Guide, read 12 Nov 2026

# SOC 2 auditors test a sample of each control population, and one firm publishes a 3 to 30 item range

**The short answer:** SOC 2 auditors test a sample of each control population rather than every instance. KirkpatrickPrice publishes a rule of at least 3 and at most 30 items, around 10 percent of the population. The AICPA sample size table could not be retrieved, so each firm's own method applies, and the company's population list sets what can be selected.

[Open the glossary](https://soc2compliancesoftware.com/glossary/) | [See the ranked index](https://soc2compliancesoftware.com/)

## Sampling means testing less than all of a population, with items chosen by the auditor

KirkpatrickPrice defines sampling as applying audit procedures to less than 100 percent of a population, and its January 31, 2018 page lists example populations: new hire training forms, employee acknowledgements of policies, antivirus reports and access control logs. The page quotes the PCAOB on the auditor using professional judgment in planning, performing and evaluating a sample and relating the result to other evidence. In a SOC 2 Type 2, each control that repeats during the period has a population.

KirkpatrickPrice says on May 24, 2019 that auditors use sampling when the population is uniform and the same standards apply across it, and that the sample is randomly selected. Its article on locations adds that when designing a sample, auditors evaluate the purpose of the sample, outliers and behavior, and consider sample risk to understand how many errors could sit in the data.

The auditor picks the items, not the company. BARR's training example shows what follows: the auditor's sample of employees included one without training, and that single person became an exception. Schellman tells companies facing a Type 2 to be ready with evidence for all samples that an auditor may select. Plan for any item in the population to be requested.

## A population is every instance of a control during the period, and the list must be complete

A population is the full list of times a control ran, or should have run, in the examination period: every employee hired, every employee terminated, every production change, every vendor review. The auditor cannot sample from a list the company has trimmed, so the list is the first piece of evidence tested. Auditors commonly ask how a list was generated and whether anything could be missing. This index presents that as practice, because the AICPA text was not retrieved.

Completeness and accuracy are separate questions. Completeness asks whether every instance is on the list: all terminations, and not most of them. Accuracy asks whether each row is right: the termination date matches the HR record, not the date a ticket was updated. BARR's consultant says one employee is easy to overlook at a larger company without a tool to track compliance, and the sample is where that gap surfaces.

Schellman says insufficient documentation is one of the biggest causes of audit overrun, because auditors then change method or delay timelines, and both add fees. A population that has to be rebuilt in the middle of fieldwork is the same problem in another form. Schellman adds that a prepared company spends less time chasing evidence and squeezing in calendar time for interviews.

## Sample sizes are set by each firm, so the same control can draw different counts

This index did not retrieve an AICPA sample size table. The AICPA's audit sampling guide page returned HTTP 404 on 12 Nov 2026, and the SOC 2 guide is sold rather than posted. No authoritative table by control frequency is reproduced here. What follows is one firm's published practice and should not be read as the standard. Ask the audit firm for its sampling methodology when scoping.

KirkpatrickPrice's May 24, 2019 page says it samples anywhere from 10 to 30 percent of a population, never fewer than three. Three new hires are all tested, and 100 new hires give a sample of ten. Its video transcript says that of a population of 1,000, the auditor may take a maximum of 30. Its January 31, 2018 SOC 1 page says a population of 100 might give a random sample of 30, so even one firm's pages differ.

For contrast, Insight Assurance's page on PCI DSS describes the PCI SSC method: a population of four to thirty components might need a sample of three, and a population over a thousand caps at thirty. That is a different framework with a defined method. None was found for SOC 2 in the sources read on 12 Nov 2026, which is why the planning memo or engagement letter is where a company should ask how many items the firm tests per control and per frequency.

| Population | Items tested | Source page |
|---|---|---|
| 3 employees hired | All 3 | KirkpatrickPrice, May 24, 2019 |
| 100 employees hired | 10 | KirkpatrickPrice, May 24, 2019 |
| 100 items, large-population example | 30 | KirkpatrickPrice, January 31, 2018 (SOC 1) |
| 1,000 items | Maximum of 30 | KirkpatrickPrice, May 24, 2019 video transcript |

## The company controls the list, and grouping the list by process keeps the sample honest

KirkpatrickPrice says its sample is randomly selected. A random draw from a complete list is fair to the company. A draw from a partial list is not, because the items left out can never be picked, and the omission may show up later as a surprise. A control that ran three times is tested three times, as in the firm's own example. A control that ran 400 times is the one where the list needs checking.

