Incident Response Checklist

Name a commander who is not debugging, keep a timestamped log, restore service before finding the cause, and treat data as involved until you can show it is not.

The interactive tool loads here when JavaScript runs.

About this tool

What is the Incident Response Checklist?

Incidents go badly for organisational reasons long before they go badly for technical ones. Three people quietly investigating the same thing, nobody talking to customers, and a restart that destroyed the only evidence. So the first section of this list is roles and communication rather than diagnosis, because that is what makes the technical work possible.

The order is deliberate and contradicts instinct. Restore service first and understand it later. Root cause is a review activity, not an outage activity, and the time spent finding it is time the customer is still down.

  • Roles and communication first, because that is what makes the technical work possible
  • Says plainly to restore service before finding the cause
  • Includes evidence preservation, which the pressure to restart destroys
  • A breach section built around statutory notification clocks
  • Editable and exportable as PDF, CSV, Markdown or a Google Form
Read the guide: Why Checklists Work, and How to Use One Properly

Updated

How to use the Incident Response Checklist

  1. Declare the incident in one named channel and name a commander who will not debug.
  2. Name a scribe to keep a timestamped log, and post a customer-facing update within minutes.
  3. Restore service first: roll back, fail over, or disable the feature. Preserve evidence before restarting.
  4. If data may be involved, bring in legal and privacy immediately and rotate exposed credentials.
  5. Afterwards, hold a blameless review within a week and give every action an owner and a date.

Worked examples

Alerts just fired

Input

First ten minutes

Output

Incident declared, commander and scribe named, first update posted

Coordination before diagnosis is what prevents three parallel investigations.

Service is down

Input

Stabilise section

Output

Rolled back, evidence preserved, one change at a time

Root cause is a review activity, not an outage activity.

Possible data exposure

Input

Breach section

Output

Treated as a breach, legal engaged, credentials rotated

Statutory clocks often start from the moment you became aware.

Background and insight

The single most valuable role is a commander who is not debugging. Their job is coordination, decisions and communication, and a commander with their hands in the system stops coordinating within minutes. The second is a scribe keeping a timestamped log of what was observed, what was changed and what happened next, because the review afterwards is built from that log and memory is not good enough.

Two technical disciplines prevent most compounding. Make one change at a time and announce it first, or you cannot tell which fix worked and occasionally they interact badly. And preserve evidence before restarting anything: logs, metrics, a snapshot, the failing request. A restart is often what destroys the only record of what happened, and the pressure to restart is highest at exactly that moment.

If data may be involved, the posture changes. Treat it as a possible breach until you can demonstrate otherwise, because deciding early that it was nothing and discovering later that it was not removes every option you had. Notification duties run on short statutory clocks in many jurisdictions, often days from awareness, which is why legal and privacy are involved immediately rather than after the technical work. And saying nothing public about scope until you know it is not evasion: an early wrong number has to be corrected in public, which costs more trust than saying the facts are still being established.

On communication, a committed interval beats a good update. Every thirty minutes, even when the update is that there is no news, because silence is what makes people escalate and flood support. One status page and one voice: support, sales and engineering saying different things is its own incident.

Afterwards, blameless review is not politeness. It is the only way to get the detail you need, because people hide the facts that matter when the review appears to be looking for someone to hold responsible. The four numbers worth extracting are time to detection, time to first action, time to mitigation and time to resolution; those gaps are where the improvements live. And if a customer detected it before your monitoring did, that is itself the most valuable finding.

Common use cases

  • Small teams without a formal process

    The roles and the communication cadence, which matter more than tooling.

  • On-call rotations

    Something to open at 3am when nobody is thinking clearly.

  • Suspected data breaches

    What to do first, and what not to say publicly, while the clock runs.

  • Running the review

    A blameless structure with the four timings that actually drive improvement.

Tips and limitations

  • The commander must not be debugging. Hands in the system means coordination stops.
  • Announce each change in the channel before making it, and make only one at a time.
  • Post an update every thirty minutes even when there is no news. Silence causes escalation.
  • Ask what changed in the last few hours. Most incidents have a recent change behind them.
  • If a customer detected it first, that is the finding. Fix the detection, not just the fault.

Sources and further reading

  1. Computer Security Incident Handling Guide (SP 800-61r2), National Institute of Standards and Technology. The standard incident handling lifecycle
  2. Incident management, National Cyber Security Centre. Roles, communication and preparation
  3. Personal data breaches, Information Commissioner's Office. Notification thresholds and the reporting clock

Frequently asked questions

What should happen in the first ten minutes of an incident?

Declare it out loud in one named channel, name a commander who will not be debugging, name a scribe to keep a timestamped log, agree a severity, and post an initial customer-facing update. All five are coordination rather than diagnosis, and all five make the diagnosis faster.

Should we fix it or find the root cause first?

Fix it. Roll back, fail over, disable the feature or scale up, and understand it afterwards. Root cause is a review activity. The one caveat is preserving evidence, logs, metrics and the failing request, before a restart destroys it.

Why does the incident commander not work on the fix?

Because coordination stops the moment their hands are in the system. Their job is deciding, sequencing changes, and communicating, and all three are full-time during an incident. It is the most commonly skipped role and the most valuable.

What should we do first if data may be exposed?

Treat it as a breach until you can show it is not, bring in whoever owns legal and privacy immediately because notification clocks often run in days from awareness, rotate credentials and keys that may be exposed, and say nothing public about scope until you know it.

How often should we update the status page?

On a committed interval, typically every thirty minutes, and hold it even when the update is that there is no news. Silence is what makes customers escalate and floods support faster than you can answer. One page, one voice.

Is there an in-depth guide to the Incident Response Checklist?

Yes. “Why Checklists Work, and How to Use One Properly” covers the background, how it works, common pitfalls and worked examples. Read the full guide.