What Should You Monitor After Launching Software?
After launching software, monitor availability, errors, response times, failed background jobs and the user journeys your business depends on. Give every actionable alert an owner, a backup and a clear escalation route. Set expectations around the harm a failure could cause and the support you have agreed, rather than copying another product's thresholds.
What should your monitoring brief cover?
Start with the outcomes customers need. A page loading is useful evidence, but it does not tell you whether someone can sign in, save their work or finish an order.
Ask your build or maintenance partner to complete this brief:
| Area | What to watch | What the alert should explain |
|---|---|---|
| Availability | Whether important services and pages can be reached and used | Which capability is unavailable and where the check failed |
| Errors | Failed requests, crashes and unsuccessful operations | What failed, how often and which users or actions are affected |
| Response times | How long important actions take from the user's perspective | Which action slowed down and how it compares with its agreed expectation |
| Background jobs | Failed, overdue or unfinished work | What has not completed and which business outcome is waiting |
| Critical journeys | Whether a meaningful sequence reaches its expected result | Which step failed and whether the final result exists |
For each row, record the owner, support hours, alert destination and first response action. Treat this as part of the handover, alongside access and documentation.
How should you check availability?
Define availability around the capabilities customers need. Include the public entry point, signing in and the main area where work happens. Where relevant, check access from outside the application itself.
A check that only confirms a page responds should have a narrow label. Call it a page availability check, rather than evidence that the whole product works.
Ask what happens when the monitoring check stops running. Include a way to identify missing monitoring results, so silence is not automatically treated as success.
For a product used only during agreed business hours, decide separately whether an overnight failure needs immediate action or a response before the next working period. Document that decision in your software maintenance and support agreement.
Which errors deserve attention?
Prioritise errors by the action they interrupt and the harm they create. A failed save, an unsuccessful sign-in and a broken optional image should not automatically have the same urgency.
Ask for error reporting that groups repeated failures and provides enough context to investigate:
- The action that failed and when it happened
- The affected product version
- Whether the failure is new or recurring
- The scope of affected users or operations
- A reference that support can use to find the relevant diagnostic record
Keep passwords, access credentials and unnecessary personal information out of diagnostic records. Decide who can access those records and how long they should be retained.
Also define failures that do not appear as crashes. If an action returns a success message but the expected record is missing, include that outcome in a journey or completion check.
How should you monitor response times?
Choose actions that matter to users: opening their workspace, searching, saving or submitting a request. Agree where each measurement starts and ends.
Measure the experience you intend to promise. If the promise is that a customer can open and use a page promptly, measuring only the initial response leaves part of that experience outside the check.
Ask to see the spread of response times, including slower experiences, rather than relying on a single average. Separate important actions so a quick, frequently used page does not dominate the view of a slower submission flow.
Set expectations using user needs, launch testing and observed behaviour. Review them as usage changes. For preparation before release, see load testing before a software launch.
What should you watch in background jobs?
List work that continues after the user leaves a screen. Examples include sending confirmations, generating reports and synchronising records.
For each job, define:
- What confirms that the work started
- What proves that it completed successfully
- When unfinished work becomes overdue
- How failed attempts and repeated retries are recorded
- What must be checked before replaying the work
Watch for expected work that never starts as well as work that reports an error. Track how long the oldest unfinished item has waited, alongside the amount of pending work.
Do not treat every retry as an incident. Agree when automatic recovery is acceptable and when a person must investigate. Before allowing a manual retry, confirm how duplicate emails, records or other repeated effects will be prevented.
How do you monitor a critical user journey?
Choose a small set of journeys whose failure would interrupt the service you sell. Define success as an observable result.
For an illustrative booking product, the journey might be:
- Sign in with a dedicated test account.
- Find an available test appointment.
- Submit a booking.
- Confirm that the booking appears in the account.
- Confirm that the expected notification reaches its test destination.
Checking the final record matters as much as seeing the confirmation screen.
Agree how test activity stays separate from real customer activity, how it is cleaned up and which outside services it may contact. Avoid checks that create real charges or send messages to customers.
Combine these controlled checks with error and performance information from actual use. Record what each method covers and what remains outside its view.
Who should own alerts and escalation?
Every urgent alert needs a named responsible role and an agreed route to a person. A shared inbox is a destination; it does not establish responsibility.
Your monitoring brief should answer:
| Decision | What to agree |
|---|---|
| Primary owner | Who acknowledges the alert and begins investigation |
| Backup | Who takes over if the primary owner is unavailable |
| Coverage | When someone is expected to respond |
| Escalation | What triggers wider technical or business involvement |
| Communication | Who updates affected customers and the business owner |
| Closure | Who verifies recovery and records follow-up work |
Distinguish acknowledgement, mitigation and resolution. Receiving an alert, restoring a usable service and fixing the underlying cause are separate milestones.
Set escalation rules around impact and uncertainty. Suspected data loss, for example, deserves an explicit route even before the full scope is known. Rehearse the contact path before relying on it.
How is operational monitoring different from product analytics?
Operational monitoring asks whether the software is working as intended. Product analytics asks how people use it and where the product could improve.
| Question | Main purpose |
|---|---|
| Can customers submit an application successfully? | Operational monitoring |
| Where do customers leave the application process? | Product analytics |
| Did the report finish and become available? | Operational monitoring |
| Which reports do customers return to? | Product analytics |
Use both when investigating a change in behaviour. If fewer people complete a journey, check for failures and delays before deciding the design needs changing.
Keep alerting responsibilities separate from product review. An urgent failure needs an operational response; an adoption question needs investigation and a product decision.
What common monitoring mistakes should founders avoid?
Treating a green dashboard as proof of a working product. Ask which customer outcomes it actually checks.
Alerting on everything. Define the action expected from each notification. Use a review queue for signals that do not require immediate intervention.
Copying universal thresholds. Agree tolerances for your product, its users and its support coverage.
Leaving ownership with an individual informally. Record the backup and update responsibilities when the team or supplier changes.
Closing an incident after a restart. Verify the affected journey, check unfinished work and assign any remaining repair.
Ignoring monitoring after launch week. Review checks when journeys, dependencies or support arrangements change. Remove obsolete alerts and add coverage for new business risks.
What questions should founders ask before signing off?
Do we need round-the-clock monitoring?
Separate continuous checking from continuous human response. Agree which incidents require immediate attention and which can wait for staffed hours. Make that coverage explicit in the support agreement.
Should every alert come to the founder?
Route technical alerts to the responsible support team. Agree when the founder needs an update or must make a business decision, such as approving a temporary service restriction.
Can monitoring replace testing?
Use testing to assess a change before release and monitoring to watch the running service afterwards. Include both in the maintenance brief, with clear responsibility for investigating failures.
What should we receive at handover?
Request the monitoring brief, access arrangements, alert owners, escalation contacts and evidence that important alerts reach their recipients. Include a demonstration of a failed check and recovery, plus instructions for keeping the checks current.
*Need a clear plan for keeping your software dependable after launch? Start a conversation →*