# Audit Sampling Bank Transactions: Size and Selection - Author: PreciseBankFlow (https://www.precisebankflow.com) - Published: 2025-02-23 - Updated: 2025-02-23 - Category: Bank statements in an audit - Canonical: https://www.precisebankflow.com/blog/audit-sampling-bank-transactions > Audit sampling bank transactions: choosing the population, sample size, random or monetary unit selection, and evaluating the errors you find. Audit sampling of bank transactions means testing fewer than all the statement lines or ledger entries in a population and drawing a conclusion about the whole. You define the population and its direction, set a sample size from tolerable misstatement and risk, select items randomly or by monetary unit, then project any errors found to the population. ## Defining the population before audit sampling bank transactions The population is the complete set of bank transactions your conclusion will cover, and it has to match the assertion you are testing. Bank statement lines tested back to the ledger test whether recorded entries are complete. Ledger entries tested out to the statements test whether recorded entries really happened. Pick the population from the side you are testing from. ### Direction decides the population This is the step most often rushed. Suppose the goal is to find disbursements that cleared the bank but were never recorded. Sampling from the cash disbursements journal cannot find them, because an unrecorded payment is not in the journal. The population has to be the bank statement debits. The reverse test, checking that recorded payments are real, samples from the journal and looks for each item on the statements. [Vouching and tracing, and how the direction of a test decides its assertion](/blog/vouching-and-tracing), sets out the two directions in detail. ### Proving the population is complete Before selecting anything, show that the population you hold is all of it. For bank statement lines, that means every statement for every account in the period, each one footing from opening to closing balance, and each closing balance carrying forward to the next opening balance. A missing month or a statement that does not foot makes every sample drawn from it unreliable. The [bank statement audit guide to footing and continuity](/blog/bank-statement-audit-footing-continuity) covers those checks. For ledger entries, agree the total of the extract to the general ledger account activity. If the client uses Sage, its reconciliation and cleared-item reports give a convenient listing; the [walkthrough of Sage bank reconciliation in Sage 50 and Sage Intacct](/blog/sage-bank-reconciliation) shows where those reports sit. ### Splitting the population Bank activity often mixes very different items: payroll transfers, card settlements, vendor checks and one-off wires. Separating them into groups, called strata, lets you test each group in the way that suits it. Large or unusual items usually come out of the sampled population altogether and are tested individually. ## How big should a sample of bank transactions be? Sample size rises with the amount of assurance you need from the test, the size of the population in dollars, and the misstatement you expect to find. It falls as tolerable misstatement rises. Tolerable misstatement is the largest error in the population you could accept and still conclude the account is fairly stated. ### The factors in practice - **Risk of material misstatement.** When inherent and control risk for cash are high, the test has to carry more of the assurance, and the sample grows. - **Tolerable misstatement.** This is set from materiality for the engagement. Cash often has a low threshold because users expect the cash figure to be close to exact. - **Expected misstatement.** If prior years or other tests suggest errors, plan for them. A sample designed for zero errors has no room for even one. - **Population value.** In monetary unit sampling, sample size depends on the dollar value of the population, not the number of lines. - **Assurance from other procedures.** Analytics, confirmations and control tests that already address the same risk reduce the sample this test needs. ### Setting a monetary unit interval Monetary unit sampling treats every dollar in the population as a sampling unit, so large transactions are more likely to be picked. The usual shortcut is to divide tolerable misstatement by a confidence factor to get a sampling interval, then divide the population value by the interval to get the sample size. Firms publish their own factor tables. A factor of 3.0 is the common textbook figure for roughly 95 percent confidence with no misstatement expected, and that figure is used in the worked example below. Use your firm's tables on a real engagement. ## Random, systematic or monetary unit selection of bank transactions Selection is how individual bank transactions are picked once the sample size is known. Random selection gives every line an equal chance and suits tests where any line could hold an error. Systematic selection takes every nth item after a random start. Monetary unit selection weights each line by its amount, which suits tests aimed at overstatement. ### Random selection Number the population lines and draw random numbers with a tool you can document, such as a spreadsheet's random function with a recorded seed. Random selection treats a 15.00 bank fee and a 150,000.00 wire alike. That fits a test for errors of any size, such as checking whether each recorded payment has an approved invoice. ### Systematic selection Systematic selection is easier to run on a long printout. It is only safe if the population has no pattern that lines up with the interval. Bank activity often does: payroll every other Friday, rent on the first of the month. If the interval happens to match a cycle, the sample can miss a whole class of transactions or catch only one kind. Use a random start, and check the selections look varied