On this page
- What “Sample Size” Means in a SOC 2 Audit
- How Auditors Actually Decide the Sample Size
- Illustrative Sample-Size Ranges Commonly Observed in Practice
- How a Sample Plays Out: Two Worked Examples
- What Is the Difference Between a Deviation and an Exception?
- Can an Auditor Test Every Item Instead of Sampling?
- What to Ask a Firm About Sampling
SOC 2 sample size is the number of control instances an auditor tests out of everything that happened during the observation period. No standard fixes it. AT-C section 205, the AICPA attestation standard for SOC 2 examinations, asks only for a sample large enough to keep sampling risk acceptably low and selected so it can reasonably be expected to represent the population. Firms size it from control frequency, population size and the deviation rate they will tolerate: 2 items for a quarterly control, 5 to 9 for a weekly one, and 25 for a large population such as deployments, rising to 40 and 60 if deviations appear.
What “Sample Size” Means in a SOC 2 Audit
In a SOC 2 Type 2 audit, the auditor identifies every time a control operated during the observation period (each access change, each production deployment, each access-review cycle) and tests a subset rather than every instance. The auditor then concludes whether the control operated effectively across the whole period based on how that subset performed.
A Type 2 report asks whether a control operated consistently across the window, not whether it was well designed on one day. The auditor defines the population (say, every access change logged in the identity provider over 12 months), pulls a subset, and tests each item against the control as written. Our Type 2 controls guide explains how period testing works more broadly, and our SOC 2 audit report guide shows where exceptions appear in the finished report.
How Auditors Actually Decide the Sample Size
No number is required. AT-C section 205 says the sample size must be enough to reduce sampling risk to an acceptably low level, and the items must be selected so the auditor can reasonably expect the sample to be representative of the population. How many items that means is each CPA firm’s methodology, built from four inputs: control frequency, population size, tolerable deviation rate, and risk.
The requirement sits in AT-C section 205 (Assertion-Based Examination Engagements), paragraph .32, which covers any attestation procedure that uses sampling. The same paragraph says that a selected item the auditor cannot test, because the evidence does not exist, is treated as a deviation, and that the auditor must investigate the nature and cause of every deviation found. The standard names no sample sizes. It points to the AICPA Audit Guide Audit Sampling as guidance that “may be useful”, which makes the guide’s tables a reference, not a rule.
The four inputs work like this:
- How often the control operates. A quarterly review produces four instances in a 12-month period; a control that fires on every deployment or login can produce thousands.
- How many instances exist in the observation window once frequency is applied.
- The tolerable deviation rate: how much failure the auditor will accept before concluding the control did not operate effectively. A firm that tolerates less takes a larger sample.
- Risk. Linford & Co, a CPA firm, frames it as the impact, in qualitative and quantitative terms, if the control is not operating effectively.
Because each firm applies its own methodology, two firms can size the same population differently and both be defensible.
Illustrative Sample-Size Ranges Commonly Observed in Practice
For a control that operates a few times a year, auditors test most of the population: two of four quarterly instances, two to four of twelve monthly ones. A weekly control gets five to nine samples. Samples of 25 to 60 are for large populations, such as daily or event-driven controls with 250 or more instances in the period: 25 is the usual start, and a deviation pushes it to 40, then 60.
The table is a sense check, not a rule. The quarterly to weekly rows are the AICPA’s small-population table, adapted from table 3-5 of the AICPA Audit Guide Audit Sampling (2019 edition). Linford & Co’s published table for less frequent controls gives the same figures for annual (1), quarterly (2), monthly (2 to 4) and weekly (5 to 9) controls. The large-population row is the approach Linford publishes for its SOC 1 and SOC 2 examinations.
| Control frequency | Population in a 12-month period | Sample size |
|---|---|---|
| Annual (e.g., a risk assessment, a DR test) | 1 | 1 (the whole population) |
| Quarterly (e.g., a vendor risk review) | 4 | 2 |
| Monthly (e.g., a monthly reconciliation) | 12 | 2 to 4 |
| Semimonthly | 24 | 3 to 8 |
| Weekly (e.g., a weekly log review) | 52 | 5 to 9 |
| Daily, or event-driven (e.g., access changes, deployments) | 250 or more | 25 to start; 40 after one deviation, 60 after two |
The last row is the one most people mean by “the standard SOC 2 sample size”. There isn’t one, but 25 is a common starting point for a large population. Linford’s table for large populations at a 90% confidence level shows how the tolerable rate moves the numbers:
| Deviations found | 8 to 10% tolerable rate | 5 to 7% tolerable rate |
|---|---|---|
| 0 | 25 | 40 |
| 1 | 40 | 60 |
| 2 | 60 | 90 |
A shorter observation period shrinks every population: a monthly control in a six-month period has six instances, not twelve (our observation period guide covers how long a window to choose). To have each population ready before the auditor asks, see what to pull by evidence source and our SOC 2 compliance checklist.
