You can buy a tool in an afternoon. Getting people to use it well takes longer, and it is where most of the value is won or lost. Two well-known controlled studies give a useful picture of what to expect, with some important limits.
Two studies worth knowing
Customer support: a 14% average gain, 34% for newer staff
Economists Erik Brynjolfsson, Danielle Li and Lindsey Raymond studied 5,179 customer support agents after a company introduced a generative AI conversational assistant. Access increased productivity, measured by issues resolved per hour, by 14% on average. Novice and low-skilled workers improved by 34%, while experienced and highly skilled workers saw minimal impact. The authors also found evidence that the AI spread the practices of more able workers, and that it improved customer sentiment and employee retention.[1]
Professional writing: faster and better
In an experiment by Shakked Noy and Whitney Zhang published in Science, 453 college-educated professionals completed occupation-specific writing tasks. Those given access to ChatGPT (version 3.5) finished about 40% faster, and independent evaluators rated the quality of their output 18% higher.[2]
What not to conclude
- These are specific tasks. A support chat or a 20 to 30 minute writing exercise is not your whole job.
- The tools were earlier versions. Current tools differ, and so will your results.
- They measure individuals in a controlled setting. Real workplaces have handoffs, exceptions and habits.
- Averages hide spread. The support study found most of the gain among less experienced staff.
- Neither study proves a revenue effect. Time saved is not profit unless it is used well.
What you can reasonably take from them
- The biggest gains may go to newer or less experienced staff. AI can carry some of the knowledge your best people have. If your business depends on a few veterans, capturing their methods in prompts, templates and checklists can lift everyone else.
- Drafting and routine communication are strong candidates. Writing is where the evidence is clearest.
- Experienced staff may see smaller gains. Do not be surprised if your best people are unimpressed. They may be right for their own work, and their role may be reviewing what AI produces for others.
- Quality can improve along with speed. But only if people review output rather than accepting it blindly.
Why adoption stalls
The CPA Practice Advisor article on adoption argues that the barriers are mostly behavioral: fear that AI will be used to cut jobs ("a trust problem before it becomes a skills problem"), concern that it devalues professional expertise, ambiguity about what use is permitted and a habit of measuring training attendance instead of changes in work.[3] Those apply well beyond accounting.
A rollout that works with people
- Say what it is for. If you will not reduce headcount because of it, say so plainly. If you might, be honest about that too.
- Pick two or three willing champions and give them time to experiment on real work.
- Start with one task per role. For an office manager, maybe meeting notes. For an estimator, maybe a first draft of scope text.
- Write down what works. Collect prompts, templates and examples in a shared place.
- Set the rules. An AI policy tells people what data they can use and what must be reviewed.
- Measure the work, not the logins. Minutes per task, turnaround time, error rates.
- Decide where saved time goes. Homebase found 87% of adopters said AI saved them time, and 57% put the time toward growing the business while 53% used it for neglected tasks.[4] Those are outcomes you can plan for.
Training that sticks
Short sessions on real tasks beat one long workshop. Have people bring a task from this week, do it with the tool, compare it with the old way and write down the steps. Follow up two weeks later and ask what they kept using and why. Whatever people stopped using tells you something about the tool, the task or the training.
Treat early errors as information. Create a way to report them without blame, review the common ones monthly and adjust your guidance. The goal is a team that trusts the tool exactly as much as it has earned.
Sources
- Brynjolfsson, Li and Raymond, "Generative AI at Work," NBER Working Paper 31161
- MIT Economics, summary of Noy and Zhang, Science (2023)
- CPA Practice Advisor, "Why CPA Firms Need a Behavioral AI Adoption Playbook" (Sept. 2026)
- Homebase, "2026 Main Street AI Gap Report"
Figures are quoted from the sources above and were reviewed in October 2026. Where a source is a secondary summary of other research, we say so in the text. Survey results reflect each survey's own sample and definitions. Examples labeled as illustrations use made-up numbers to show the method. Nothing here is a forecast or a guarantee of results.
Want this worked out for your business?
Tell us what is wasting your time.