Yesterday’s article covered what ISA 530 requires. Today we move from the standard to the desk: how do you actually choose a sampling method, size the sample, pick the items and evaluate what you find? Whether you are testing payment vouchers at a parastatal, loan files at a bank or grant expenditure at an NGO, the steps below will help you produce a defensible sample.
Sampling versus selecting specific items
Before choosing a method, be clear about what audit sampling is. Under ISA 530, audit sampling means applying a procedure to less than 100% of a population so that every sampling unit has a chance of being selected, which lets you draw a conclusion about the whole population. If every item cannot be selected, you are not sampling, and you cannot project your results.
That distinction matters because many auditors label any partial testing as “sampling”. Picking the ten largest payments, every journal posted after year-end, or all transactions with a related party is a perfectly valid approach, but it is targeted selection of specific items, not sampling. It tells you about the items you tested and nothing about the rest. A good plan often combines both: test the high-value and high-risk items in full, then sample the remainder.
The four methods you will use most
1. Simple random sampling
Every item has an equal chance of selection, typically using a random number generator. It is the cleanest statistical method and works well when the population is fairly homogeneous, such as a list of fixed asset tags or staff files. In Excel, add a helper column with =RAND(), copy and paste it as values (so it stops recalculating), sort by that column and take the first n rows. Keep the file so the selection can be re-performed.
2. Systematic (interval) sampling
Divide the population size by the sample size to get an interval, choose a random starting point within the first interval, then select every nth item. For example, 2,400 receipts with a sample of 60 gives an interval of 40; a random start of 17 selects items 17, 57, 97 and so on. It is quick and spreads the sample across the period. The risk is a hidden pattern in the population: if a clerk batches every 40th receipt for a particular cashier, your sample could be badly biased. Check the ordering of the population before using it.
3. Monetary unit sampling (MUS)
MUS treats each individual dollar (or ZiG) as the sampling unit, so larger balances are more likely to be selected. It suits substantive tests for overstatement, such as receivables, payables confirmations or expenditure. A common approach is:
- Sampling interval = tolerable misstatement ÷ reliability (confidence) factor.
- Sample size = population value ÷ sampling interval.
- Any item larger than the interval will always be selected, so many teams pull these out and test them in full.
Using a commonly quoted factor of 3.0 for 95% confidence with no misstatements expected, a population of USD 1,500,000 and tolerable misstatement of USD 75,000 gives an interval of USD 25,000 and a sample of 60 items. MUS is less effective at finding understatements, because an understated or omitted balance has fewer dollars to be picked.
4. Judgemental (non-statistical) sampling
The auditor selects items using professional judgement, often a haphazard selection with no conscious bias. It is widely used in internal audit and is acceptable under ISA 530 provided the sample is still expected to be representative. The weakness is that results cannot be measured statistically, so the evaluation depends on how well you documented why the sample was appropriate. Block selection (for example, “all March vouchers”) is rarely suitable for drawing conclusions about a full year.
A step-by-step sampling plan
- State the objective. What assertion or control are you testing, and what counts as a deviation or misstatement? Define this before you look at a single item.
- Define and validate the population. Reconcile the listing to the general ledger or system totals. A sample from an incomplete population proves nothing about the missing part.
- Stratify and isolate key items. Separate high-value, unusual or high-risk items for full testing. Split mixed-currency records into USD and ZiG strata rather than combining amounts recorded at different rates.
- Choose the method. Controls testing usually suits random or systematic selection; balance testing for overstatement usually suits MUS.
- Set the sample size. Base it on the risk assessment, tolerable rate or misstatement, expected errors and the confidence you need. Higher risk means a larger sample.
- Select and perform the tests. If a selected document cannot be found, treat it as a deviation; do not quietly swap it for another item.
- Evaluate. Investigate the nature and cause of each exception, project misstatements to the population, and compare the result with the tolerable amount.
- Conclude and document. Record whether the sample provides a reasonable basis for your conclusion, and what you did if it did not.
Template: sampling plan summary
Use a one-page summary like the one below at the front of each tested area.
| Field | Example entry |
|---|---|
| Audit area / objective | Payments: verify that expenditure is properly authorised and supported |
| Population and source | All payment vouchers, 1 Jan – 30 Jun 2026, ERP extract |
| Population reconciled to | General ledger expenditure total (difference: nil) |
| Sampling unit | Individual payment voucher |
| Items tested in full | All payments above USD 25,000 (14 items) |
| Method for remainder | Systematic, random start 17, interval 40 |
| Sample size and basis | 60, based on high control risk and zero expected deviations |
| Definition of deviation | Missing approval, approval above delegated limit, or no supporting invoice |
| Results | 3 deviations (5%) |
| Conclusion | Deviation rate exceeds tolerable rate of 3%; control not relied on; extend substantive testing |
Common mistakes to avoid
- Projecting results from a targeted selection. Testing the 20 largest invoices and concluding “the population is fairly stated” is not supported.
- Never reconciling the population. Extracts from payroll or ERP systems are often filtered or incomplete.
- Replacing awkward items. Swapping a missing voucher for an easier one destroys the sample’s integrity.
- Dismissing errors as “isolated”. ISA 530 only allows an anomaly in extremely rare circumstances and with a high degree of certainty that it is not representative. Most errors must be projected.
- Weak documentation. If a reviewer cannot re-perform the selection, the work is hard to rely on.
Where populations run into hundreds of thousands of rows, such as mobile money transactions or bank ledgers, Excel can slow down. A dedicated tool like ProDataTool can help you summarise, stratify and filter large datasets before you draw your sample.
Conclusion
Good sampling is less about formulas and more about discipline: define the objective, prove the population is complete, separate the items that need full testing, choose a method that fits the test, and document every decision. Do that consistently and your conclusions will hold up.
Keeping sampling plans, working papers, findings and evidence organised across many engagements is where many audit teams struggle. Auditors Track GRC helps internal audit departments across Africa digitise their work, with risk assessments, audit reviews, findings and report writing, evidence attachments, and built-in libraries of controls, audits, working papers and risks. Dashboards by department and engagement type give audit leaders a clear view of progress. Book a free demo of Auditors Track GRC to see how it fits your audit methodology.