About this article: This article is reviewed against the cited sources. This is source research, not lab validation. Check the stated versions and your own requirements before making changes.
A useful post-incident report lets you judge, without taking the provider's word, three things: what happened to your systems and when, why it happened, and whether the promised fix is specific enough to check. A vague summary usually fails on one of those. It gives no times, names a trigger instead of a cause, or promises improvements without owners or dates.
This article proposes a set of headings you can ask a provider to report against. The headings are our recommendation. They are built from what large technology companies publish about their own post-incident reviews. We found no source that fixes a delivery deadline for a customer report, so agreeing one is part of the recommendation.
Why it matters
An incident that is explained only as resolved leaves the customer unable to decide anything. Should the outage be treated as a one-off or as evidence of a weakness? Does the application need a different design? Is the next review of the contract due?
Google's engineers put the cost of the missing step plainly in the SRE Workbook, quoting Ben Treynor Sloss, Google's VP for 24/7 Operations: "To our users, a postmortem without subsequent action is indistinguishable from no postmortem". The customer version of that point is that a report without checkable actions is a statement of regret.
Three different documents
People use the word report for three different things, and it helps to separate them.
- Status-page updates are written during the incident, to inform. They are short and often revised.
- The internal post-incident review, often called a postmortem, is written by the provider's own engineers for learning. Google's SRE book defines it as "a written record of an incident", covering its impact, the actions taken to mitigate or resolve it, the root causes and the follow-up actions to prevent a recurrence. It is blameless, meaning it focuses on contributing causes rather than on indicting individuals.
- The customer-facing review is a version of the second for people outside the provider. PagerDuty's public guide shows that the two are separate products: its internal process includes an external message to customers that is written and reviewed before it is sent, as a distinct item.
It is reasonable to ask for the third. You should not assume you will get the second, which may contain information about other customers and internal systems.
What the large providers publish
Public examples show the shape of a customer-facing review.
Microsoft Azure keeps post-incident reviews on its status history page, each retained for five years. Its reviews use headings along these lines: "What happened?", "What went wrong and why?", "How did we respond?" and "How are we making incidents like this less likely or less impactful?". The response section is a timeline with UTC timestamps. The corrective-actions section marks each item as completed or gives an estimated completion date.
Azure's reviews also address the customer's side. The one we read states that there was nothing customers could have done to avoid or minimise the impact of that incident, and in general it points to guidance and notes that the impact times are the overall incident duration and "not specific to any individual customer".
Amazon Web Services publishes post-event summaries only for broad events. Its page says AWS is committed to a public summary for issues with a broad and significant impact, and that each will provide "the issue's scope of impact, factors that contributed to the issue and the actions taken to address the risks identified", and stay available for a minimum of five years. Smaller incidents that affect only you may not get a public document at all.
Two lessons follow. First, providers that publish these reviews treat a structured account of impact, cause, response and corrective action as normal. Second, the public ones cover broad incidents, so a smaller provider and a narrower incident calls for you to ask.
Headings to ask for
Our recommendation is to request a report with these headings, adapted to the incident.
- Summary and status. One paragraph: what failed, for whom, whether it is closed.
- Impact on your systems. Start and end times in a stated time zone, and how they were measured. Provider-wide durations are not the same as your own, as Azure's wording above shows.
- Timeline. Detection, escalation, mitigation and recovery, each with a time and the evidence behind it. PagerDuty's guide asks for a timeline of "important changes in status/impact and key actions taken by responders", and, for each item, a metric or other source for the data.
- Trigger and cause. The trigger is what set it off. The cause is why the system was vulnerable to that trigger. Google's review criteria ask, among other things, "Was the root cause sufficiently deep?"
- Detection. How the provider found out, and how long that took. A monitoring failure is itself a finding.
- Data. Whether any data was lost, changed or accessed, and how the provider knows.
- Corrective actions. Each with an owner role, a due date, a status and a way to verify it.
- What you should do. Anything the customer could change.
- Open questions. What the provider does not yet know.
- Contact and review date. Who can answer questions, and when the actions will be reviewed.
For a security incident, add what personal information may be involved. Our companion article on notifiable data breaches explains why that matters.
NIST's incident response profile supports the general idea. It says to prepare an "after-action report" at the end of recovery that documents the incident, the response and recovery actions taken, and lessons learned; and in the communication category it says to follow the rules in contracts for incident information sharing with suppliers. That second point is your lever: put the report into the contract.
When to ask for a review
Not every blip deserves a formal report. Google's SRE book suggests that organisations should define in advance when a postmortem is required, so that everyone knows. Its example triggers include user-visible downtime or degradation beyond a threshold, data loss of any kind, an on-call engineer having to intervene (such as rolling back a release), a resolution time above some threshold, and a monitoring failure. It also notes that any stakeholder may request one.
