toward better answers. In domains where you lack expertise, AI will make you faster at producing mediocre work - which might actually be worse than producing nothing, because mediocre work delivered with confidence is hard to distinguish from good work until it causes problems.
The 78-person sample is modest, and a single company in financial services doesn't represent every industry. But the finding aligns with a growing body of research suggesting AI's biggest productivity gains go to people who already know what good looks like.