Editorial illustration; the status labels are representative examples, not a live monitoring screen.
When a familiar service stops working, the first few minutes are often wasted on the wrong question. Someone restarts a laptop, changes a DNS setting, disables a browser extension and then reports that “the internet is broken”. Those actions may change the evidence without identifying the failure.
This guide is for a person who needs to decide what to do when an email provider, collaboration tool, cloud dashboard or website suddenly becomes unreliable. The aim is not to explain every type of outage. It is to separate a local problem from a shared incident, protect unsaved work, and create a short record that another person can use.
Summary
- Reproduce the smallest failure and note the exact service, account and time before changing anything.
- Compare one independent connection or device, then check the provider’s official status channel and your own monitoring.
- Preserve unsaved work and avoid risky production changes until the failure boundary is clearer.
- Record the evidence, impact and next check so the incident does not become a collection of guesses.
Start with the user-visible failure
Write one sentence that could be tested: “Outlook web loads, but sending a new message returns an error,” or “The site is reachable from my phone but not from the office network.” Avoid starting with a diagnosis such as “DNS is down”. A diagnosis can be wrong; the observation is still useful.
Record the local time, the URL or application, whether you were logged in, and what you expected to happen. If a form or document is open, save a local copy or take another safe route before refreshing. Protecting work is part of incident response, not an optional extra.
1. Check whether the failure is repeatable
Repeat the same action once. Do not click ten times and then call the number of clicks evidence. Note whether the page fails immediately, times out, returns an error, or loads but behaves incorrectly.
Use a second page on the same service if it is safe to do so. A broken dashboard does not prove that the whole provider is unavailable. Conversely, a homepage loading does not prove that login, uploads or API requests are healthy.
If the error includes a request ID or incident code, copy it into your private incident note. Do not post access tokens, customer data or full private URLs in a public status thread.
2. Compare one independent path
The most useful quick comparison changes one boundary at a time:
- Try the same task from a second browser profile or device.
- If appropriate, compare the office or home network with mobile data.
- Ask one colleague in a different location to test the same user journey.
Do not change the browser, device, network and account together. If the second test succeeds, you have learned that the failure is conditional, not necessarily that the first device is faulty.
A phone test can mislead when the mobile app uses a different backend from the web application. Describe what you tested, including platform and route, rather than writing “works on mobile”.
3. Check the provider’s official signal
Use the provider’s status page, official support account or in-product incident notice. Search results and social posts can be useful leads, but they are not the source of truth for your account or region. Look for the affected component, start time, current impact and whether the provider is investigating or recovering.
Treat an all-clear cautiously. A service may recover gradually, or only one region, plan or API endpoint may still be affected. Keep testing the exact action that matters to you.
Current interest in Malaysia around a large Microsoft service disruption is a useful reminder that an outage can become widely visible before every user sees the same symptom. The trend signal can tell you that many people are asking; it cannot tell you whether your particular request is failing for the same reason.
4. Check your own boundary without guessing
If the problem appears local, inspect the least disruptive clues first:
- Can other sites and services load on the same connection?
- Does the browser show a certificate, DNS, authentication or timeout error?
- Does the application have a current status, account or session warning?
- Did a network, security or proxy change happen just before the failure?
Avoid changing DNS, firewall rules or endpoint security policy as a first response. Those changes can create a second incident and make a later comparison harder. If you manage the network, capture the current configuration before making a controlled test.
For browser failures, the workflow in Browser Console Errors Are Clues: A Practical Triage Workflow helps connect the visible action to the first relevant error. For a slow rather than unavailable page, How to Use Chrome DevTools to Find the One Request Making a Website Slow is the better fit.
5. Decide what not to touch
During uncertainty, restraint is a technical skill. Do not restart a production service, purge every cache, rotate credentials or deploy a “quick fix” merely because a dependent service is failing. Ask what evidence would justify the change and what recovery path exists if it makes things worse.
There are exceptions. If you have evidence of an active security incident, follow the documented containment plan. If unsaved work is at risk, take a safe copy. If a provider instructs customers to change a setting, verify the instruction through an official channel and record the version or date.
6. Record impact in a useful way
A short outage note should include:
- service and feature affected;
- first observed time and latest test time;
- affected users, region or network;
- exact action and result;
- comparison that succeeded or failed;
- provider status signal and its timestamp;
- workaround, if one is safe;
- next check and owner.
“Teams down” is less useful than “Three Kuala Lumpur users cannot send new messages in the web client; reading existing messages works; mobile data shows the same error; provider status is investigating.” The second note supports a decision.
Example: a failed cloud dashboard
Imagine a dashboard returns a blank page after login. The homepage loads, a colleague in another country sees the dashboard, and a private window shows the same blank result. The first conclusion should not be “clear all cookies”. Compare the browser console and the network request generated after login, check the provider status page, and save the request ID. If the failure follows your account rather than your device, that is information the provider needs.
If the dashboard is a deployment control panel, do not start a second deployment while its state is unclear. Confirm whether the first action completed by checking the official deployment log or target environment. Duplicate actions can turn a visibility problem into a real configuration problem.
Common mistakes
- Treating a popular status post as proof that your account has the same failure.
- Changing several network or browser settings before recording the original symptom.
- Sharing screenshots that expose tokens, email addresses or private URLs.
- Reporting only that a homepage loads, when the important action is login or publishing.
- Repeating a failed action so often that rate limits or duplicate transactions become possible.
My Take
Good outage triage is deliberately boring. The valuable result is a boundary: local device, local network, account, provider component or uncertain. Once that boundary is written down, the next action becomes smaller and safer. That is more useful than a confident explanation produced before anyone compared the evidence.
Conclusion
Describe the smallest failure, repeat it once, compare one independent path, check the provider’s official signal and record the impact. Preserve work and avoid production changes until the evidence supports them. If the service recovers, keep the note: the next outage will be easier to recognise and explain.