Our recommendation is to borrow that idea for the contract. Agree the thresholds that entitle you to a written review, for example any outage of a stated length, any loss or exposure of data, and any incident where the provider's own monitoring did not detect the problem first. Without agreed thresholds, the provider decides alone whether you receive a review, and a quiet incident is the one most likely to be left unexplained.
Timing: early summary, full review later
A complete review takes time to write well. PagerDuty's internal process, as an example of one company's practice, schedules the review meeting "within 3 calendar days for a SEV-1 and 5 business days for a SEV-2", and the written postmortem is reviewed at that meeting. That is one company's internal timetable, not a standard for customer reports.
A sensible customer arrangement (our recommendation) has two stages: a short written summary soon after the incident, stating what is known and what is not, followed by the full review once the investigation is complete, with both dates agreed in advance. If the second date slips, the provider should say so and give a new one.
Judging whether the fix is real
Google's SRE Workbook gives an example of a weak action list. Its criticisms translate well for a customer reading a provider's list:
- Actions that are only mitigations, with nothing preventing a recurrence.
- Vague verbs such as improve or make better, which make success impossible to measure.
- Every action at the same priority, so nobody can tell what comes first.
- No owner and no tracking, so actions are forgotten.
- An action that tries to make people less error-prone, when changing automated systems and processes is generally more reliable.
Apply that to what you receive. A good action says what will change, by when, and how the provider will show it worked.
Warning signs in a vague summary
- No times, or times with no time zone.
- A single named cause, often human error, with no contributing factors.
- Language such as isolated incident without evidence.
- Promised monitoring improvements with no description of what will be monitored.
- Actions with no dates.
- No statement about data.
- No mention of how the provider detected the issue.
Questions to put to your provider
- When an incident affects us, will we get a written review, and within what time?
- Which headings does your standard review use, and may we have a recent example with other customers' details removed?
- How do you measure the impact on our systems, not only on the platform?
- Do your reviews list each corrective action with an owner and date?
- How do you tell us when an action is completed?
- Will you preserve the logs behind the timeline for a stated period?
A worked example (illustrative)
The following is invented. After a two-hour outage, a provider sends a customer a three-line email: the issue was caused by a configuration change, it has been rolled back, and monitoring will be improved.
The customer replies asking for a review against the headings above. The second version states the impact window for the customer's service in AEDT with the source of those times, a timeline with detection at ten minutes, the contributing causes (a change with no staged rollout and a missing alert), and three actions with a role, a date and a check each. It also states that no data was lost, and says how that was confirmed. The customer can now decide that the change process is the weakness, and ask for a date by which the staged rollout will be demonstrated.
Read next
Related articles you may want next.
- TroubleshootingThe first hour of a suspected security incident on a hosted serverWhat to do, preserve and avoid in the first hour of a suspected compromise on a hosted server, before your provider or responders take over.Read article
- TroubleshootingWindows event logs: retention, forwarding and keeping the evidence an investigation needsHow long Windows event logs last by default, why they overwrite, and how forwarding and size settings keep evidence for an investigation.Read article
- TroubleshootingSQL Requests Are Waiting: Find the Head Blocker Before Ending SessionsBuild a time-correlated blocking chain and identify the transaction owner before deciding on intervention.Read article
Risks and limits
- A report is written by the party that had the incident; it may be incomplete or favourable. Independent evidence, such as your own monitoring, is still needed.
- Large providers publish reviews for broad incidents; smaller providers may not. Detail may be limited by confidentiality and by other customers' privacy.
- The structures we cite are examples, not standards. No source we read fixes a deadline for delivery.
- A blameless internal review is a learning tool; a customer report should still name the failed systems and decisions.
Next step
When you next receive a vague incident summary, reply with the headings above and ask for a review against them, including a delivery date. Keep a copy with the contract. If you are reviewing a hosted service, the Managed Services page is a starting point for that conversation; ask for the scope in writing.
Sources
- Postmortem Culture: Learning from Failure, Site Reliability Engineering, Google, date unavailable.
- Postmortem Culture: Learning from Failure, The Site Reliability Workbook, Google, date unavailable.
- Azure status history, Microsoft Azure, as served on 15 November 2026.
- Post-Event Summaries, Amazon Web Services, date unavailable.
- Postmortem Process, PagerDuty Incident Response Guide, PagerDuty, date unavailable.
- SP 800-61 Rev. 3, NIST, April 2025.
All accessed 15 November 2026.
Learning notes
- A post-incident report should let you check the fix, not just accept it.
- A trigger is what set the incident off; a cause is why the system was vulnerable.
- Ask for your own impact window, not just the platform's.
- Put the report, its headings and its delivery time into the contract.
This article was originally published on .