The files arrived late. Someone restored delivery. The client received an explanation. The team still needs to understand what happened.
A postmortem gives that work a record. Keep the impact, timeline, unresolved questions, and corrective actions together. Make each conclusion answerable to evidence.
Google's SRE book describes a postmortem as a written incident record covering impact, response, causes, and follow-up actions. Its blameless approach examines contributing conditions without indicting individuals or teams.
The method below adapts that practice for a small service business. The template and review rules are editorial proposals. The delayed-delivery incident is fictional. It describes no client engagement or measured business result.
Open the Record After Stabilizing the Work
Give immediate response one job: restore the affected service under an authorized plan. Give the postmortem another: explain the incident and assign learning work.
During response, capture short timestamped observations when doing so does not obstruct recovery. Preserve the available records. Leave the complete narrative until someone can review it.
Choose postmortem triggers before the next incident. A small team might use a missed client delivery commitment, a failed recovery attempt, or a failure discovered by the client. These are proposed triggers. Set thresholds around the service your team provides.
Also let someone request a review when an incident exposes an unexplained weakness. A recovered workflow can still contain an unknown dependency.
Name an author and a reviewer. The author gathers the record. The reviewer checks the evidence, impact boundary, and proposed actions. Confirm who can authorize changes that emerge from the review.
A business continuity tabletop exercise rehearses an imagined interruption. This record examines an event that has happened. Keep those purposes distinct when choosing the evidence.
Define the Impact Before Explaining the Cause
Start with the promised output. State what the recipient could not do and when that limitation ended.
For the fictional incident, a service team prepares a scheduled document package. Six client packages are due at 10:00. At the deadline, four packages remain undelivered. The other two have confirmed delivery receipts.
All four delayed packages reach their intended destinations by 11:25. Recipient receipt records establish delivery. They do not establish when anyone read the documents or whether downstream work was delayed.
Record those limits. The known impact is four missed delivery deadlines. Downstream client impact remains unknown until someone obtains evidence. Do not convert an unreturned question into a claim of no harm.
Separate the delivery interruption from the response. The deadline slip begins at 10:00 in this example. Detection occurs at 10:07. Recovery completes at 11:25. Those timestamps describe different events.
Keep the affected population explicit. Four of six scheduled packages missed this deadline. The record says nothing about other delivery cycles unless the reviewer examines them.
Build a Timeline With Evidence Labels
Use a single time zone and a named clock reference. Record when an event occurred separately from when someone reported it.
The following timeline is entirely invented. Its evidence labels demonstrate a recording method rather than reporting actual logs.
| Time, UTC | Confirmed event | Evidence or limitation |
|---|---|---|
| 09:40 | Six packages enter the fictional delivery queue | Queue snapshot records six identifiers |
| 09:44 | Two packages receive delivery confirmations | Receipt records match two queue identifiers |
| 09:46 | Four packages remain queued | Snapshot shows queued status; cause unknown |
| 10:00 | Four scheduled deliveries miss their deadline | Deadline record compared with available receipts |
| 10:07 | A coordinator notices missing confirmations | Timestamped incident note |
| 10:12 | The response owner pauses automatic retries | Response note records the decision and scope |
| 10:28 | A sample package reaches a controlled destination | Test receipt; no proof of full recovery |
| 10:50 | Manual delivery starts for the affected packages | Approved response record |
| 11:25 | All four delayed packages have delivery confirmations | Reconciliation against the original queue |
Add interpretations below the table. For example: “The retry behavior may have contributed to the delay.” Label that sentence as a hypothesis with a question owner.
Keep unknowns visible too. This fictional record does not establish when or why the four packages stopped progressing. The 09:46 snapshot records their state at one moment. It does not prove that pausing retries repaired anything.
If accounts conflict, preserve both accounts until evidence resolves them. An interview recollection can guide investigation while remaining different from a timestamped system observation.
Record the limits of each evidence source. A missing receipt can establish an unresolved delivery state without proving a failed delivery. A screenshot shows one observed moment. It does not establish the state throughout an interval.
Ask the author to separate the earliest confirmed failure from the possible start of the interruption. If the last successful check was earlier, record that interval. Leave the start unknown when the available records cannot narrow it.
Give every unresolved interpretation a specific next question. Instead of “check the system,” ask which record could show whether the delivery service accepted the missing packages. The investigation owner can then state whether that record exists and what it establishes.
Examine the Conditions Around Each Decision
Ask what the responding person could see at the time. Then ask which information or control was missing.
In the fictional incident, the queue shows that work was accepted. The coordinator's checklist treats acceptance as completion. Nobody compares scheduled packages with delivery receipts before the deadline.
That distinction gives the review a concrete condition to examine. Acceptance into a queue and confirmed delivery describe different states. The procedure uses evidence for the earlier state to close the later task.