Insight Assurance's PCI page makes a point that carries over: categorize the population by type and function before sizing the sample. Lumping a web server farm and administrative workstations into one bucket undercounts a population that was more varied than the label suggested. For a SOC 2, split a population when it has two paths, such as code releases through pull requests and infrastructure changes made in a console.

A single combined list hides which path an exception came from, and the auditor may have to ask again. Name each population the way the control matrix names the control, and state the path in a column. Where the company expects the auditor to treat two paths as one, say so at planning, and keep the reason on file in case a reviewer later questions the grouping.

## One exception in a sample can widen the testing

BARR's consultant describes the sequence. The auditor asks the client about suspect evidence, and there is often an explanation. In one case a vulnerability tracking tool showed unresolved items because it was not tagging severity correctly. The client fixed the tagging process and the issue did not become an exception. In other cases the client confirms that the control was not designed or operating properly, and the audit team decides the effect on the criteria.

Insight Assurance's PCI page says that when one item in a sample fails, the assessor considers whether the failure points to a wider gap across the population. That can mean expanding the sample, retesting a wider set after remediation, or treating the category as non-compliant until verified. SOC 2 firms apply their own methods, so ask your firm whether an expanded sample is possible and who must supply the extra items.

Get ahead of it. If one terminated user kept access, check every terminated user in the period yourself and tell the auditor what you found. That matches the steps Schellman lists: validate the finding, identify mitigating practices, correct the issue, find the root cause and write the remediation plan. A company that arrives with the whole-population result gives the auditor less reason to widen the test.

## A clean population export saves days of fieldwork

Export from the system of record, not from a spreadsheet someone maintains. Each file needs a header with the source system, the query or filter, the run date and the period covered, then one row per instance with a unique ID, a date and an owner, and totals that reconcile to the source. Save the query or a screenshot of the filter with the file so the auditor can see how the list was produced.

Cut populations at the period boundaries and say so. If the period is January 1 to December 31, the list should show its first and last dates, with no instances outside them. Do not edit rows by hand. If a row is wrong, fix the source and run the export again, because a manually changed cell is the first thing a careful reviewer will question and the hardest to defend.

Build the populations before fieldwork, one per control, in the order of the control matrix. A company that can produce a complete list of terminations, changes, hires and vendor reviews within an hour is a company whose controls are being tracked. Where the list cannot be produced, that is a control gap before it is an evidence problem, and it is cheaper to find during readiness than during testing.

## More guides

- [What a SOC 2 Type 2 auditor asks for: population lists, samples and dated evidence](https://soc2compliancesoftware.com/guides/what-an-auditor-asks-for/)
- [The SOC 2 observation window: 3 to 12 months, and every control must run through it](https://soc2compliancesoftware.com/guides/the-observation-window/)
- [What software cannot do in a SOC 2 examination: the opinion, the assertion and the judgment calls](https://soc2compliancesoftware.com/guides/what-automation-cannot-do/)

## Sources

- [KirkpatrickPrice: Auditing Basics, How Does Sampling Work](https://kirkpatrickprice.com/video/auditing-basics-sampling-work/), read 12 Nov 2026
- [KirkpatrickPrice: Understanding Your SOC 1 Report, How Does Sampling Work](https://kirkpatrickprice.com/video/understanding-your-soc-1-report-how-does-sampling-work/), read 12 Nov 2026
- [KirkpatrickPrice: Understanding Your Audit, Locations and Sampling](https://kirkpatrickprice.com/blog/understanding-your-audit-locations-and-sampling/), read 12 Nov 2026
- [BARR: Common SOC Report Exceptions and How To Mitigate Them](https://www.barradvisory.com/resource/common-soc-report-exceptions/), read 12 Nov 2026
- [Insight Assurance: How PCI DSS Assessors Sample Your Environment](https://insightassurance.com/insights/blog/pci-dss-sampling-methodology/), read 12 Nov 2026
- [Schellman: Do You Need a Readiness Assessment](https://www.schellman.com/blog/audit-readiness/do-you-need-a-readiness-assessment), read 12 Nov 2026
- [Schellman: How to Scope a SOC 2 Audit](https://www.schellman.com/blog/soc-examinations/how-to-scope-a-soc-2-audit), read 12 Nov 2026
- [Schellman: How to Address a Testing Exception in Your SOC Report](https://www.schellman.com/blog/soc-examinations/how-to-address-testing-exceptions-in-soc), read 12 Nov 2026
Last reviewed 12 Nov 2026.

Canonical: https://soc2compliancesoftware.com/guides/evidence-sampling/
