Running the Programme, Not the Project

Lesson 7: Running the Programme, Not the Project

The first report is a project with a finish line. Everything after it is a programme, and the programme is where companies quietly succeed or quietly rot. The test is simple: on any given day, could you produce a stranger's requested evidence for any control in scope within an hour, for the whole current period? If the answer is “give us three weeks”, you are running a project.

Assign owners, not tasks

Every control needs one accountable person — not a team, a name. The work distributes predictably:

OwnerTypical controls
Head of security / CISO (or fractional)Programme policy, risk assessment, incident response, vendor reviews, board reporting
Platform / DevOpsAccess provisioning and reviews, encryption, logging, vulnerability SLAs, backup and restore tests, change pipeline
Engineering leadsCode review, deployment approval, SDLC and threat modelling, dependency hygiene
IT / opsEndpoint management, asset inventory, onboarding and offboarding execution, facilities
People / HRBackground checks, training completion, acknowledgements, performance reviews, termination feeds
Legal / financeContracts, DPAs, vendor terms, insurance, customer commitment language

Put it in one register: control ID, description, owner, frequency, system of record, evidence location, last performed, next due. That register is your operating system. Without it, every audit becomes archaeology.

The evidence calendar

CadenceWhat runs
Continuous / automatedConfig drift and CIS checks, cloud posture, endpoint compliance, dependency scanning, log shipping, uptime and alerting
Per eventAccess requests and approvals, terminations, incidents, emergency changes, vendor onboarding, customer deletion requests
MonthlyVulnerability review and remediation tickets, patch compliance report, backup success review, new-hire training within window
QuarterlyUser access reviews across in-scope systems, privileged access review, control self-assessment sample, risk register review
AnnualRisk assessment, policy review and re-approval, security awareness refresher, penetration test, DR/BCP test, tabletop exercise, vendor reassessments, management review, board update, insurance renewal
Design for the sample you'll be asked for. The auditor tests a control by pulling a population and sampling instances. So make the population trivially producible: one query or export per control, from the system of record, covering arbitrary date ranges. Companies that keep evidence in screenshots and email attachments spend weeks assembling; companies with a queryable system of record spend an afternoon.

Compliance automation: what it buys, what it can't

Platforms in this category (Vanta, Drata, Secureframe, Thoropass, Sprinto and others) integrate with your cloud and SaaS stack, run continuous tests against a control library, host policies and training, track vendors, and expose an evidence vault your auditor can read. Realistically they:

  • Do cut evidence collection from weeks to days, standardise policy drafting, keep the calendar, and make gaps visible between audits rather than during one.
  • Do not design your controls, fix your architecture, attend the interviews, decide your scope, or take responsibility for the assertion. They also do not prevent exceptions caused by human process — a platform can remind you about an access review; only a person can perform it.
Keep the auditor's judgment independent of your vendor. The AICPA has publicly flagged business arrangements between CPA firms and compliance tool providers, and has referred unlicensed or non-peer-reviewed firms to regulators. Practical rules: choose your auditor yourself rather than accepting a single bundled recommendation, confirm the firm is licensed and peer-reviewed, and be sceptical of an engagement that promises a report in two weeks with no evidence period — “quick-turn” SOC reports are exactly what peer reviewers are now scrutinising.

The exception list to design against

These recur across nearly every report. Each is a scheduling problem, not a technology problem:

ExceptionRoot causeFix
Late access removal after terminationHR feed not connected to IdPAutomate termination → deprovision; alert on accounts with no HR match
Missed quarterly access reviewNo owner, no calendarOwner + recurring ticket generated automatically, with a completion SLA
Unticketed production changeConsole hotfix habitBreak-glass path with after-the-fact ticket; audit-log alert on out-of-band deploys
Vulnerability SLA breachesBacklog with no severity tiersSeverity definitions, auto-ticketing, monthly review, documented risk acceptances with expiry
Stale asset inventoryManual spreadsheetCloud asset export reconciled monthly to the list
Training or policy acknowledgements incompleteNew hires, contractorsEnrolment in onboarding automation; completion report reviewed monthly
DR test not performedNever scheduled, or fearedBook it early in the window; a failed test with a remediation plan is a better artifact than no test
Vendor review overdueInventory decayTiered reassessment dates in the register, owned by security

Incidents during the window

You will probably have one. The finding is never “an incident occurred”; it is “the incident was handled outside the documented process” or “no record exists”. So: declare it, log it, follow the plan, preserve evidence, communicate on the documented channels, hold the post-incident review, and track corrective actions. A well-documented incident with a clean response is often the best evidence in the report that your controls work.

Selling with it

  • Trust centre — a public page listing your frameworks, report availability, subprocessors, and security contacts. It removes the first three emails of every enterprise deal.
  • NDA-gated distribution. Reports are restricted-use; most companies require an NDA and log who received which version.
  • Answer bank. Map questionnaire answers to report sections and control IDs, so sales and security answer consistently. Inconsistency between a questionnaire and the report is the fastest way to lose a deal's trust.
  • Know your CUECs. Hand them to solutions engineers; they define what the customer must configure.

Leverage it across frameworks

The controls you build for SOC 2 are largely the controls ISO 27001, HIPAA and PCI DSS want; AICPA publishes mappings to ISO 27001 and GDPR, and overlapping coverage is typically the majority of the work. Common patterns: Security + Availability is ~90% of ISO 27001 from a test perspective; HIPAA needs the privacy and breach-notification layer added; PCI needs the cardholder-data scope carved out. If you ship AI, the new SOC 2 + AIUC-1 joint engagement is where enterprise buyers are heading, with annual renewal and quarterly agent retesting.

A 90-day plan

  1. Days 1–14: pick scope (categories, boundary, window), appoint the programme owner, budget, and decide Type I vs Type II and the window start date.
  2. Days 15–30: gap analysis against the in-scope criteria; inventory assets, systems and vendors; list the controls you can already evidence and the ones you can't.
  3. Days 31–60: adopt the policy set with approvals; stand up identity, MFA, logging and change-control basics; start the monthly and quarterly cadences so the clock runs.
  4. Days 61–90: book the pen test and DR test, run the risk assessment, engage the auditor for a scoping call, and begin drafting the system description.
The one metric that predicts a clean report: percentage of controls with evidence for every required instance in the current period, measured monthly. If that number sits at 100% for two quarters, your audit is a formality. If it sits at 60%, no amount of preparation in the final month will fix it — because the instances you missed are in the past and cannot be recreated.

🧠 Knowledge Check

1. What is the most reliable predictor of a clean Type II report?

2. Which of these can a compliance automation platform NOT do for you?

3. A production incident occurs mid-window and is resolved with no ticket. What is the audit-relevant problem?

Further Reading