On December 9, OpenAI released its largest internal survey on "AI productivity" to date—covering 9,000 employees across core roles such as data science, engineering, communications, and accounting. The results show that AI tools save employees 40-60 minutes per day on average, and 75% of respondents clearly feel an improvement in "speed or quality." Messages related to coding in the engineering/IT department have increased by 36% over half a year, and enterprise paid seats for the ChatGPT workspace have exceeded 7 million.

Hard metric: 40-60 minutes saved per day  

- Data science teams save the most: an average of 58 minutes per day, used for feature cleaning and code debugging  

- Communications roles: drafting copy + multilingual translation cut work hours by nearly 50%  

- Accounting department: report reconciliation and VBA generation have moved from "hour-level" to "minute-level"

Soft metric: 3/4 of employees claim to have "accelerated"  

- Frequent deep users (calling more than 20 times daily + using multiple tools) score 27% higher in efficiency than light users  

- Low-code roles are rising: employees who previously did not code now use AI to generate scripts, SQL, and macros; coding messages have increased by 36% over half a year.

External skepticism vs. internal data  

A study by MIT in August pointed out that "the ROI of enterprise AI is almost zero"; Harvard and Stanford also warned about the proliferation of "workslop" (low-quality superficial content). OpenAI Chief Operating Officer Brad Lightcap responded: "External conclusions do not align with actual adoption."

- Enterprise adoption is faster than consumer side: financial, healthcare, and manufacturing POC → mass production cycles have been shortened to 90 days  

- 1 million companies have paid to use ChatGPT, with 7 million seats, up 120% year-on-year

The overlooked "capability spillover"  

Chief Economist Ronnie Chatterjee emphasized: "75% of employees say 'I can now do things I couldn't before'—this is a dimension that cannot be captured by simple ROI numbers." Typical examples include:

- Marketing staff using Python to run A/B tests without waiting for the data team  

- Legal assistants generating 200 initial NDAs in 30 minutes, with 94% accuracy

Next step: turning "feelings" into "hard metrics"  

OpenAI plans to launch the "Productivity Dashboard" in Q1 2025—real-time tracking of time saved, lines of code, and task completion rates. Companies will be able to compare departments and industry benchmarks to manage budgets and performance.

At the same time, the company is collaborating with academic institutions to assess the long-term quality of AI output using a "double-blind + peer review" approach, to avoid "self-promotion" disputes.

Industry Insight

When "saving 60 minutes" meets the critique of "workslop," AI companies are solving it by quantifying "subjective experience" into a "dashboard" and accepting third-party academic audits. If OpenAI can continue to expand the "capability spillover" effect among its 7 million paid seats, it may offer a new ROI calculation model for the enterprise software market. AIbase will continue to track the release of its dashboard and the progress of academic reviews.