When AI saves time, who does the checking?

AI can get a report onto your desk sooner. To see whether it saves the business time, count the work needed to check and correct it before it is ready to use.

At NoVacancy Sydney in September 2026, I heard Scott Boyes, CEO of Trilogy Hotels, talk about using AI to prepare his month-end board report and how much time and work it saved him. It made me think of conversations we’ve had with managers and team members from many organisations about the time they spend checking AI-generated work that staff and clients send them.

When someone tells me AI saved them time on a report, I want to know what happened after they sent it. Did the person receiving it have what they needed, or did they have to go back to the person who wrote it?

Whether you use AI to draft an email, analyse data or prepare a report, count the work needed to reach a finished result. That includes your own checking and corrections, as well as work someone else has to do.

What happens after the draft is sent?

Say someone uses AI to draft the commentary for a monthly sales report. The draft is ready sooner. It says sales fell because a promotion ended. The spreadsheet shows when sales fell. The manager still has to ask what connects the two.

The employee checks the promotion dates and looks at which products accounted for the fall. They revise the commentary and send it back. The manager then checks the changes. Here’s how the staff time might add up.

Illustrative example: staff time for one monthly sales report
Work Usual process AI-assisted process
Employee gathers figures and drafts the report 30 min 15 min
Manager checks the report, follows up and reviews changes 10 min 20 min
Employee makes corrections 5 min 10 min
Total staff time 45 min 45 min

Both reports take 45 minutes of staff time. The employee still comes out 10 minutes ahead overall. The manager has to find an extra 10 minutes. If that manager is also the person everyone else is waiting on, the change can affect when the report gets out.

For the next report, the employee could link the figures to their source and flag any explanation they still need to check. If the cause of the fall is unclear, the commentary can say what the figures show and leave the cause open. That gives the manager a clearer starting point. Time both sides again to see whether the extra preparation makes review quicker.

When a client brings an AI-assisted summary

A client might use AI to turn emails and notes into a timeline before meeting their lawyer. That can help the lawyer get across what happened. If the timeline leaves out where key dates came from, someone may still have to go back through the emails or ask the client for the records.

In the 2026 Victorian Lawyer Census, conducted in Victoria, Australia, 46% of respondents to a question about client AI use reported that it increased the time and effort needed to deliver services. Another 19% reported a reduction. The overall response rate for the voluntary census was 7.2%, and the findings reflect Victorian practitioners’ own observations. Victorian Legal Services Board and Commissioner, The Evolving Use of Generative AI, pp. 17 and 33

The same issue arises for accountants. A client may send an AI-generated summary instead of a shoebox full of receipts. It looks more organised, but can the figures be traced back to the records?

If clients send you summaries like this, ask for the records behind the details you’ll rely on. Then pay attention to what the summary helps you find and what you still have to check. That gives you a basis for telling the next client what would be useful to send.

Practical Takeaway

Pick one recurring task where you use AI. Compare reports or tasks of similar scope and use the same review standard. If you don’t know how long the usual process takes, time the next typical run.

  • Staff time. Record how long each person spends preparing the material, drafting, checking and correcting it. Include time spent getting the material ready for AI, checking its output and answering follow-up questions.
  • Turnaround. Record when the job started, when the finished work was ready to use and whether it met the deadline.
  • Quality. Check whether both versions meet the same standard. For a sales report, the figures should match their source and the commentary should have evidence behind it.

Treat the first comparison as a trial. If one month had more to report on or the figures were harder to pull together, run the comparison again on more similar work.

Repeat the approach if the finished work meets the same standard, arrives when it’s needed and takes less staff time, with a workload the reviewer can manage.

Adjust and retest if extra checking used up the drafting time you saved or waiting for review made the work late. If the reviewer had to chase figures, send the sources with the next draft. If the work sat waiting for review, look at when the reviewer can pick it up.

Keep the usual process for now if quality drops or the person checking the work can’t absorb the extra time.

If the new process takes the same staff time but delivers a better report or gets it ready sooner, record that benefit. It may be worth keeping the change for that reason.

Related reading
AI licence counts and policies don’t tell you whether AI is working in your organisation