Substantive analytical procedures that hold up
When analytical procedures count as substantive audit evidence under ISA 520, how to build an expectation, set a threshold and chase differences.
By Achekam · · 5 min read · For Accountants & audit firms in South Africa, Nigeria, Kenya
"Salaries up in line with inflation. No further work." Every reviewer has met that line on a payroll paper, and every inspector has written it up. It compares this year with last year, accepts the client's explanation and calls the result evidence. It is not. A substantive analytical procedure is a test with a design: an expectation built from data the auditor trusts, a difference decided in advance, and a follow-through when the actual figure falls outside it.
Done properly, analytics can be the most efficient evidence on a file, particularly for large volumes of predictable transactions such as payroll, rent, interest and depreciation. Done loosely, they are the weakest part of it. This article sets out where the line falls.
Analytics at planning, as evidence and at completion
Analytical procedures appear three times in an audit, and only one of those uses is substantive.
| Stage | Standard | Purpose | Evidence for an assertion? |
|---|---|---|---|
| Planning | ISA 315 | Spot unusual movements that point to risk | No |
| Fieldwork | ISA 520 | Test a balance or class of transactions | Yes, if designed as a test |
| Completion | ISA 520 | Check the statements make sense as a whole | No, it is an overall review |
The planning comparison and the substantive test can look alike on paper, which is how the one-line conclusion above slips through. The difference is precision. A planning analytic asks "does anything look odd?"; a substantive analytic asks "is this balance materially right?", and it needs an expectation sharp enough to answer.
Analytics also suit some risks better than others. ISA 330 requires tests of details as part of the response when the approach to a significant risk consists only of substantive procedures, so an analytic on its own will not carry a significant risk such as revenue recognition. ACCA's article on responses to assessed risks explains how auditors respond to assessed risks of material misstatement under ISA 330 using tests of controls and substantive procedures, and is a good refresher on where analytics fit among them.
Building an independent expectation
The expectation is the test. Build it from data that does not come from the figure you are testing, and judge that data before you use it:
- Source. Information from outside the client, or from a part of the client independent of the accounting function, is more reliable than the ledger itself.
- Controls. If the data is the client's own, has it been subject to controls, or tested?
- Disaggregation. A monthly expectation by location or product is far more precise than one annual total.
- Relationships. The best expectations rest on a real driver: headcount and pay rates for salaries, the lease schedule and occupancy for rent, average borrowings and the agreed rate for interest.
A worked example in words: for a manufacturer's factory wages, the senior takes the monthly headcount from the HR system, the pay rates in the union agreement and the month the annual increase took effect, and calculates expected wages for each month. Overtime is modelled separately from the time-and-attendance system. The resulting expectation is compared, month by month, with the wages accounts in the trial balance. A month that departs from it stands out in a way an annual comparison would hide.
Deciding what difference is acceptable
ISA 520 requires the auditor to decide, before comparing, how large a difference can be accepted without further investigation. Decide it first and write it down; a threshold chosen after seeing the result is not a threshold.
The amount follows from materiality and from how much assurance the analytic is meant to give. In Achekam, staff propose overall, performance and clearly-trivial materiality from a benchmark with a written rationale, and the engagement partner approves it, never the preparer. Achekam suggests no percentage, either for materiality or for the analytic threshold, so the judgement and its reasoning belong on the paper. If materiality is later revised, the new approval replaces the old one and the earlier figure is kept, so revisit any threshold built on it.
Investigating differences with the client
When the actual figure falls outside the threshold, the standard asks for two things: inquiry of management, and evidence that corroborates the answer. The second is where files fail. "Higher overtime in the peak season" is an explanation, not evidence. Corroboration means looking at the overtime records, the approval of those hours and the payroll reports that support them.
If management cannot explain the difference, or the explanation does not hold, perform other procedures. A difference that cannot be explained may be a misstatement.
Achekam's work items keep this honest. Each work item lists the procedures to perform and the documents required, and a step is marked done only with a written response and its documents, or set aside with a reason. Staff can ask AI to draft a procedure's response or a conclusion on a working paper, and the draft goes into the file only when a person saves it. Treat any such draft as wording, not as corroboration: the evidence for the explanation still has to be attached by someone who looked at it.
Workbooks refreshed from the trial balance
An analytic is only as good as the actual figure it is compared with. A working paper in Achekam can have a spreadsheet workbook whose figures are refreshed from the engagement's trial balance, and any figure the refresh could not read is listed, so a comparison against a stale draft is visible rather than silent. Refreshing is refused on a locked file or on a paper under review, which stops the numbers moving while a reviewer reads them.
Lead schedules are also built from the engagement's trial balance, with each figure traceable to its lines. If an unexplained difference turns out to be an error, staff record an audit adjustment on the affected line with a reason, and the finding is raised from the working paper. How workbooks and lead schedules link together is covered in workbooks and lead schedules from the trial balance, and the completion-stage review in the final analytical review.
A test for your next file
Pick one area where the team currently relies on a year-on-year comparison. Rebuild it with a driver-based expectation, a threshold recorded before the comparison, and corroborated explanations for every month outside it. If the rebuilt paper takes longer than tests of details would have, analytics were the wrong tool for that area; if it does not, you have a stronger paper for no extra cost.
General guidance only, not advice on your own circumstances. Tax and payroll rules change: check the current position with the authority or your adviser.