before you test them. ### Monetary unit selection Monetary unit selection cumulates the line amounts and picks the line holding every interval-th dollar after a random start. Any line larger than the interval is certain to be selected. That makes it efficient for testing whether recorded amounts are overstated. It is weak at finding understatement, because a line recorded at zero or far too low has little chance of being picked. For completeness tests, use random selection from the side where the missing item would appear. ### Haphazard and judgmental picks Picking items by eye is not sampling, because the results cannot be projected. Judgmental picks of high-risk items are often worth testing, but report them as targeted testing, separate from the sample. ## Worked example: monetary unit sampling of bank disbursements The numbers below are an example built for this post. The test is whether recorded bank disbursements for the year are overstated, with each selected item vouched to an invoice, approval and the cleared statement line. | Step | Figure | How it is reached | | --- | --- | --- | | Population: recorded disbursements | 2,400,000.00 across 3,100 lines | Agreed to ledger activity | | Tolerable misstatement | 120,000.00 | From engagement materiality | | Confidence factor, no expected errors | 3.0 | Textbook figure for about 95 percent | | Sampling interval | 40,000.00 | 120,000 / 3.0 | | Sample size | 60 selections | 2,400,000 / 40,000 | | Lines over 40,000.00 | 7 lines, all selected | Larger than the interval | | Random start | 18,214.00 | From a recorded random draw | The first selection is the line that contains the 18,214th dollar when the disbursements are added in ledger order. The next is the line containing dollar 58,214, then 98,214, and so on. Any line of 40,000.00 or more will always contain at least one selection point. A line that holds two selection points counts once, so the final count of distinct items can be slightly below 60. ### The error found Suppose 59 items vouch cleanly and one does not. A payment recorded at 8,000.00 was for an invoice of 7,200.00; the vendor refunded the extra in the next month. The misstatement is 800.00, which is 10 percent of the recorded amount. That 10 percent is called the tainting. ## How do you evaluate misstatements found in a bank transaction sample? Evaluate a misstatement by projecting it to the population, adding an allowance for sampling risk, and comparing the total with tolerable misstatement. If the total stays below tolerable misstatement, the sample supports the account. If it does not, extend testing, test the area another way, or ask the client to find and correct the errors. ### Projecting the error In monetary unit sampling, an item smaller than the interval stands for the whole interval. The projected misstatement is the tainting times the interval: 10 percent of 40,000.00 is 4,000.00. Items larger than the interval are not projected; their actual error is simply added. ### Adding the allowance for sampling risk The upper limit on misstatement combines three parts: 1. **Basic precision**, the confidence factor times the interval: 3.0 times 40,000.00, or 120,000.00. 2. **Projected misstatement**, 4,000.00 from the one error. 3. **An incremental allowance** for having found an error, taken from the firm's tables. With the common textbook factors, it is about 3,000.00 here. The upper limit is about 127,000.00, above tolerable misstatement of 120,000.00. The sample does not support the conclusion on its own, even though the actual error found was only 800.00. This is what designing for zero expected errors means in practice: one error of any size pushes the limit over. Planning with a small expected misstatement would have produced a larger sample and left room for it. ### Looking at the cause, not just the amount A misstatement also has a cause. An overpayment that the vendor later refunded points at a weak invoice check in disbursements. That may matter more than the amount. Consider whether the same weakness could produce errors the sample would not catch, such as payments for goods never received. Unrecorded payables are the obvious companion risk; the [search for unrecorded liabilities using bank records](/blog/search-for-unrecorded-liabilities) tests payments after year end for obligations missing from the balance sheet. ## When audit sampling of bank transactions is the wrong tool Sampling is the wrong tool when the population is small enough to test every item, when a few large or unusual items make up most of the balance, or when the risk is a specific fraud pattern a random sample would rarely hit. In those cases test the key items individually or examine the whole population with data analysis. ### Testing everything A small account with 40 transactions a year can be tested in full in less time than it takes to document a sampling plan. Testing every item also means there is no sampling risk to evaluate, and the conclusion is direct. ### Hunting for specific patterns Duplicate payments, payments on weekends, round-dollar wires and gaps in check sequences are rare in a population, and a sample of 60 is unlikely to land on them. Full-population checks find them reliably. PreciseBankFlow foots every statement and lists possible duplicates and check number gaps across all lines, which gives you the population-wide exceptions to follow up. The sample then covers what those checks cannot, such as whether a payment was approved and for a real purpose. The decision on how much assurance each piece provides remains yours. ### When the population itself is unreliable If the statements do not foot or a month is missing, the problem is the population, not the sample. Fix the completeness question first, ideally with statements obtained from the bank, and only then sample. If you want those population checks run before you design the sample, [PreciseBankFlow](/get-started) is opening to audit firms in small groups.