
Governance & Compliance
https://hmmnm.com/iso-27001-from-scratch-part-2-risk-assessment-that-auditors-accept/
TL;DR: Clause 6.1.2 of ISO 27001 asks you to define a repeatable risk methodology and actually apply it — auditors fail you when scores are inconsistent, owners are missing, or residual risk is never re-evaluated. This article walks through building an asset-based methodology with likelihood × impact scoring, a defensible risk register with inherent and residual scores, and the four treatment decisions, using the 50-person SaaS company from Part 1 as the worked example.
Authorization note: everything in this article is documentation and spreadsheet work on your own lab machine. If you add the optional scanning steps, run them only against systems you own or have written permission to test.
Clause 6.1.2 is short, which is why people underestimate it. You must:
Auditors check three things: is your methodology defined, did you apply it consistently, and does your evidence connect (risk → treatment → control → statement of applicability)? Break any of those links and you get a nonconformity.
There are two dominant approaches:
| Approach | How it works | Best for | Weakness |
|---|---|---|---|
| Asset-based | Inventory assets, assign each a value, enumerate threats to each asset | First-time certifications, smaller orgs | Can become a checklist that misses novel attack paths |
| Scenario-based | Write risk scenarios (“attacker steals production DB backup from misconfigured S3 bucket”) | Mature orgs, threat-modeling cultures | Harder to prove completeness; scenarios can be cherry-picked |
For a first certification at a 50-person SaaS, go asset-based with scenario overlays: inventory your assets, then for the crown-jewel assets write 2–3 concrete scenarios each. Auditors like this because it’s systematic (asset-based) but not shallow (scenarios show real analysis).
You cannot assess risk to assets you haven’t listed. Keep it simple — for the Part 1 SaaS company:
| Asset | Type | Owner | Value |
|---|---|---|---|
| Production PostgreSQL cluster | Information + system | Head of Engineering | Very High (customer data) |
| SaaS web application | Application | CTO | Very High |
| AWS account & infrastructure | System | Platform Lead | High |
| Employee laptops (~50) | Hardware | IT Manager | Medium |
| Google Workspace (email, docs) | SaaS service | IT Manager | High |
| Customer credentials in identity provider | Information | Security Lead | Very High |
| Source code repository | Information | Head of Engineering | High |
Document the value scale too — “Very High” must mean something, e.g. “regulatory or contractual breach, existential business impact.”
If your lab mirrors the SaaS environment, pull an asset list from cloud tooling so your inventory isn’t fantasy. For an AWS lab account:
aws ec2 describe-instances
--filters "Name=instance-state-name,Values=running"
--query 'Reservations[].Instances[].[InstanceId,InstanceType,Tags[?Key==`Name`].Value|[0]]'
--output table
------------------------------------------------------------------------------
| DescribeInstances |
+---------------------+-------------+----------------------------------------+
| i-0a1b2c3d4e5f | t3.medium | prod-app-01 |
| i-0f6e5d4c3b2a | db.r6g.large | prod-postgres-01 |
+---------------------+-------------+----------------------------------------+
Expected output: a table of running instances. Reconcile this against your spreadsheet. Any instance not in the inventory is a finding in itself.
Use a 5×5 matrix. Five-point scales are granular enough to be meaningful and coarse enough that two people can actually agree on a score. Document the definitions — auditors ask for them.
| Score | Label | Definition (write something like this) |
|---|---|---|
| 1 | Rare | Less than once per 5 years; would require exceptional circumstances |
| 2 | Unlikely | Once every 2–5 years; industry reports few comparable incidents |
| 3 | Possible | Once every 1–2 years; comparable incidents reported in our sector |
| 4 | Likely | Expected within 12 months; near-misses have occurred internally |
| 5 | Almost certain | Expected multiple times per year or already occurring |
| Score | Label | Definition |
|---|---|---|
| 1 | Negligible | No material harm; absorbed in normal operations |
| 2 | Minor | Limited internal disruption or minor data exposure; no customer impact |
| 3 | Moderate | Customer-visible outage or limited data exposure; contractual penalties possible |
| 4 | Major | Sustained outage, reportable breach, regulatory fines, significant churn |
| 5 | Severe | Existential: loss of major customers, fines threatening viability, breach of all customer data |
Risk score = likelihood × impact, mapped to an acceptance band:
| Score | Rating | Required action |
|---|---|---|
| 15–25 | Critical | Treat immediately; executive visibility; escalate if not treated in 30 days |
| 8–12 | High | Treat within 90 days |
| 4–6 | Medium | Treat within 6 months or formally accept |
| 1–3 | Low | Monitor; accept by default |
Calibration tip: score three risks as a group before going solo. If engineering scores “developer laptop stolen” likelihood 4 and security scores it 2, you have a definition problem, not a judgment problem. Fix the definitions.
