The Realistic Copilot: Why 2.8 Percent Time Savings Is Not a Failure

The marketing promises sound like hours a day. The largest study so far on chatbots at work reaches a sober conclusion. Anders Humlum and Emilie Vestergaard analysed data from around 25,000 workers in 7,000 Danish workplacesS. Across the whole working day, chatbots saved 2.8 % of working timeS on average. The researchers found no measurable effect on wages or hours worked.

Why that is not a failure

Let us do the maths. 2.8 % of a 40-hour week is just over an hour. At a full cost of 50 euros per hour, that is about 56 euros a week or roughly 240 euros a month. A business licence costs 20 to 30 US dollars a month. Even at the sober average, the assistant clearly pays for itself.

A second finding matters even more. Where employers actively supported their staff, with training and an official company chatbot, workers reported larger gains. So 2.8 % is not a law of nature. It is a starting value.

That matches how companies see themselves. In McKinsey's August 2026 survey, 80 % of respondentsU report personal productivity gains. Self-reports regularly overstate such effects, though. That is why it pays to measure yourself.

Five routines worth building

1. Email triage: Have unread mail sorted and get drafts for what is urgent.
2. Drafts: Never start proposals, concepts or summaries from scratch again.
3. Meeting prep: Build an agenda from the last minutes and open tasks.
4. Research with sources: Accept answers only with evidence, and open the sources.
5. Spreadsheet analysis: Have anomalies found, and double-check every number once.

The workflow, using email triage as the example

Sort my unread emails into three groups: reply today, this week, read only. Justify each assignment in a few words. For the "today" group, write draft replies in my tone: friendly, direct, five sentences at most. Do not invent commitments or dates.

Limits and risks

• Time that evaporates: Saved time fills up by itself. Decide what it will be used for.
• Review effort: Every draft has to be read. Forget that and you are flattering the savings.
• Self-reporting: Perceived time savings are not a measurement. Our article on the METR study shows how far apart the two can be.

How to start

1. Pick two of the five routines and use them daily for two weeks.
2. Every Friday, note how much time you think you saved and where rework was needed.
3. Keep what works and drop the rest.

Sources

1. SHumlum, Vestergaard: Large Language Models, Small Labor Market Effects, NBER WP 33777, 2025
2. UMcKinsey: The state of AI in 2026, On the road to ROI, 25 Aug 2026