Keeping pace with AI, one useful tool at a time
“The important thing is not to stop questioning. Curiosity has its own reason for existing.” – Albert Einstein
KEEPING up with artificial intelligence (AI) can feel like a second job. Just as I begin to understand one tool, another arrives with unfamiliar terminology and fresh promises. Finding time to make sense of it all alongside accounting work is a challenge.
But the reading becomes worthwhile when a development addresses a familiar frustration: explaining the same procedure repeatedly, sorting through inconsistent records or preparing a report that follows the same steps every month. Technology becomes interesting when we see what it might help us do.
A Mercor study published on Oct. 1 offers one example. Across four simplified month-end accounting exercises, Claude Opus 5 met every grading criterion in 20 attempts, each completed in under 10 minutes. Models from just 18 months earlier had scored below the participating junior accountants’ average.
The scope matters. These were clearly defined exercises, not a test of everything an accountant does. They could not capture the full demands of client communication, professional judgment or understanding a business over time.
Still, the findings give us reason to revisit earlier impressions. A task that resisted automation a year ago may deserve another look today.
Consider a monthly expense review. The arithmetic may be straightforward, but getting the records ready often takes longer. Descriptions vary, supporting documents come in different formats, and some transactions require closer examination before they can be classified.
Given approved expense categories, an AI model could suggest classifications and flag uncertain items for the accountant’s review.
Specialized tools are emerging for such work.
In September, TypeSafe introduced Jev, a model that chooses among predefined options and returns probability estimates. An accounting team could explore using it to classify documents or route transactions to reviewers. Its estimates, however, are assessments, not proof that a choice is correct.
Consistency
The next challenge is consistency. A monthly review should follow an established procedure, apply the required checks and produce a report in the agreed format.
Reusable AI ”skills” can help preserve that procedure. These packages contain instructions, reference materials and, when needed, supporting programs. For an expense review, a skill could include approved categories, examples of difficult classifications and directions for reporting exceptions.
As recurring problems emerge, the accountant can refine the instructions. Knowledge scattered across conversations can become a shared resource that colleagues can inspect, update and use.
Each part of the process needs a clear role. Conventional software calculates totals and compares expenses with budgets. The model interprets descriptions and suggests classifications. The accountant reviews unusual items and approves their treatment.
The value lies in fitting AI into a process we can check. Before relying on its output, we should compare it with previously reviewed work and examine the errors.
We must also protect confidential information and establish who approves the results. Any time saved should be weighed against the effort required to correct mistakes and maintain the process.
One anonymized historical expense file, used with an approved tool, could be a sensible starting point. Ask the model to suggest classifications, compare them with the reviewed answers and record where it goes wrong. That exercise can reveal more about its usefulness than an impressive demonstration.
I expect many of today’s practical limitations to ease as AI improves. Some tasks we have long accepted as slow or cumbersome may take less time, giving us more room to understand the business and exercise judgment. Greater accuracy, however, will still depend on sound procedures and careful review.
That prospect gives me a reason to keep learning. Becoming a consummate accountant means using the capacity we gain to see further into the numbers, ask better questions and bring insight to a decision while there is still time to shape it.
---------------------------------------------------------------------------------------------------------------------------
Manuel Guilius A. Pamorca is a member of the Association of Certified Public Accountants in Public Practice (Acpapp) Media Affairs.
Comments