How a Sample Plays Out: Two Worked Examples
A quarterly access review has four instances in a year and the auditor tests two, so each carries half the result. A deployment approval control can have over a thousand instances and the auditor tests 25, then more if one fails. These examples are illustrative: a hypothetical company, the sample sizes above, and the evidence our controls library lists for each control.
Quarterly user access review (AC-06)
AC-06 in the library is a quarterly review of every account and its permissions on in-scope systems, with a keep or remove decision recorded for each and removals completed by a deadline. It maps to CC6.2 and CC6.3. Over a 12-month period the population is four reviews, so the auditor selects two.
For each selected quarter, the library lists three items the auditor asks for: the dated export of accounts and permissions that was reviewed, the sign-off showing the reviewer, the accounts covered, the decisions and the changes made, and the removal records showing that access marked for removal was actually removed. From those the auditor can check that the export covers the in-scope systems, that every account has a decision, and that each removal decision has a ticket or identity-provider change behind it. The other two quarters are not inspected, but the library’s Type 2 note for AC-06 asks for a completed review in every quarter, because a quarter with no review leaves nothing to inspect.
The common gap the library records is a reviewer approving the whole list without checking real permissions, or removals decided in the review that never happen. One failure in two tested reviews is a 50% deviation rate with almost nothing left to extend into. Linford’s own example for a monthly control says that when the population is small, any deviation means the control did not operate effectively. A quarterly control is smaller still, so finish the removals inside the quarter and keep the export with the sign-off.
Change approval on about 1,100 deployments (CM-01, CM-03, CM-04)
Take a company that deploys about 1,100 times in 12 months. Its control is that no change merges to the production branch until someone other than the author approves it, enforced by branch protection (CM-01, mapped to CC8.1 and CC5.1), and that every release is recorded in a deployment log tied to an approved change (CM-03). The auditor needs a deployment list they can trust, because a sample says nothing about changes that were never on it. The list should match source-control and pipeline history.
From 1,100 items, Linford’s approach starts at 25. For each sampled deployment the library’s CM-01 evidence is the merged pull request with its reviewer and approval time, the branch protection settings that were on throughout the period, and the log of anyone who bypassed them. CM-03 adds the deployment record linking the release to a commit and an approved change.
Now suppose one of the 25 was approved by someone who also wrote the change, or approved after it merged. Under Linford’s published approach the sample extends to 40, so 15 more items are drawn from the same population. The auditor also has to investigate the cause (AT-C section 205, paragraph .32). An urgent hotfix approved a day late reads differently from an administrator who routinely bypasses branch protection, and the bypass log shows which it was. CM-04 covers the first case: an emergency path that lets an urgent fix ship first, then requires review and approval within a set number of business days. It helps only if every emergency change can be identified, and the common gap the library records is hotfixes that go to production with no ticket and nobody able to list them.
If the 15 extra items come back clean, the result is one deviation in 40, which the auditor typically describes in the test results. Whether it affects the opinion is a separate judgment, covered in our guide to SOC 2 exceptions and qualified opinions. In Linford’s example a second deviation extends the sample to 60, and a further one after that stops sampling with a conclusion that the control did not operate effectively. Other firms set their own stopping points. Our change management controls guide shows the ticket-to-deploy packet an auditor traces for each sampled deployment and how the exception can read in a report.
What Is the Difference Between a Deviation and an Exception?
A deviation is one sampled item that fails the test procedure: the evidence does not match the control as written. An exception is the auditor’s documented description in the report of the control tested and the deviations found. A single deviation can lead to an exception, and deviations can also lead the auditor to extend testing before deciding.
Auditors treat the deviation as the raw fact and the exception as the narrative built around it. Missing evidence counts too: under AT-C section 205 an item the auditor cannot test is treated as a deviation, so a ticket that was never written can fail a sample as surely as a wrong approval.
Can an Auditor Test Every Item Instead of Sampling?
Yes. AT-C section 205, paragraph .32, begins “If sampling is used”, so sampling is a choice, and Linford notes that block sampling can cover 100% of a population. A small population is often tested in full: an annual control has one instance and a monthly control in a short window may have only a few. For high-volume controls, whether a firm will test everything is a question about its methodology, and it depends on the evidence being complete and machine-readable.
What to Ask a Firm About Sampling
Ask each firm for its sampling table, the tolerable deviation rate behind it, and the point at which it stops extending a sample. Those answers put firms on the same footing. Our directory of SOC 2 audit firms compares them on pricing, timelines and testing approach.
More in Audit Preparation