Insights/Guides

How to communicate with customers during a software outage

What to say in the first 15 minutes, how often to update, which channel does which job, templates for each stage, and what your SLA really promises.

Thinkiyo team··13 min read

Say something within 15 minutes, before you know the cause. Name what customers are seeing, say you are working on it, and give the time of the next update. Then keep that promise on a fixed rhythm, with the same words on every channel, until it is fixed. Afterwards, tell everyone it is over, say what they need to redo, explain how credits work, and publish a short account of what happened.

This is the communication playbook. For handling the flood of tickets itself, see how to handle a spike of support tickets during a SaaS outage.

What the published playbooks agree on

Three public sources cover most of it, and they point the same way.

Speak before you understand. Google's SRE workbook says that unless you acknowledge an incident is happening and being addressed, people will assume nothing is being done. Atlassian's Statuspage guidance is to acknowledge quickly, summarise the known impact and promise further updates.

Fast first message, steady rhythm. PagerDuty's own incident guide has the first message posted within 5 minutes of the incident call starting, a second with the scope within 5 minutes of that, then updates at least every 20 minutes for the first two hours. Atlassian suggests every 30 minutes, or whatever suits the situation. Those are two vendors' practices. Neither is an industry benchmark.

One voice, separate from the fixers. Google's model has a communication role that is the public face of the response, while only the operations team changes the system. PagerDuty calls the person a customer liaison: they write the message, get the incident commander's approval, then send it.

Templates written in advance. Google recommends two or three ready-to-use templates.

No guesses. PagerDuty's rules for the liaison are "Never lie, and never guess" and "Don't estimate resolution times".

What contracts and regulators require

The principle that holds almost everywhere: for most software companies, what you owe customers during an outage is set by your contract, and the contract usually promises an uptime figure and a credit. It rarely promises a message. Regulated sectors are the exception. This section is general information and not legal advice. Check your own contracts and your own market.

What published SLAs look like. Two examples, as their pages read in November 2026.

Google WorkspaceAmazon EC2 (region level)
Promise99.9% monthly uptime99.99% monthly uptime
Credit3, 7 or 15 days of service, by how far uptime falls10%, 30% or 100% service credit, by how far uptime falls
How to get itSupport case within 30 days of becoming eligible, or the credit is forfeitedSupport case received by the end of the second billing cycle after the incident
LimitCapped at 15 days a month. Sole and exclusive remedy under the SLASole and exclusive remedy, unless the agreement says otherwise

The pattern to notice: uptime is measured over a month, the customer has to claim, there is a deadline, and the credit is the only remedy the SLA gives. Read yours before the next incident.

Where a rule adds a duty.

  • EU financial entities. Under DORA, which has applied since 17 January 2025, a major ICT incident that affects clients' financial interests must be communicated to clients "without undue delay", along with the measures taken to limit the harm. That is Article 19(3), which we read in this reproduction of the regulation. Confirm against the official text.
  • UK financial firms. The FCA's operational resilience rules came into force on 31 March 2022. Firms must set impact tolerances for the maximum tolerable disruption to important business services.
  • Australian banks, insurers and super funds. APRA's standard CPS 230 has regulated entities notify APRA of operational risk incidents and of breaches of tolerance for a critical operation.

If you sell to firms like these, expect your contract with them to pass some of that clock down to you. If personal data is exposed, a separate set of deadlines starts. The market table for that is in the ticket spike post.

The first 15 minutes

Every time in this guide is a suggested starting point, not a benchmark. Adjust to your product and team size.

  1. Minute 0 to 5. Someone declares an incident. Google's test is simple: if the outage is visible to customers, it is an incident. One person takes the communication role and stops answering tickets.
  2. Minute 5 to 10. That person collects three facts from support and engineering: what customers see, who is affected, when it started. Unknowns stay unknown.
  3. Minute 10 to 15. The first message goes on the status page, approved by whoever leads the incident. It contains the symptom, the words "we're investigating", and the time of the next update.

At a five-person company the incident lead and the communicator may be the same founder. Write the message, then go back to the fix.

The first hour

  • Post a second message once you know the scope: which features, which regions or plans, and what still works.
  • Share the status text with every support agent as the only approved wording.
  • Have account owners write personally to customers with contractual SLAs. PagerDuty's guide also has premium support customers told individually.
  • Tag every related ticket, so the list exists when you need to close the loop.
  • Pause scheduled marketing emails and social posts.

