Short answer
Plan the audit like a test. Fix the scope and the criteria, pick an auditor who did not do the work, and decide the sample before you start. Then ask each process owner for records, not descriptions. Write every nonconformity in three parts: the requirement, what you observed, and the records that show it. Keep opportunities for improvement separate. The free audit plan and checklist templates below follow this structure, and the worked example follows one finding from the audit to CAPA.
What ISO 13485 expects
Internal audits sit in clause 8.2.4 of ISO 13485:2016 (“Internal audit”). In outline, the clause expects audits at planned intervals to check that the QMS conforms to your own arrangements, the standard and applicable regulatory requirements, and that it is effectively implemented. It also expects a documented procedure, an audit programme that takes account of the status and importance of processes and of earlier results, auditors who do not audit their own work, records of audits and results, and timely action on findings. Read the exact wording in your licensed copy. ISO 19011 (now in its 2026 edition) gives guidance on running an audit programme and on auditor competence.
In the EU, Article 10(9) of the MDR lists what a manufacturer’s quality management system must address. Internal audit is how most manufacturers show that those processes work. If you also sell in the United States: the Quality Management System Regulation has applied since 2 February 2026. In its QMSR FAQ, FDA says it now has the authority to inspect management review, quality audit and supplier audit reports, because the old exception was not carried over. Write your reports so they can stand on their own.
Plan: scope, criteria, auditor, sample
Most weak audits fail at the plan, not on the day. Decide these four things in writing, and have the plan approved before the audit:
- Scope. Processes, products, versions, sites and the period of records. “Development” is not a scope. “Requirements-to-verification links and release evidence for candidate 1.0” is.
- Criteria. Your own controlled procedures, with versions, plus the clauses and regulatory requirements in scope. An audit against “ISO 13485” in general cannot produce a precise finding.
- Auditor competence and independence. Record the auditor’s training and experience. Exclude each auditor from the areas they work in. In a small company the usual answer is cross-auditing plus an external auditor for the QA function itself.
- Sample. How many records, which ones, and why. With six requirements in a release you can check all six. With six hundred, sample the riskiest ones, the most recently changed ones, and a random set. Write the rationale down.
Over the programme cycle, cover every QMS process. Audit more often where processes changed, failed before, or matter most for device safety.
What to ask in a software team
Ask for the record first, then talk about it. The table below lists questions and evidence for the processes where software teams most often slip. Clause numbers point to ISO 13485:2016 unless stated otherwise. Use them as a starting question set in the checklist template, one row per question: the document reviewed, the ISO clause, your observation, and Pass or Fail.
| Process | Ask | Objective evidence to request | Where software teams often fall short |
|---|---|---|---|
| Design and development planning (7.3.2; IEC 62304 5.1) | Does the plan match how the team actually works? | Approved development plan; its revision history; the sprint or release cadence it describes | Plan written for a waterfall project, while the team ships every two weeks |
| Requirements and traceability (7.3.3, 7.3.10) | Can each sampled requirement be traced to its source, risk control and verification on the same baseline? | Requirements export with approvals; trace export; test results with build IDs | Links exist but tests ran on an earlier baseline; late requirements skip test planning |
| Verification and validation (7.3.6, 7.3.7; IEC 62304 5.7) | Was the released build the one that was tested? Was validation done with intended users? | Executed test records; anomaly list; validation report on an identified build | “Done” in the tracker used as if it meant “verified” |
| Release and changes (7.3.8, 7.3.9; IEC 62304 5.8, 8.2) | Was each change assessed before implementation, and each release authorised on complete evidence? | Change requests with impact and regulatory assessment; release record; known anomalies | Hotfixes released outside the change process |
| Configuration and problems (IEC 62304 8.1, 9) | Can you rebuild the release? Were problem reports evaluated for safety? | Configuration baseline; SOUP list or SBOM; problem reports with safety evaluation | Bugs closed as “won’t fix” with no recorded safety evaluation |
| Risk management (ISO 14971; IEC 62304 7) | Is the risk file current for this baseline? Are risk controls implemented and verified? | Risk management file revisions; links from risk controls to requirements and tests | Risk file last updated at the first release |
| Suppliers and outsourcing (4.1.5, 7.4.1–7.4.3) | Were suppliers evaluated in proportion to risk, and are their changes controlled? | Supplier evaluation and decision; quality agreement; acceptance records; register | Cloud and development suppliers missing from the approved supplier list |
| Feedback, complaints and reporting (8.2.1–8.2.3) | Are complaints assessed for reportability, with a decision recorded? | Complaint records; reportability assessments; PMS data | Support tickets never screened as possible complaints |
| CAPA (8.5.2; EU MDR Art. 10(9)(l)) | Is the need for CAPA assessed? Are causes evidenced? Was effectiveness checked against criteria set in advance? | CAPA log; investigations; effectiveness plans and results | CAPAs closed when training is done |
| QMS software (4.1.6) | Are the tools that hold QMS records validated for that use? | Validation records for the tracker, CI and document system; configuration change log | Workflow changed by an admin without revalidation |
Sampling tip: follow one trail end to end. Take one requirement, follow it to its tests, the release, the change that last touched it and the risk control behind it. One trail done properly shows more than ten documents opened and closed. Our guide to supplier qualification for outsourced SaMD development covers what good evidence looks like.