Another condition concerns recovery. The team has a manual delivery procedure, but its owner is absent from the working checklist. The response owner spends part of the incident confirming who can authorize it.
Neither observation establishes the original queue failure. Both describe conditions that the fictional team can address while that cause remains unresolved.
Trace these conditions through a business system dependency map. Follow the promised output backward to required data, delivery service, confirmation record, and approval authority.
Avoid forcing every finding into one root cause. Keep separate questions for the queue interruption, missed detection, and recovery decision. Each may need different evidence and a different action.
Use a Postmortem Template With Inspectable Fields
Copy the fields below into one incident document. Link controlled evidence where appropriate. Exclude personal data and private log contents from broadly shared versions.
| Field | What to record |
|---|---|
| Record identity | Incident identifier, author, reviewer, version, and review date |
| Service and boundary | Promised output, affected workflow, and exclusions |
| Impact | Affected items, missed commitments, duration, and unknown downstream effects |
| Detection | First observed symptom, observer, timestamp, and missing earlier signals |
| Response | Decisions, authorized scope, mitigations, and evidence used to end response |
| Timeline | Event time, evidence reference, and confidence label |
| Contributing conditions | Observed weaknesses and the evidence supporting each one |
| Open questions | Unresolved cause, evidence needed, question owner, and review point |
| Corrective actions | Specific change, owner, target date, verifier, and completion evidence |
| Follow-up | Accepted evidence, remaining limits, and next review decision |
Keep a short summary above the fields. A new reader should identify the failed output and current state before reading the timeline.
Write “unknown” when the evidence is absent. State the attempted check if it helps the reviewer understand the gap. A blank cell leaves the reason for missing information unclear.
Version the record when later evidence changes a conclusion. Preserve the earlier interpretation in a dated note. Readers need to see why the action plan changed.
Turn Findings Into Bounded Actions
An action needs an owner who accepts the work and a verifier who knows what completion means. A role can hold both responsibilities on a small team. Make that choice explicit.
“Be more careful” gives the verifier no observable change. “Compare scheduled package identifiers with confirmed receipt identifiers before closing delivery” identifies a check.
For the fictional incident, propose these actions:
| Action | Owner and target | Verification evidence |
|---|---|---|
| Add a receipt reconciliation step | Delivery coordinator, before the next scheduled cycle | Controlled exercise identifies a package with no receipt and leaves delivery open |
| Confirm manual-delivery authority | Service lead, before approving the revised response procedure | Approved procedure names the authorizing role and backup |
| Investigate stalled queue behavior | Technical owner, review at the next incident follow-up | Reproduction evidence or a documented unresolved question |
Keep investigation separate from correction. Asking a technical owner to inspect queue behavior does not establish that the queue has been repaired.
Before accepting an action, check the dependency it introduces. A receipt comparison requires accessible receipts. A backup authorizer requires someone available and authorized to make that decision.
Put accepted procedural changes into the team's SOP documentation. Keep the postmortem as the reason for the change. The procedure should contain the current instructions.
Review One Completed Action and One Open Action
At the fictional follow-up, the delivery coordinator presents a controlled exercise. The sample contains six queued package identifiers and only five delivery receipts.
The revised checklist leaves the unmatched package open. The coordinator records the missing receipt, names the next check, and avoids marking the cycle complete. The reviewer accepts the reconciliation action for this exercise.
That evidence shows the revised check detected this sample mismatch. It does not demonstrate performance across every live delivery condition. Record the exercise boundary beside the completion decision.
The technical investigation remains open. The owner has not reproduced the stalled queue behavior. No available record establishes whether the failure began before or after the delivery service accepted the four packages.
Keep that uncertainty in the action register. Record what was inspected and what evidence is still needed. The service lead must decide whether to continue investigation, narrow it, or accept a documented limit.
Assign the next review point. Leaving the task open without an owner or date gives the team no next action. Closing it because the symptom disappeared invents certainty.
Finish the Review Without Claiming Immunity
Google's postmortem guidance includes review for completeness, impact, and action quality. Apply those questions to this smaller record before accepting it.
Ask whether a reader can trace the impact to evidence. Check whether the timeline distinguishes observation from interpretation. Confirm that each accepted action has an owner and a completion condition.
Then separate document acceptance from action completion. The narrative may be ready while the queue investigation remains unresolved. State both statuses in the same record.
Share a suitable version with people who need the learning. Keep restricted evidence in its controlled location. A client-facing explanation can summarize confirmed impact and response without copying internal investigation material.
Reopen the record if new evidence changes the impact or causal account. Link a later incident when it tests the same assumption. Completed actions reduce particular weaknesses only within the evidence available.
Keep the open questions owned.
Sources
- Google SRE: Postmortem Culture, Learning from Failure. Incident records, blameless investigation, and review of impact and follow-up actions. Checked 2026.10.28. Google's practice informs this editorial adaptation.