Control Testing and Audit Evidence
how to prove a control works, not just that it exists
Writing a control is easy. Showing that it really works, every time is the harder part, and it is the heart of auditing and assurance.[1]
Two questions for every control
- Is it designed well? If it ran perfectly, would it reduce the risk it targets? (A policy that says "review logs" but names no owner or frequency is badly designed.)
- Does it operate effectively? Was it actually performed as described, consistently, over the period?
Both must be true. A well-designed control that nobody performs fails the second test; a control performed daily that does not address the risk fails the first.
Ways to test
- Inquiry: ask people to explain the control.
- Observation: watch it being done.
- Inspection: examine records, tickets, configurations and logs.
- Re-performance: repeat the control yourself and compare results.
Combine them. Inquiry alone is weak.
Populations and samples
Define the population (all access approvals in the period), check it is complete and accurate (where did the list come from, and could anything be missing?), then select a sample. Larger populations and higher risk call for larger samples; for controls run by automated systems, one test may be enough if the configuration is stable and controlled. Record how the sample was chosen.
What good evidence looks like
- Relevant: it proves the specific control attribute you are testing.
- Reliable: system-generated beats typed-in; from an independent source beats from the control owner; original beats a copy.
- Dated and attributable: shows when, by whom and for which system. Screenshots without dates or names are weak.
- Complete: covers the whole period and population, not a favorable slice.
- Reproducible: another person could get the same result by following your steps.
Examples
- Control: terminated staff lose access within one day. Test: take the HR leavers list (check its completeness), select a sample, and compare each leaver's termination date with the account disable time in the directory logs.
- Control: changes are approved before release. Test: sample deployments and trace each to an approved ticket that predates it.
Documenting it
Record the objective, procedure, population, sample, evidence references, results, exceptions and conclusion, in working papers that someone else could review. Where tests fail, record the cause and impact, not only the fact.[2]
Automation
Many controls can be tested continuously from system data, which gives more coverage and less point-in-time stress. Automated evidence still needs a check that the data source is complete and unaltered.
Sources
Test yourself
My notes
Private — saved only on this device, never sent anywhere.