Update rhythm

StageSuggested timingWhat the message carries
AcknowledgeWithin 15 minutes of declaringSymptom, "investigating", next update time
ScopeWithin 30 minutesWho and what is affected, what still works, any workaround
UpdatesEvery 30 minutes for the first two hoursWhat changed, what was ruled out, next update time
Long incidentHourly after two hours, said out loudThat the fix will take time, and the new rhythm
ResolvedWhen the incident lead confirms recoveryTime fixed, what to redo, whether data was lost
Follow-upWithin 5 working daysWhat happened, what changes, how to claim credits

Post whenever the impact changes, whatever the clock says. And avoid empty updates. PagerDuty warns about notification fatigue from content-less updates. An update with no fix still needs one new fact, even if the fact is what you ruled out.

Which channel does what

Atlassian's advice is consistent updates across all channels. The wording is shared. The jobs differ.

ChannelIts jobOwnerNever
Status pageThe single source of truth. Every other message links hereCommunication leadLeave it green while tickets pile up
In-app bannerReach people who are in the product and confusedProduct or engineeringDepend on it if the app itself is down
Social repliesOne line and the status link, where people are askingSupport or social teamExplain causes or argue in threads
EmailStart and end of serious incidents, and the follow-up noteCommunication leadSend every 30-minute update to the whole list
Ticket repliesThe approved wording, plus anything specific to that customerSupport agentsImprovise a cause or a fix time
Account callsCustomers with SLAs or large exposureAccount ownerPromise a credit before reading the contract

Host the status page somewhere that does not share infrastructure with your product.

When you don't know the cause

Say what you do know, in customer terms: the symptom, the scope, what you have ruled out and what you are trying next. As PagerDuty's guide puts it, customers don't care that application-server-123 is having issues, they care that they are not getting notifications.

The Microsoft 365 outage that began on 12 November 2026 shows the pattern at scale. Microsoft's first public post, at about 12:50pm US Eastern time, said only that it was "investigating reports of users experiencing issues accessing Exchange Online". Overnight it said its attempts to improve service were "not taking as expected". It first described the problem as limited to the Americas, then confirmed any region could be affected. On Friday it apologised and promised a full report once its assessment was done. Write "at least" in front of any scope you have not confirmed. Our lessons from the Microsoft 365 outage cover it in detail.

Three more rules for this stage:

  • Don't name a third party until you are sure, and even then own it. Atlassian notes that an incident caused by another provider is still, in your customers' eyes, a problem with your service.
  • Don't give a fix time. Give the next update time.
  • If customers ask whether their data is safe and you don't know yet, say you are checking and will confirm. Say "no data was affected" only when engineering has confirmed it.

Templates

Fill in the brackets. Times in UTC, or the time zone most of your customers use.

Acknowledge

We're investigating a problem with [feature]. Some customers are seeing [symptom in plain words]. We'll post the next update by [time].

Scope

[Feature] is failing for customers [on plan, in region, or "on all plans"]. [Other features] are working normally. [Workaround, or "There is no workaround yet."] Next update by [time].

Update with no fix

Still working on [feature]. We've ruled out [cause] and are now testing [next step]. No fix yet. Next update by [time].

Long incident

This is taking longer than we want. The fix needs [plain description], and we don't have a reliable finish time. We're moving to hourly updates. Next one by [time].

Resolved

[Feature] has been working again since [time]. Anything you [did] between [start] and [end] may need to be redone: [specific instruction]. No data was lost. [Or: what was lost.] We'll publish what happened by [date].

Follow-up email

On [date], [feature] was unavailable for [duration]. The cause was [one plain sentence]. We have [change already made] and will [change still to come] by [date]. If your plan includes an uptime commitment, here is how to claim a credit: [steps and deadline]. I'm sorry for the disruption. Reply to this email and it comes to me. [Name, role]

Hard case: a customer wants a fix time

I can't give you a time I'd trust, and I'd rather not guess and be wrong. What I can promise is an update by [time], on [status link] and here in this thread.

Service credits

Decide three things before the incident ends.

