Most startups write their first incident response plan during their first real incident. A cloud key shows up in a public repository, a customer reports that data looks wrong, or a bad deploy takes production down at two in the morning. Nobody is sure who is in charge, where to talk, what to shut off or who needs to be told. The hours that follow decide whether the incident is a bad night or a company-defining event.
This guide gives you a plan you can copy and adapt in an afternoon: roles, severity levels, a first-hour runbook, short playbooks for common incidents, communication templates, a post-incident review format and a tabletop exercise to test it. It is practical guidance, not legal advice, and you should adjust it to your own systems, customers and obligations.
What is an incident response plan?
An incident response plan is a short, written set of roles, steps and contacts that tells your team what to do when something goes wrong, so people act in a clear order instead of improvising under pressure.
An incident can be a security event, such as a leaked credential or malware, or an operational one, such as an outage. One plan can cover both, with separate playbooks for the common cases. The test of a good plan is whether a tired engineer can follow it at night.
Why do startups need an incident response plan?
Startups need one because customers, auditors and regulators increasingly expect it, and because speed in the first hour limits damage.
- Enterprise security questionnaires ask whether you have a documented, tested plan. See our guide to SOC 2 for startups.
- Some laws and contracts set deadlines for reporting certain incidents, so you need to know quickly what happened.
- A clear plan reduces panic, duplicated work and unforced errors such as deleting evidence.
NIST's incident response guidance, SP 800-61 Revision 3 (April 2025), now frames incident response around the Cybersecurity Framework 2.0 functions: Govern, Identify, Protect, Detect, Respond and Recover. The older four-phase model of preparation, detection and analysis, containment, eradication and recovery, and post-incident activity is still a useful way to think. A startup plan needs to cover the same ground in plain language.
The template: what to put in your plan
1. Purpose and scope
One paragraph: what counts as an incident, which systems and data are covered and who the plan applies to. Example: "This plan covers any event that threatens the confidentiality, integrity or availability of our systems or customer data, including security events and production outages."
2. Roles
Name a person, and a backup, for each role. In a small team one person may hold several, but the roles must be clear.
- Incident Commander: makes decisions, assigns work and keeps the response organised. Does not debug.
- Technical Lead: investigates, contains and fixes.
- Communications Lead: writes internal and customer updates and handles questions.
- Scribe: records a timestamped timeline of what was seen, decided and done.
- Executive sponsor: decides on major actions such as taking a service offline and involves legal advisers.
Keep a contact list with phone numbers, not only chat handles, because chat may be unavailable.
3. Severity levels
Agree severity levels in advance so nobody argues about them during an incident. These are examples to adjust:
- SEV1 Critical: confirmed data exposure, ransomware or a full production outage. Respond immediately, assign a commander within 15 minutes and update stakeholders every 30 minutes.
- SEV2 High: a major feature is down, or a compromise is suspected but limited. Respond within one hour, with updates every hour.
- SEV3 Medium: degraded service or a contained security event. Handle in business hours.
- SEV4 Low: minor issue or near miss. Track it in the backlog and review it.
When unsure, choose the higher severity. You can downgrade later.
4. How incidents are reported
Make reporting easy. One dedicated channel, one email address and one phone number are enough. Tell every employee what to report and that reporting a mistake quickly is rewarded, not punished. A good report includes what was seen, when, where and who saw it.
5. The first-60-minutes runbook
- Declare the incident and set a severity. Name the commander.
- Open a single channel and a call for the response. Keep it separate from normal chat.
- Start the timeline. The scribe logs every action with a time.
- Assess scope: which systems, accounts and data might be affected? What is known and what is not?
- Contain: stop the spread without destroying evidence. Revoke or disable credentials, isolate hosts, block addresses, take a bad release out of service.
- Preserve evidence: take snapshots, keep logs and do not wipe or rebuild systems before you have captured what you need.
- Communicate: send the first internal update, and decide whether customers, partners or authorities must be told.
- Eradicate and recover: remove the cause, restore from known good state, rotate secrets and verify that normal operation is back.
- Monitor: watch closely for several days for signs the problem returned.
6. Playbooks for the most likely incidents
Keep each playbook to a page. Start with these three.
Leaked cloud credential or API key
- Deactivate the key. Do not delete it yet, because you need it for the investigation.
- Review audit logs for everything the key did, including from unfamiliar addresses.
- Look for new users, roles, access keys and resources created by the key, and remove them.
- Rotate any secrets the key could reach, and find where the key leaked.
# Deactivate the exposed key (example values)
aws iam update-access-key --user-name USERNAME \
--access-key-id AKIAXXXXXXXXXXXXXXXX --status Inactive
# Review what that key did
aws cloudtrail lookup-events \
--lookup-attributes AttributeKey=AccessKeyId,AttributeValue=AKIAXXXXXXXXXXXXXXXX \
--max-results 50
Our guide to secrets management in CI/CD explains how to stop it happening again.
Suspected compromised user account
- End active sessions and reset the password and multi-factor authentication.
- Revoke tokens and application authorisations.
- Check for new mail forwarding rules, new devices and unusual logins.
- Review what the account accessed, and tell the user what to watch for.
Publicly exposed storage or database
- Block public access immediately.
- Check access logs to see whether anyone outside accessed the data, and what data it held.
- Fix the configuration and add a guardrail to stop it recurring.
- Involve your legal adviser if personal or customer data may have been exposed.
