An audit record should help someone explain an important change later. A stream of generic “updated” messages is difficult to investigate, while copying every request body can create an unnecessary second store of private customer data.
Choose the events that matter to accountability and recovery. Then define a consistent record shape that preserves who acted, what changed, where it happened, and whether the change succeeded.
Start with decisions the business may need to reconstruct
Typical candidates include access grants, ownership transfers, billing overrides, important status changes, exports, and deletion. Ordinary page views may belong in a different telemetry system with different retention and access rules.
For each audit event, define the actor, organisation, resource, action, time, outcome, correlation reference, and the minimum before-and-after context needed. Use stable identifiers so renamed users or resources do not break the history.
Review a fictional request cancellation
Imagine a staff member cancels a service request after speaking to a customer. The business may later need to know the previous status, new status, responsible staff member, reason category, and related request. It may not need the entire customer's address and conversation copied into the audit event.
Test cancellation, rejection because the request is already completed, an interrupted response after a successful write, and a support-admin repair. The audit should distinguish a successful state change from a denied attempt and preserve the real support actor rather than impersonating the customer owner.
If the application writes the business change and audit event separately, consider what happens when one succeeds and the other fails. Choose a consistency mechanism appropriate to the architecture and verify it with failure injection in a controlled environment.
Design retrieval and retention with the event
Staff need to filter by organisation, resource, actor, action, and time. An archive nobody can search is weak operational evidence. At the same time, permission to view a service request should not automatically grant access to every administrative audit event.
Define how deleted accounts and removed memberships appear in history. Avoid relying exclusively on current profile lookups. Apply the organisation's retention and privacy rules to the audit store, including exports and backups where relevant.
The tradeoff is forensic detail versus complexity and exposure. Rich change payloads may speed investigation but retain more sensitive data. Minimal event names are easy to store but may leave the reviewer unable to explain the outcome. Use a few realistic investigations to find the useful middle.
The StackFlight portfolio account provides related private-SaaS context. The cancellation model is an illustrative design, not a description of a client's incident history.
Questions about SaaS audits
Is an application log the same as an audit trail?
Not automatically. Audit records need defined meaning, access, retention, and consistency with important business actions.
Should failed attempts be retained?
Some are useful, especially sensitive administrative attempts. Record their outcome clearly and keep noise manageable under the chosen policy.
What should be reviewed first?
Trace one important change through the stored record, audit event, support view, and deletion lifecycle. Bring missing context to a data-model review.