Once a well-prepared working paper is in hand, what else does the accountant need to check?
In Brief
AI quickly analyses the contracts it receives. Experienced accountants can identify company circumstances missing from those documents and consider the relevant facts and standards together. Accepting an AI-generated result also requires an explanation of the basis for agreeing with it.
About five minutes is enough to turn several contracts into a draft review working paper. The change described in the previous article makes work easier, but it also leaves a question. What should accountants do to avoid being replaced by AI?
Using AI in practice makes it easier to see which tasks it can handle and which aspects need further attention. Those differences help explain the judgements accountants can contribute.
1. Contract analysis and Excel development
Reviewing the structuring of a special purpose company (SPC), or a share acquisition or disposal, often means reading several contracts together. Even a careful reader can miss an important phrase. In one case I experienced, AI organised its review of ten complex contracts into a standard working paper in about five minutes. Checking and revising the output required additional time, but the speed of producing a first draft was difficult to compare with the old process.
I see a similar change in building Excel templates. It takes far less time to link worksheets and set up calculations. I still find accountants more adaptable when tailoring a file to the work at hand, but AI has a clear advantage in the speed of building it.
AI is beginning to perform tasks quickly that once required considerable time and skill. Reading documents quickly or being proficient in Excel may no longer be enough to explain an accountant’s competitive advantage. The question becomes what I can contribute once the output is ready.
2. Non-public information and informed judgement
There is an area where I still find experienced accountants stronger : finding and using a company’s non-public information, and weighing the relevant facts and standards together. At least in the work I have experienced, accountants retain an advantage here.
Not all the information needed for a judgement is public. Earlier interviews with company staff, internal analyses and the background to transactions examined in previous audits often remain in the company’s or audit team’s records. Even a good understanding of the standards and comparable cases can lead to an unsuitable conclusion if this information is missed.
Suppose the team member responsible for an account is unaware of an earlier interview and provides AI only with the contracts currently available. AI may produce a logical working paper based on those contracts. But if a fact established in that interview also affects the current judgement, the analysis is incomplete without it. If the relevant records have not been provided or made accessible to AI, there is little reason to expect the output to reflect them.
An in-charge or reviewer familiar with the company’s history can ask further questions. How did the company explain its reasons for entering the transaction? Do the conditions that mattered to the earlier judgement still apply? Finding records that the preparer did not know existed may change the conclusion.
The accountant’s strength lies in connecting that information to the current issue. Does the contract wording reflect how the transaction works in practice? Why do earlier explanations differ from the current records, and are those differences significant under the applicable standard? Considering the materials together helps determine what else to check. Experience does not make memory or earlier judgements reliable by default, however. The original records need to be reread and their continuing relevance assessed. If AI’s judgement remains unsuitable even after all the necessary information has been provided, the reasoning used to apply the standard to the facts also needs review.
As more past working papers and internal records become accessible to AI, the advantage that comes from finding information may narrow. Accountants need to keep developing their experience as that changes. Judging whether a record applies to the current transaction, and deciding which conflicting explanations need further investigation, will still matter.
3. Review issues and further work
AI also helps with responses to comments from an in-charge or project manager (PM). Finding relevant material and revising documents has become easier. Before drafting a response, though, the preparer needs to understand the reviewer’s concern.
The next step depends on whether the explanation is incomplete, necessary evidence has not been checked, or the conclusion itself needs reconsideration.
Example
Suppose a contract review working paper receives this comment
“The contract can be terminated early. Please establish why the termination terms do not affect the conclusion.”
The preparer needs to check the original contract to establish who can terminate, under what conditions and with what settlement obligations. Separate agreements that modify those terms also need attention. The reviewer is asking whether the original conclusion still holds once the relevant terms are taken into account.
AI can organise the clauses and suggest issues to examine. The preparer must establish whether that analysis fits the actual contract and the company’s circumstances. Discovering a previously missed condition may require a change in the conclusion, rather than simply a fuller response.
Relevant standards — Professional scepticism and judgement
For audits subject to US PCAOB standards, AS 1000 paragraphs .11–.12 require critical evaluation of audit evidence and the application of training, knowledge and experience in making judgements. Both supporting and contradictory information must be considered. Using AI to draft a review response should not become a search only for explanations that support the existing conclusion.
4. Evidence and documentation of judgement
My answer, for now, to the question of avoiding replacement by AI is this: I should be able to explain the judgements I made, even when AI helped me do the work.
That does not always mean finding an error or changing the conclusion. If I have checked that the necessary information is included and that the standard has been applied appropriately to the company’s circumstances, agreeing with AI can itself be the outcome of a review.
The following hypothetical examples show what an accountant might check and how the findings could be reflected in a working paper.
| Ai’s Draft | What the accountant established | What to reflect in the working paper |
|---|---|---|
| SPC contract review After reviewing the loan agreement provided, AI stated that Company A had no obligation under a funding shortfall agreement. | A prior working paper mentioned a separate agreement. Once obtained, it showed that A had to cover any shortfall in the SPC’s debt repayment funds up to a specified limit. | Correct the statement that no obligation exists. Document its triggering conditions and funding limit, and assess the effect on the original conclusion. |
| Application of a comparable case AI cited a case in which one company made key decisions alone and proposed the same conclusion | The actual investment agreement required another investor’s prior consent for key decisions. | Revise the assumption that the company decides alone. Identify the matters requiring consent and each party’s rights, then reassess whether the case’s conclusion applies. |
| Excel template A monthly sales file calculated December revenue at KRW 1 billion. | The source data included KRW 50 million of sales on 31 December, but the formula’s date range ended on 30 December. | Correct December revenue to KRW 1.05 billion. Check other months using the same formula and any linked tables for related effects. |
A completed working paper should retain a record of that process. The next reviewer needs to understand what additional material was checked and why the conclusion was retained or revised.
Key Takeaways
- Experienced accountants demonstrate their strength by finding non-public company information and transaction background, then connecting it to the current issue.
- A response to a review comment should address the concern raised and assess how newly established facts affect the conclusion.
- Even when an AI conclusion is retained, the basis for accepting it should be explained and documented.
These abilities develop gradually through reading contracts, consulting standards and receiving review comments. An accountant who already has that experience starts from a different position from a new associate learning through AI-generated working papers from the outset.
When AI writes the working paper and drafts responses to review comments, where will new accountants develop that judgement?
The next article considers that question.
References
- Public Company Accounting Oversight Board (PCAOB), AS 1000 — General Responsibilities of the Auditor in Conducting an Audit, paragraphs .11–.12.