For an outage caused by a bad release, roll back first and investigate afterwards. Our production readiness checklist and AWS cloud security checklist cover the controls that make these incidents rarer and shorter.
7. Communication
Keep internal updates on a fixed rhythm, such as every 30 minutes during a SEV1, even if the update is "no change". For customers, say what you know, avoid speculation and promise a next update time. A simple template:
Subject: Security incident notice: [Company name]
What happened: On [date and time, with time zone], we detected [short, factual description].
What it affected: [systems and types of data, and what was not affected].
What we have done: [containment steps taken].
What you should do: [for example, rotate credentials or watch for phishing].
What happens next: We will send another update by [time]. Contact: [email address].
Do not assign blame, guess at causes or minimise. Have one person approve every external message.
8. Legal and notification obligations
Some incidents trigger legal or contractual duties to notify. Examples include the GDPR and UK GDPR requirement to notify the supervisory authority of certain personal data breaches within 72 hours of becoming aware of them, India's CERT-In directions requiring certain incident types to be reported within six hours, notification laws in individual US states, sector rules and the terms of your customer contracts. Rules differ by location and industry and change over time. Ask your legal adviser which apply to you, write the answer into the plan with who decides, and add your cyber insurer's contact if you have one.
9. Evidence and logs
Decide in advance where logs live, how long you keep them and who can access them. Turn on audit logging for your cloud accounts and key systems before you need it, and keep copies outside the systems that might be compromised. Do not delete or rebuild affected systems until evidence is preserved.
10. Post-incident review
Hold a blameless review within about five working days, while memories are fresh. Cover:
- A timeline of what happened, from the scribe's notes.
- Impact: customers, data, downtime and cost.
- Root cause and contributing factors.
- What went well and what did not.
- Actions with an owner and a due date, such as new alerts, tighter permissions or a runbook change.
Update the plan with what you learned, and track the actions to closure. Our guide to reducing mean time to remediation and our post on real-time incident coordination go deeper.
11. Testing and maintenance
Review the plan every six months and after every significant incident or system change. Check contact details, roles and the playbooks. Run a tabletop exercise at least twice a year.
How do you run a 45-minute tabletop exercise?
A tabletop exercise is a talk-through of a made-up incident with the people who would respond, to find gaps before a real one does.
- Set the scene (5 minutes). "At 02:10 an alert shows an unfamiliar address using an engineer's cloud key to list storage buckets."
- Walk through the response (25 minutes). Add new information every few minutes: "A new user account has appeared," "A customer emails about strange activity," "A journalist asks about a breach." At each step ask who does what, which tool they use and who approves.
- Review (15 minutes). What was unclear? Which contact was missing? What would we change in the plan?
Write the gaps down as actions with owners.
What mistakes do startups make with incident response?
- Having no named incident commander, so everyone debugs and nobody decides.
- Deleting or rebuilding systems before preserving evidence.
- Debating severity instead of acting.
- Not recording a timeline, which makes reviews and legal questions harder.
- Communicating late, vaguely or inconsistently.
- Never testing the plan, or never updating it after an incident.
- Blaming the person who made the mistake, which teaches people to hide problems.
How can T-Mat Global help with incident response?
T-Mat Global Technologies is a founder-led, DPIIT-recognized technology delivery partner from India. We bring production expertise from the Fortune 100 ecosystem to startups and engineering teams across the US, UAE and UK. We help with cloud infrastructure, DevOps, CI/CD, cloud security, vulnerability remediation and production support, and engagements can be project-based, ongoing maintenance or managed infrastructure support. Our founder, Sainath Mitalakar, is an AWS Certified DevOps Engineer Professional with production expertise from the Fortune 100 ecosystem.
We can help you draft and test your plan, set up logging and alerting, write playbooks for your systems and support incident response and ongoing security operations. See our managed services and cloud security pages, and our other services.
Get a free assessment
Want help building or testing your incident response plan? Contact T-Mat Global to discuss your requirements. Visit t-matglobal.com or email info@t-matglobal.com, and book your free assessment.
Frequently asked questions
What is an incident response plan?
It is a short written plan of roles, steps and contacts that tells your team what to do when a security or operational incident happens.
What should an incident response plan include?
A purpose and scope, named roles, severity levels, reporting channels, a first-hour runbook, playbooks for common incidents, communication templates, legal and notification contacts, evidence handling and a post-incident review process.
Who should be the incident commander at a startup?
Someone senior and calm who can make decisions and coordinate people, usually the CTO or an engineering lead, with a named backup. The commander coordinates and does not debug.
How often should we test the plan?
Review it every six months and after any major incident or system change, and run a tabletop exercise at least twice a year.
Do we have to notify customers or regulators?
Sometimes. Duties depend on the type of data, your location, your customers' locations, your sector and your contracts. Ask your legal adviser which apply and write the answer into the plan.
What is the first thing to do in a security incident?
Declare the incident, name a commander, open one response channel and start a timeline. Then contain the problem without destroying evidence.
What is a blameless post-incident review?
A review that focuses on how the system and process allowed the problem, not on who made a mistake. It produces actions with owners and dates, and it encourages honest reporting.
Is NIST SP 800-61 required?
No. It is free guidance from the US National Institute of Standards and Technology, aligned with the Cybersecurity Framework 2.0, and it is widely used as a reference. Many frameworks and auditors recognise its approach.
