Microsoft's Work Trend Index measures opinion, not AI's effect

What Microsoft actually measured

On 9 May 2023, Microsoft published Will AI Fix Work?, an entry in its annual Work Trend Index series. The report describes an online survey of 31,000 full-time workers across 31 countries, conducted between 1 February and 14 March 2023, roughly 1,000 respondents per market, alongside separate analysis of Microsoft 365 and LinkedIn usage signals such as meeting counts and email volume. This is a self-reported opinion survey plus company telemetry, commissioned and published by a vendor that sells AI productivity tools, not a randomised or observational study of what AI assistance actually does to output.

What it found, described as what it is

Respondents reported feeling overloaded: 64% said they struggled with time and energy for their job, and 68% said they lacked uninterrupted focus time during the workday. Self-reported time split showed 57% spent on communication (meetings, email, chat) against 43% on creation. On AI specifically, 70% said they would delegate as much work as possible to it to reduce their load, and majorities said they would be comfortable using it for administrative, analytical and creative tasks, while 49% worried it could replace their job. These are attitudes and self-assessments of current strain, gathered in the weeks after ChatGPT's public debut, not measurements of what happened when people actually used AI tools at work.

Why the vendor-survey label matters

None of this makes the numbers false, but they answer a different question than a study like the NBER paper on generative AI in customer support, which measured resolved chats per hour before and after a real tool rollout. A survey asking whether people feel overwhelmed, or whether they would like AI to help, cannot show whether AI assistance would actually reduce that overload, and Microsoft's own description of the series across years mixes self-reported survey waves with anonymised product-usage signals without separating, in the top-line figures, which claims rest on which method. Citing “70% would delegate work to AI” as evidence that AI improves productivity conflates a stated preference with a measured outcome.

Questions to ask before you adopt it

• Is a cited statistic a stated opinion, a self-report of current behaviour, or a measured before-and-after result?
• Does the publisher sell the product the survey's respondents are describing?
• Would you make the same decision if the number were reported by an unaffiliated researcher instead?

A large, well-fielded vendor survey is a legitimate source for what workers say they feel and want. It is not a substitute for a study of what a tool actually changes, and the two should never be quoted with the same confidence.