Your team is using AI more often. Excellent—but is the work getting better, or are drafts simply arriving faster in somebody else’s inbox?
New research puts adoption in context
On September 15, Google announced an interactive, open-access experience for its AI & Economy ATLAS, making occupational and geographic usage data easier to explore. It also highlighted research with Google DeepMind and MIT FutureTech on scientific work. Google’s ATLAS update explains the findings.
That research includes a survey of more than 600 scientists in the US and UK and an analysis of 2,600 specialised AI models. Nearly half of surveyed scientists reported daily AI use, and respondents reported saving just under seven hours a week. Google also describes bottlenecks in testing and validation.
Those are reported findings from a particular study population. They do not establish that every occupation will achieve the same savings, or that time saved automatically produces additional discoveries.
Guava analysis: follow the work to completion
For business leaders, the useful question is what happens after an AI-assisted task. If a sales proposal takes less time to draft but legal review takes longer, the organisation may have gained less than the drafting team reports.
- Record a baseline for a complete workflow before introducing AI.
- Measure completion time, rework, quality, and review effort.
- Track whether the next team in the process receives more usable work.
- Ask staff where the remaining delays occur.
- Check whether the benefit continues beyond the first enthusiastic week.
Start with one workflow and a manageable sample. For example, compare customer-response preparation from the initial request through review and delivery. Keep the task mix comparable and record unusual cases so they do not distort the result.
Use adoption data to ask better questions
External research can help you choose areas to investigate. It should not become a performance target copied into every department.
A team with lower usage may face unsuitable tasks, unclear permissions, or a tool that does not fit its needs. Interview the people doing the work before treating low adoption as a training problem.
Decide in advance what would justify expanding the pilot. A sensible goal might combine faster turnaround with an acceptable error rate and manageable review effort. The outcome you want is better work delivered, not a busier prompt box.
Explore Guava’s AI and Copilot collection for more ideas on putting AI to work thoughtfully.
