How to Design a Customer Status Page and Incident Communication Plan
How to design a customer status page and incident communication plan for a custom SaaS: components, severity tiers, update templates, and drills.
Design a customer status page and incident communication plan in this order: decide which parts of the product customers depend on, write tiers that decide how fast you publish, then draft the update templates before you need them. A status page is a public web page that reports whether each part of a service is working. The plan is what keeps that page honest at 2am.
Start with the components customers depend on
A component is a named part of the product that a customer would recognise and would complain about if it stopped, such as sign-in, the API, billing, or reporting. Most custom SaaS products need between four and seven public components. The test is blunt: a customer should be able to point at one row and say that is what is broken for me.
Do not mirror your internal architecture onto the page. Nobody cares that a queue worker is unhappy. If a database problem stops sign-in, the customer-facing component is sign-in.
Keep the mapping in a short internal document, sometimes called a service map, which records which backend services sit behind each public component. You will use it to decide what to mark as affected at 3am, when nobody wants to read code.
What should a customer status page display?
A customer status page should show named components with a current state each, any active incident with timestamped updates, upcoming scheduled maintenance, and a readable history of past incidents, so customers can judge reliability without contacting support.
Keep the state vocabulary small and define each state in one line. Operational, degraded performance, partial outage, major outage, and under maintenance covers almost everything. Ambiguous labels cause arguments during an incident, and arguments cause delay. If your team cannot agree whether something is degraded or a partial outage, the label is doing too little work.
Let customers subscribe rather than refresh. Email, RSS (a feed format that lets other tools poll for updates), and webhooks (an HTTP callback that pushes an update into a customer's own tooling) matter for B2B buyers who route alerts into their own on-call systems. Assume that most people who open the page during an incident arrived from a support reply, so put the link in help centre macros, error pages, and email signatures.
One rule is worth enforcing above all: never show green when customers are failing. A page that reports operational while logins fail does more damage than having no page, because it tells customers their own monitoring is wrong.
How fast should you communicate during an incident?
Publish the first public update within roughly 15 to 30 minutes of confirming customer impact, then refresh it every 30 to 60 minutes until resolution, even when the only news is that there is no news.
Speed beats completeness in the first update. "We are investigating reports of failed logins. Next update by 14:40" is a complete, useful message. Waiting for root cause means publishing after your customers have already told each other on social media.
Severity tiers turn that principle into something a tired engineer can follow. A severity tier is a label that groups incidents by customer impact, and it decides who gets paged, who writes, and how often you update. Most teams label them SEV1 to SEV4, with SEV1 the most serious.
| Tier | Typical trigger | First public update | Update cadence | Who publishes |
|---|---|---|---|---|
| SEV1 | Major outage, most customers cannot use a core component | within 30 minutes | every 30 minutes | on-call engineer, approved by engineering lead |
| SEV2 | Partial outage or degraded performance for a subset | within 1 hour | every 60 minutes | on-call engineer |
| SEV3 | Minor issue with a workaround available | within 4 business hours | at start and at resolution | support lead |
| Maintenance | Planned work with expected disruption | at least 3 days ahead | only if the plan changes | platform owner |
Every tier needs a named owner for communication, separate from the person fixing the problem. One person writes, someone else approves on SEV1, and the same text appears on the status page and in the customer email so nobody has to reconcile two versions.
What must an incident communication plan define?
An incident communication plan must define severity tiers, one named person authorised to publish, the update templates, the channels used per tier, and a review step that publishes a postmortem within roughly five business days.
Write the plan as a numbered checklist and keep it somewhere reachable when your own systems are down. A wiki hosted on the infrastructure that is currently failing is a well known failure mode. Print it, or keep a static copy on a separate host.
- Tiers and triggers, each with one concrete example so the labels stop being abstract.
- Roles: an incident commander (the person who runs the response and makes the call, not necessarily the most senior engineer), a communications lead, and a scribe who keeps a timestamped timeline.
- Channels per tier: status page always, email to affected accounts for SEV1 and SEV2, in-app banner when the product is partly usable.
- Three templates: investigating, identified, resolved.
- The postmortem owner and the publication deadline.
A postmortem is a written review published after the incident that explains what happened, the customer impact, and what changes. Customers read those more carefully than they read your marketing pages.
How do you write an update customers can actually read?
Say what is affected, what customers will see, what you know and have ruled out, and when the next update arrives, and avoid cause speculation until it is confirmed because a wrong cause costs more trust than silence.
A four line template is enough:
- Status: investigating, identified, monitoring, or resolved.
- Impact: which components, which customers, what fails when they try.
- Action: what the team is doing right now, in one sentence.
- Next update: a clock time, not a duration.
Give a next update time you can keep, never a fix estimate you cannot. "Next update at 15:10" is a promise you control. "Fix expected in 20 minutes" is a promise you do not.
Put the status page inside the build scope
Treat the status page as part of the product, not a marketing add-on. It needs read access to health checks (automated probes that report whether a service responds), the service map that connects those probes to public components, and an audit trail of who published what. On a custom SaaS build this is a small slice of scope, typically one to three weeks of engineering, and it is far cheaper to add during the build than to retrofit after the first outage.
Availability, the share of time a component responds correctly, deserves a reality check at the same time. 99.9% availability over a month allows roughly 43 minutes of full downtime; 99.95% allows roughly 22 minutes, before counting partial degradation. If you promise 99.95% in a contract, your communication plan has to support a first update inside 15 minutes, because the budget for silence is close to zero.
Host the page outside the system it describes. A status page running in the same region as the product goes down with it, which is the one moment it needs to work.
How do you know the plan works before a real incident?
Run a game day once a quarter, roughly 60 to 90 minutes, where one person simulates an incident and the communications lead publishes to a staging status page, a private copy used for practice, while someone else times each step.
Measure three things: time from simulated detection to first published update, whether the cadence held, and whether the postmortem shipped within five business days. A workable goal for a small team is a first update under 30 minutes on 90% of SEV1 incidents within two quarters, though the first attempt is usually slower. The gap between the written plan and the drill is where real incidents hurt.
Frequently asked questions
What should a SaaS status page show during an outage?
It should show the affected components marked with a clear state, an active incident entry with a timestamped update, and a stated time for the next update. Scheduled maintenance and a short history of past incidents belong on the same page so customers can judge reliability without asking support.
How often should you update customers during an incident?
For a serious incident, publish the first note within roughly 15 to 30 minutes of confirming customer impact, then update every 30 to 60 minutes until it is resolved. Keep updating even when nothing has changed, because a stale timestamp reads as neglect and pushes customers toward public complaints.
Should a status page be hosted separately from the product?
Yes, host it on infrastructure that does not depend on the system it describes. If the page runs in the same region as the product, a regional failure takes down both the service and the place customers go to check on it. A static copy of the plan should also live outside your internal wiki.
Do you still need a status page if you send incident emails?
Yes, because the two do different jobs. Email reaches people who are not looking, while the status page is the single place a customer can check at any time without waiting for a message, and it holds the timestamped record you can point to afterwards.