What counts as objective evidence
Objective evidence is something another person could check later: a record with an ID, a version, a date and an owner. Good evidence in a software team includes:
- an export of requirements and trace links for a named baseline, saved as a controlled record
- CI test results tied to a commit or build ID, and a reviewer’s approval
- a merged pull request with its approvals and required checks
- a signed (or electronically approved) release record that names the build
- a change request showing the impact assessment before the change was implemented
Interview answers help you find records; on their own they are not evidence. A screenshot can be evidence if it shows enough context to identify the record, its version and the date. Write down exactly what you looked at, including what you looked for and did not find.
Worked example: one limited audit
Fictional example
Example Manufacturer A develops Example SaMD A, a fictional web application that shows laboratory reports to an authorised reviewer and sends reviewed reports to the hospital record system through a supplier-maintained API. Before the release review of candidate 1.0, it ran audit EX-AUD-01 on 3 August 2026:
- Scope: requirements-to-verification links and release evidence for candidate 1.0; supplier approval and agreement for the API supplier; the CAPA procedure.
- Criteria: internal procedures EX-QP-DEV v1, EX-QP-SUP v1 and EX-QP-CAPA v1. These are invented for the example and are not quotations from any standard.
- Auditor: Audit Lead A, with a competence record and no role in the sampled records.
- Sample: all six requirements and their tests; the supplier’s evaluation, approval and quality agreement; the CAPA procedure (no CAPA records existed yet).
Outcome: one nonconformity and one opportunity for improvement within the sample. Complaint handling, training and the rest of the QMS were outside the scope and were recorded as not audited, so the report draws no conclusion about them. If your procedure grades findings, classify them as defined in the plan. The plan template uses these definitions:
| Classification | Meaning |
|---|---|
| Major nonconformity | One or more requirements of the management system or standard were not met, and this could lead to a significant deficiency in supplying a product or service. |
| Minor nonconformity | A single lapse in the management system that will not lead to a significant deficiency in supplying a product or service. |
| Opportunity for improvement | A deficiency, problem or chance to improve practice that does not qualify as a nonconformity. Not a softened nonconformity. |
Writing findings
A finding someone can act on has three separate parts: the requirement, the observation and the objective evidence. Leave one out and the finding becomes an opinion that is easy to dispute and hard to close. Here is the example nonconformity, EX-NC-001, next to a weak version of the same finding:
| Weak | Well-formed | |
|---|---|---|
| Requirement | (not stated) | EX-QP-DEV v1 §4: every in-scope requirement links to an approved test and reviewed result before release review. |
| Observation | “Traceability is poor.” | In candidate 1.0, requirement EX-REQ-006 had no linked test and no execution result. |
| Evidence | (not stated) | Requirements export and test index for candidate 1.0, reviewed on 3 August 2026. Five of six requirements were linked to passed tests. |
The observation describes what was found, not why. The cause (here, a workflow that let a requirement reach “Done” without a test link) comes later, from the investigation. The audit also raised EX-OFI-001: the test protocol template has no build ID field. Every sampled record still identified its build through CI links, so the requirement was met. The finding is recorded as an opportunity for improvement, not as a nonconformity.
After the audit
- Report promptly, with the scope as performed, any deviations from the plan, the findings and a conclusion limited to the sample.
- Correct and contain. In the example, the candidate was held from release review the next day.
- Assess the need for CAPA. Not every finding needs one. This one did: the same workflow gap was open for every requirement and every release. The investigation and effectiveness checks are covered in CAPA for medical software.
- Verify closure with evidence, and keep the original finding. Do not edit an audit record later to show what happened next.
- Report to management review, including trends across audits.
Common mistakes
- Auditing the procedure instead of the records: “we have an SOP for that” is not an audit result.
- The QA manager auditing the QA processes they run.
- Findings with no requirement, or with a cause written in place of an observation.
- Recording opportunities for improvement as minor nonconformities, or the other way round.
- Conclusions about the whole QMS from a sample of one process.
- Findings “closed” on the date the action was planned, not when it was verified.
- Ignoring the tools: workflow changes in Jira or CI made without revalidation.
Download the templates
Free to download, no sign-up. Both open in Word or LibreOffice Writer.
- Internal audit plan — templateCompany and sites, planned dates, scope and objectives, criteria, auditors and auditees; the audit process with linked-process examples; an agenda with major, minor and opportunity-for-improvement definitions; plan approval.Download DOCX
- Internal audit checklist — templateAudit number, date, auditors and auditees, then one row per question: document reference, ISO 13485 clause, question, observation or response, and Pass/Fail.Download DOCX
Each template has a document control block and [placeholders] for your company name, product name and document IDs. Finding grades, response times and audit frequency must follow your own audit procedure. Completing the templates does not guarantee the outcome of a certification audit.
Want an outside view? I run internal audits and mock certification audits as part of ISO 13485 QMS implementation, and internal audits on an agreed schedule within ongoing QA/RA support. For the full procedure set, including an internal audit SOP and programme, see the ISO 13485 QMS templates.
Sources
- ISO 13485:2016 – official sample pages incl. table of contents (iTeh Standards) — clause titles
- DQS – ISO 13485 internal audit (certification-body article, 22 October 2025) — planned intervals; auditors do not audit their own work
- ISO 19011:2026 – Guidelines for auditing management systems (ISO)
- IEC 62304:2006+AMD1:2015 – preview incl. table of contents
- Regulation (EU) 2017/745, Article 10 (as adopted; legislation.gov.uk) — Article 10(9)
- FDA – Quality Management System Regulation: Frequently Asked Questions (content current as of 5 November 2026) — question 8
- FDA – Medical Devices; Quality System Regulation Amendments, final rule (Federal Register, 2 February 2024) — effective 2 February 2026