Skip to content

Guide

What does research say about AI and productivity at work?

· MaxiMind

Published studies show that AI helps people work faster and better on writing drafts and answering customer questions, in the studies below, and that it helps newcomers the most. But research also shows AI can be wrong while sounding completely sure, and people who trust it can be led into mistakes. So the safe way to use it is: AI drafts, a real person checks, the owner approves.

Here are three studies, explained in plain words. The numbers are averages under each study's own conditions. They are not a promise of results for your shop or school.

Does AI help people write faster?

Yes, at least for everyday office writing. Researchers Shakked Noy and Whitney Zhang gave working professionals writing tasks typical of their jobs. Half could use ChatGPT, half could not.

The group using AI took about 40% less time. Their writing was rated about 18% higher in quality.

Source: Noy & Zhang, Science 381 (2023).

What it means for an owner: first drafts are something AI does well. Replies to customers, social posts, notices: AI writes the first version, a person edits and signs off.

Does AI help new staff or experienced staff more?

New staff. Erik Brynjolfsson, Danielle Li and Lindsey Raymond followed 5,179 customer support agents using an AI assistant that suggested replies.

On average, issues resolved per hour went up about 14%. For novice and less-skilled agents, the gain was about 34%. For the most experienced agents, the effect was minimal.

Source: Brynjolfsson, Li & Raymond, "Generative AI at Work", Quarterly Journal of Economics 140(2), 2025 (NBER working paper w31161).

What it means for an owner: AI works like a coach sitting next to a new hire. It suggests the kind of answers your best people give. But the staff member still reads the reply and sends it.

When does AI make people more likely to get it wrong?

When AI is given a task beyond what it can do, and the person using it doesn't notice. Fabrizio Dell'Acqua and colleagues at Harvard Business School ran an experiment with BCG involving 758 consultants.

On many tasks inside AI's abilities, consultants using AI did better. But one problem was chosen on purpose to sit outside what AI could do at the time. On that task, consultants using AI were 19 percentage points less likely to get the right answer than those working without AI.

The authors call this the "jagged frontier". AI's abilities are uneven: two tasks can look equally hard, and AI may handle one very well and get the other wrong. From the outside, it is hard to tell in advance.

Source: Dell'Acqua et al., "Navigating the Jagged Technological Frontier", Harvard Business School & BCG (2023).

So is AI good or bad for a small business?

All three studies point the same way: AI is strong in the right places, and risky when nobody checks.

  • Strong: drafts, repetitive work, common questions, helping beginners get up to speed.
  • Weak: hard judgement calls, and anything that depends on a customer's particular situation.
  • Biggest risk: an answer that sounds certain but is wrong, and nobody reads it before it goes out.

How do you use AI safely?

  • Step 1. Pick drafting or repetitive work. For example customer replies, social posts, the weekly report.
  • Step 2. Let AI write the draft, never the final version. The final version always passes through a person.
  • Step 3. Have someone read it before it goes out. Check names, times, numbers and every promise to a customer.
  • Step 4. Don't trust AI straight away on new or hard tasks. Compare its work with how you do it now, a few times first.
  • Step 5. The owner approves anything about money. Discounts, refunds and commitments to customers always wait for you.

How does MaxiMind put this into practice?

MaxiMind's AI operations team works in exactly these three layers: AI drafts, a real person supervises, the owner approves. AI employees handle drafts and repetitive work. A real person at MaxiMind watches and checks. Anything about money, discounts or promises to customers waits for your approval. Website chat and phone answering are coming soon.

Want someone to set this up around your own business? See MaxiMind's AI operations team. Send us a few lines at maximind.dev.

Questions

Will AI definitely make my staff more productive?

Nobody can promise that. The studies measured averages under their own conditions. In your business, results depend on the work you hand over, how it is used and who checks it.

Who gets the most help from AI?

In the customer support study, newer and less-skilled agents gained the most: about 34%, against an average of about 14%. The most experienced agents saw minimal change.

Does AI make mistakes?

Yes. AI can be wrong while sounding very confident. In the Harvard Business School and BCG study, on a task outside AI's abilities, people using AI were 19 percentage points less likely to get the right answer. That is why a person always checks.

What should I hand to AI first?

Drafting and repetitive work: customer replies, social posts, the weekly report. Keep money matters, upset customers and hard decisions for people.

Where can I read the original studies?

Each study is linked in the article: Science (Noy & Zhang, 2023), the Quarterly Journal of Economics (Brynjolfsson, Li & Raymond, 2025) and Harvard Business School with BCG (Dell'Acqua et al., 2023).

Tell us about your business

Send a few lines and we will read them and get back to you to set up a call.

Online booking opens soon. Until then, the booking button opens our contact form and we agree a time with you.

Đọc bằng tiếng Việt