Sampling documents

A building-materials wholesale base issues around 3,600 dispatch notes in a quarter. An auditor cannot read them all, and does not need to: a well-chosen sample gives a reliable picture of the whole flow. The hard part lies elsewhere: how to choose the sample, how many documents are enough, and what to do when an error turns up. Without answers to those three questions, "we tested on a sample basis" is an empty phrase.

Define the population first

A sample is drawn from a population, so first make sure the population itself is complete. The auditor needs not a tidy list prepared by the storekeeper or department head, but a report pulled straight from the 1C journal. Check the number sequence too: if 1,485 follows 1,482, where are the three in between? Deleted and cancelled documents go on a separate list, because a "missing" number often leads to the most interesting finding.

Which direction to test in

The direction depends on the objective. For "has every shipment actually been recorded?", the sample is taken from physical documents (the gate log, the pick list) and traced into 1C. For "is every recorded shipment real?", you go the other way: from the 1C entry back to the document and the warehouse. If you test in only one direction, a problem on the other side stays hidden.

Selection methods

• Random selection: every document has the same chance of being chosen. In Excel the numbers come from the RANDBETWEEN formula. The most impartial method.
• Systematic selection: every n-th document is taken. If you need 30 out of 3,600, that is every 120th; the starting point is picked at random.
• Targeted selection: the largest amounts, documents raised at weekends and at month end, those corrected later, those issued to one customer in a row.
• Stratified selection: the population is split into groups (for example, below and above 5 million so'm), and each group is sampled separately.

In practice the methods are combined: the main part is random, with risky documents added on top. The result of the random part can be projected onto the whole population; the targeted part cannot, and the report shows it separately.

How many documents are enough

Statistical tables exist for exact calculations, but in small and mid-sized warehouses a practical rule of thumb is more common: the more often a control operates, the larger the sample.

1. The control operates every day (every delivery note): 25–40 documents.
2. Once a week (weekly reconciliation): 5–15 documents.
3. Once a month (monthly report): 2–5 documents.

If the risk is high or errors were found last time, the sample grows. For the base in our example, 30 random and 10 targeted documents would be reasonable.

What is tested in each document

The list of attributes is written in advance so every document is tested the same way:

• a signature and date are present, and the date matches the date in 1C;
• the quantity matches what physically left the warehouse (pick list, gate log);
• the price comes from the valid price list, and any discount is authorised;
• the document was posted in 1C on the day of dispatch or within a day or two.

The result goes into a table: document number in the row, attribute in the column, "pass" or "fail" in the cell.

What to do when an error appears

Do not write an error off as chance. First find out whether it is a one-off or has a systemic cause. If one of 30 documents lacks a signature, ask for an explanation and add another 15–20 documents. Two or more errors suggest the control is not working. If the random part shows 2 errors, that is roughly 7 percent; projected onto 3,600 documents, it would be on the order of 240 similar ones. Report this as an order of magnitude, not as an exact figure.

Common mistakes

• The list is prepared by the party being audited.
• Only documents from the last week are taken, not the whole period.
• The choice is made from a "model" folder recommended by an employee.
• When an error is found, the sample is not extended; it is simply noted as a "one-off".

Conclusion

Sampling saves time, but it is reliable only when the rule is clear. OmborAudit selects documents under a rule set in advance and keeps the selection method in its working papers, so every conclusion can be rechecked.