All insights
ComplianceNovember 17, 2026

An internal IT audit checklist you can run yourself

Two days, ten samples per area, and evidence with dates. It finds more than most external assessments and it costs a fortnight of somebody's attention.

An internal IT audit sounds like something that requires an audit function, which most organisations of 50 to 500 people do not have. It does not. What it requires is somebody prepared to check a sample of things properly rather than reading the policy and assuming.

The distinction between those two is the whole discipline. A policy says leavers are disabled within 24 hours. An audit takes ten leavers from the last year and checks each one.

Sample, do not survey

The instinct is to be comprehensive: every account, every device, every control. That produces a spreadsheet nobody finishes and conclusions nobody trusts, because the later rows were skimmed.

Ten items, examined carefully, tells you more. If eight of ten leavers were disabled properly and two were not, you have found a twenty percent failure rate, which is a finding. If all ten were correct, you have reasonable confidence and you spent an hour.

Sampling properly is what separates an audit from a questionnaire
Sampling properly is what separates an audit from a questionnaire

What to cover

Scoped for a business with no internal audit function
Scoped for a business with no internal audit function

Taking the areas in turn, with the specific check for each.

Identity and access

Export every account and reconcile against the current staff list. Every account should map to a current person or a documented service. Anything else is a finding.

Then take ten leavers from the last twelve months and check each one properly: account disabled, sessions revoked, MFA methods removed, licence reclaimed, group memberships removed, and any individual sharing links revoked. The last two are the ones that fail.

Check that administrative accounts are separate from daily-use accounts. A person who reads email using an account that can also delete your tenant is a single phishing click away from a very bad day.

Finally, ask when the last access review happened and whether anything changed as a result. A review where nothing was removed is a review that did not happen properly.

Resilience

When was the last restore, what was restored, and how long did it take? A date and a duration. If the answer is a schedule or a dashboard, record it as a finding rather than accepting it.

Patch coverage: one number, trended, plus the exception list with owners and expiry dates. Exceptions with no expiry date are the finding.

Incident response: does the plan exist, is the contact list current, and has anyone exercised it? Pick up the plan and call the third number on it. If it rings the right person, that is more assurance than reading the document.

Governance

Reconcile the asset register against identity sign-ins. Devices signing in that are not in the register are unmanaged endpoints. This single comparison is the most productive governance check available.

Vendor list: does it exist, does every entry have an owner, and are the renewal dates current? Pick the three largest by spend and verify the renewal date against the actual contract.

Collect evidence as you go

A screenshot with a date, saved into a folder structured by area. This takes marginally longer than not doing it and it changes the exercise from an opinion into a record.

Two payoffs. Next year's audit becomes an update rather than a repeat, because you can compare. And when a client, an insurer or an auditor asks for evidence of a control, you have it already, which turns a two-week scramble into an email.

Rate findings by consequence

Not by the severity label a framework attaches. Ask what would actually happen.

A dormant administrator account belonging to somebody who left eight months ago is critical, regardless of what any scoring model says, because it is a working key held by somebody outside the organisation.

An unencrypted laptop belonging to a person who only ever uses web applications is a finding, and it is not the same thing.

Expressing it this way is also what gets findings funded. "Access control deficiency, rated high" produces a nod. "A former employee's administrator account has been active for eight months" produces a decision.

Write it short

Three parts.

Five findings, ranked, each with the consequence in one sentence, an owner and a date. Five, not thirty. If you found thirty, the top five are what will actually get fixed and the rest can be an appendix.

What was checked and what was not. The scope statement protects everybody. An audit that did not look at the production environment should say so plainly.

What improved since last time. The part that makes anyone want to do it again.

Cadence

Annually for the full exercise, two days of one person's time. Quarterly for the two checks that decay fastest: the leaver reconciliation and the asset-to-identity comparison. Those two are twenty minutes each once the first full pass has built the baseline.

When to bring somebody external in

Three situations. When you need independence, because a client or an insurer will not accept a self-assessment. When you suspect the internal view is incomplete, which is a normal and healthy suspicion. And when you are preparing for a formal certification, where somebody who knows what that specific assessor looks for saves considerably more than they cost.

Otherwise, the internal version run honestly finds most of what matters, and it finds it every year rather than every third year when there is budget.

Work with us

Want a second set of senior eyes on your Microsoft 365 environment?

Book a free call