Analytical procedures fail when the auditor compares this year with last year, sees a 12 percent rise and writes "reasonable". A useful procedure starts earlier: the auditor forms an independent expectation of what the number should be, decides how far a difference can go before it needs work, and only then compares. This post shows the sequence with an example.
Last reviewed: 17 October 2026. Facts checked. We could not open the ICAI text of SA 520; our earlier searches returned only secondary study material on the international standard it is based on. The steps below follow that material and general practice, so confirm the wording against SA 520 before quoting. No thresholds are given because we cannot source one; the auditor sets it using judgement and materiality.
Where it fits
Analytical procedures can be used in planning, as substantive tests and in the final review. The overview is in our SA 520 guide. This post is about the substantive use, where the procedure is meant to give evidence on a balance or a total.
The four steps
1. Develop the expectation first
Use data that is independent of the number being tested. Good sources are volumes from operations, price lists, headcount, contracts, rent agreements, loan balances and interest rates, or industry data. Avoid starting from the ledger figure and working backwards.
2. Decide the acceptable difference
Set how large a difference you will accept without further work. It should be well below materiality, so that several small unexplained differences do not add up. Record the reasoning.
3. Compare and investigate
Calculate the difference. If it is above your threshold, ask management why, then test the explanation. An explanation not backed by evidence is only a lead.
4. Conclude
State whether the expectation supports the balance, and what further work was done on any unexplained amount.
Precision depends on the data
The more disaggregated and reliable the data, the sharper the expectation. A monthly view is sharper than an annual one. A comparison by product beats a single total. Predictability matters too: rent and interest are easy to predict, discretionary expenses are not.
The inputs also need to be reliable. If you use headcount from HR to estimate payroll, test that headcount; see how to test reports in client-produced data and audit evidence.
Worked example: interest expense
An auditor tests interest on borrowings for a company with three term loans and a cash credit account.
Step 1, expectation. From loan statements, the opening balance, repayments by month and the sanctioned interest rates, the auditor computes interest for each loan. For the cash credit, the average monthly outstanding from the bank statement is multiplied by the rate.
| Facility | Expected interest (Rs lakh) |
|---|---|
| Term loan A | 42.0 |
| Term loan B | 18.5 |
| Term loan C | 9.0 |
| Cash credit | 31.0 |
| Total | 100.5 |
Step 2, threshold. The auditor sets the acceptable difference at Rs 3 lakh, a fraction of materiality, recorded with the reason.
Step 3, comparison. Interest per the ledger is Rs 109.2 lakh. The difference is Rs 8.7 lakh, above the threshold.
Investigation. Management says part of it is penal interest for a delayed instalment on loan B. The auditor asks for the bank's debit advice. It supports Rs 5.4 lakh. The remaining Rs 3.3 lakh is unexplained, and the auditor finds it is a processing fee booked as interest, which belongs in finance costs but should be presented separately.
Step 4, conclusion. After reclassifying, the difference is within the threshold. The file shows the expectation, the sources, the threshold, the explanation and the evidence.
All figures here are invented for illustration.
Mistakes to avoid
- Expectation built from the same ledger the number comes from.
- Threshold set after seeing the difference.
- Accepting "business growth" without a volume or price driver.
- Not recording the source of the data.
- Using an analytical procedure where the risk is high and the data is poor. A test of details may be better.
Where tools help
A spreadsheet or the analytical procedure expectation calculator can compute differences quickly. If you use an AI model to draft explanations, read our note on choosing an AI model for variance commentary. The expectation and the judgement stay with the auditor.
Frequently Asked Questions
Is comparing with last year an analytical procedure?
It is a start. On its own it is rarely a precise enough expectation for a substantive test.
Can analytical procedures replace tests of details?
Sometimes, for areas with predictable relationships and reliable data. Not for high-risk areas with poor data.
How is the threshold set?
By judgement, with reference to materiality and the risk. Record the reasoning.
Who explains the differences?
Management gives explanations, and the auditor corroborates them with evidence.
Related Articles
Statutory facts on this page are checked against their sources, and the page says where it relied on secondary reporting. How we verify · Report an error