Your register is the centerpiece evidence. Minimum required columns:
Inherent risk assumes no controls. Residual risk is what remains after your existing controls. The gap between them is your justification for the controls you’ve implemented. Common audit failures:
Give each register row a “linked Annex A controls” column. When the auditor picks R-003 and asks “why is A.8.24 in your SoA?”, the answer is on one line. Keep this in a simple spreadsheet or GRC tool — a CSV in version control is fine and auditable:
id,asset,description,inherent_l,inherent_i,inherent_score,existing_controls,residual_l,residual_i,residual_score,treatment,owner,annex_a_controls,review_date
R-001,prod-postgres,Customer PII exfiltration via stolen DB credentials,4,5,20,IAM MFA; encryption at rest; secret rotation,2,5,10,modify,head-eng,A.8.2 A.8.24 A.8.28,2026-06-30
R-002,laptops,Malware on unmanaged laptop spreads to internal systems,3,4,12,MDM; EDR; automatic patching,2,3,6,modify,it-mgr,A.8.7 A.8.8 A.8.16,2026-09-30
R-003,aws-infra,Public S3 bucket exposes customer backups,3,5,15,Block Public Access org policy; config scanning,1,5,5,modify,platform-lead,A.8.9 A.8.16,2026-06-30
Here’s the complete register for the Part 1 company. Note how inherent scores are high, existing controls drive residuals down, and one risk is accepted:
| ID | Risk | Inherent L×I | Existing controls | Residual L×I | Treatment | Owner | Review |
|---|---|---|---|---|---|---|---|
| R-001 | Customer PII exfiltrated via stolen DB credentials | 4×5=20 Critical | MFA on AWS IAM, encryption at rest, secret rotation every 90 days | 2×5=10 High | Modify — add DB activity monitoring (A.8.16) | Head of Engineering | 2026-06-30 |
| R-002 | Ransomware on unmanaged laptop spreads laterally | 3×4=12 High | Kandji MDM, Defender EDR, auto-patching, disk encryption | 2×3=6 Medium | Modify — quarterly phishing simulation (A.6.8) | IT Manager | 2026-09-30 |
| R-003 | Misconfigured S3 bucket exposes customer backups publicly | 3×5=15 Critical | Org-wide Block Public Access, nightly config compliance scan | 1×5=5 Medium | Modify — alert on any config drift within 1h (A.8.16) | Platform Lead | 2026-06-30 |
| R-004 | Extended AWS region outage makes SaaS unavailable >24h | 2×4=8 High | Multi-AZ deployment, daily tested backups | 2×3=6 Medium | Retain — multi-region cost not justified at current ARR; documented in management review | CTO | 2026-03-31 |
| R-005 | Departing engineer retains access to production | 4×4=16 Critical | SSO with Google Workspace, offboarding checklist, quarterly access review | 2×4=8 High | Modify — automate deprovisioning via SCIM within 4h (A.5.18) | Security Lead | 2026-06-30 |
| R-006 | Vendor payroll SaaS suffers breach leaking employee PII | 2×2=4 Medium | Vendor review at onboarding, minimal data shared | 2×2=4 Medium | Retain — within appetite; monitored | IT Manager | 2026-12-31 |
Clause 6.1.2 e) requires you to pick from four options for each risk. Getting the vocabulary right matters — auditors specifically check that you don’t just “modify” everything.
Apply controls to reduce likelihood or impact. Defensible when the residual score is above appetite and a cost-effective control exists. This should be your most common decision — typically 70–80% of rows. Always name the control and a target date.
Decide the residual risk is within appetite and do nothing further. Defensible when: the residual score falls in your accepted band, the cost of treatment exceeds the benefit, and — critically — the acceptance is documented and signed by the risk owner. “We accept everything green by default” is fine if your methodology says so; a silent acceptance of a High risk is a finding. Revisit accepted risks at the review date or when context changes.
Stop the activity that creates the risk. Defensible when the activity isn’t core to the business. Real examples: decommissioning a legacy admin portal nobody uses, dropping a data-sharing integration that can’t meet contractual security terms. If avoiding the risk would mean shutting down your core product, this isn’t avoidance — it’s fantasy. Auditors have seen it.
Shift some of the financial or operational burden to a third party. Defensible when it genuinely reduces exposure: cyber insurance for financial impact, contractual liability clauses with vendors, or moving from self-hosted Postgres to a managed database service. Not defensible as: a way to make risk disappear. You can outsource the infrastructure, but not accountability to your customers or the regulator. If R-004 were “shared” with a cloud provider’s SLA, you’d still own the customer-facing availability risk.
Small habits that make the difference between “documented” and “lived”: