By Ammar Ashour
An AI tool can draft a variance explanation before your coffee cools. Someone still has to open the ledger, check the period, and decide whether that explanation makes sense. That moment at the desk is where AI training earns its keep.
In my qualitative study of AI reskilling in accounting, participants described changes in three areas: how they used their time, how they felt about their work and future skills, and how they handled client data and responsibility for AI-assisted work. Their accounts reflect experiences in the study. They cannot establish typical time savings or predict employment outcomes across the profession.
Measure the time it takes to reach a reliable answer
Participants described spending less time on routine tasks and redirecting some of that time to review, analysis, and client work. The benefit varied. When a suggested answer needed substantial checking or correction, the initial time savings could shrink. Participants’ accounts also pointed to a broader definition of productivity that included accuracy and judgment.
Take a hypothetical month-end close. An AI tool drafts an explanation for why expenses increased. It attributes the change to rising vendor prices, but the ledger shows a quarterly insurance payment. The reviewer catches the error before the client sees it. The time to open the ledger, correct the explanation, and approve the final version belongs in the workflow measurement.
For a small pilot, compare similar tasks before and after training. Record drafting time, review and correction time, and whether the final output passed review. Use the same quality standard both times. See the financial summary review checklist.
Keep junior accountants in the review
Participants described relief from repetitive work and opportunities to do more engaging work. They also expressed concern about keeping their skills current and about what junior accountants could miss as routine assignments change.
Picture a new accountant learning why two entries with nearly identical descriptions belong in different accounts. The explanation might depend on the invoice, a conversation with the client, or what happened in the previous month. If the job becomes clicking through a list of suggested categories, the junior accountant may miss the chance to build that judgment.
Give newer staff a few real exceptions to work through with a reviewer, using data they are authorized to access. Ask them to explain what they checked, what would change their decision, and when they would escalate. That conversation can also show a manager where more coaching is needed. Read more about training entry-level accountants as workflows change.
Rehearse the decisions your AI policy covers
Participants most often discussed responsible AI use in terms of protecting client information and reviewing outputs. The study also surfaced a gap between having rules and getting practice applying them.
Use a mock client file to rehearse a decision before it comes up during a busy close. Ask the team which tool and data fields are approved, how to handle an identifier that should stay out of a prompt, how to verify the response, and who makes the final call. NIST’s generative AI risk profile identifies privacy and plausible but incorrect output as risks that organizations can address through their own controls. NIST Generative AI Profile.
At your next team meeting, pick one recurring task and schedule a short practice review. Ask the person doing the work to show where the AI helped, what they checked, and what they would do when the answer looks right but the evidence is missing.