  1. Who qualifies. Check the uptime figure against the contract's definition. A partial failure may not count as downtime under your own wording.
  2. Whether customers must claim. If your SLA needs a claim by a deadline, as the two examples above do, say so plainly in the follow-up.
  3. Whether you go beyond the contract. A goodwill credit is a business decision. If you make one, apply one rule to everyone affected. Case-by-case credits reward whoever complains loudest.

The post-incident note

Internally, Google's SRE book describes a postmortem as a written record of the incident, its impact, the actions taken, the root causes and the follow-up actions.

The customer version is shorter. Atlassian's Statuspage help reduces it to three parts: apologise, show you understand what happened, explain your remediation plan. In practice:

  • What customers experienced, with start and end times.
  • The cause, in one paragraph a non-engineer can follow.
  • What you have already changed, and what you will change, with dates.
  • What customers should do, including credits.
  • A named person to reply to.

Leave out staff names, internal system names and anything you have not verified. Link the note from the status page incident and send it to everyone you tagged.

A Monday checklist

  1. Write the acknowledge, scope, update, resolved and follow-up templates into your status tool and your helpdesk.
  2. Name a communication lead and a backup for each time zone you cover.
  3. Write down who can declare an incident. Include support.
  4. Check that your status page runs on separate infrastructure, and that two people can post to it from a phone.
  5. Read your own SLA. Note the uptime figure, how downtime is defined, how to claim, and the deadline.
  6. List the customers with contractual SLAs and who calls each one.
  7. Run one 20-minute drill: a made-up outage, a real first message, a stopwatch.

Mistakes teams make

  • Waiting for the cause before the first message.
  • Stating a scope as fact, then widening it twice.
  • Marking the incident resolved when the fix ships, before anyone has confirmed customers can work again.
  • Sending the resolved message with no instruction about work that needs redoing.
  • Leaving the credit claim process out of the follow-up.

What to measure

  • Minutes from first customer report to first public message.
  • The longest gap between updates, against the rhythm you promised.
  • Next-update promises kept, as a share of those made.
  • Share of affected customers who received the resolved message and the follow-up.
  • Contacts per incident over time. Fewer "is it just me?" tickets suggests the status page is doing its job.

What runs in Relay today

Relay pulls customer messages from connected channels (email, chat, WhatsApp, social posts, reviews and app store reviews) into one inbox, and drafts replies that a person approves and sends.

The limits, stated plainly. Each channel works only after you connect your own account or key. Drafts come from templates today. Nothing is sent automatically, so plan for approval time on a large incident. The companion listener's alerts on new negative mentions are basic today, crisis mode is switched on by hand, and an automatic volume-spike trigger is on the roadmap. Relay is not a status page tool. You will still need one.

If outage replies currently live in five tabs, talk to us. See also how Relay works.

Common questions

How often should we post updates?

Atlassian suggests every 30 minutes or whatever suits the situation, and PagerDuty uses at least every 20 minutes for the first two hours. Pick a rhythm you can keep, state the next update time in every message, and slow down openly if the incident runs long.

Should the update go on the status page or on social media first?

The status page first, then everything else links to it. Social replies should be one line and a link, because a thread is a poor place to keep a record. Email is for the start and end of serious incidents and for the follow-up note.

What do we say when we don't know the cause?

Say what customers are seeing, who is affected, what you have ruled out and when you will next update. PagerDuty's rule for its customer liaison is "Never lie, and never guess". A wrong guess gets quoted back for weeks.

Do we have to give service credits?

Only what your contract and local law require, which depends on your SLA wording and on whether the customer is a consumer. Many published SLAs make the customer claim within a deadline, as Google Workspace and Amazon EC2 do. Take legal advice on larger claims.

Does Relay post to our status page?

No. Relay is the inbox for the messages customers send you, with drafted replies a person approves. Your team posts status updates in its own status tool.

Sources

See what Relay can do for your customer service

Get a personalised 30 minute demo on your own channels and policies. We show you Relay handling your real conversations, then set it up with you.

  • Built and connected for you
  • Can run in your own cloud
  • A person on every refund

Brands we work with

  • Bolt
  • Dave & Buster's
  • ETV Win
  • Malachite Cars
  • Space Angel
  • Infinity Quest
Made by Thinkiyo© 2026 Thinkiyo. Relay is a Thinkiyo product.Examples use Lumen Parcel, a fictional brand.