Australian cloud for business software — built and run in Sydney since 2002

NetLeverage
NetLeverage

Severity levels, response and resolution times: how to read a support schedule

Technical article
Hosting

About this article: This article is reviewed against the cited sources. This is source research, not lab validation. Check the stated versions and your own requirements before making changes.

A response time in a support schedule is, in the public schedules we read, the time until a support engineer first contacts you and starts work. It is not the time until the problem is fixed. Microsoft's Azure page defines its initial response time as the period from when you submit a request to when an engineer "contacts you and starts working on your support request". Nothing in that definition promises a resolution.

Severity decides which response time applies. In the three schedules we compared, the customer proposes the severity when the request is raised, each severity has a written definition of business impact, and at least one vendor (Azure) says it can lower the level. The practical skill is to read three things together: the severity definition, the clock that response starts and stops, and the hours in which the clock runs. This guide is for a buyer comparing support terms. The large vendors below are examples of how definitions are structured, not hosting-provider contracts.

Why it matters

In our view, many disputes about support come down to wording that was never read. A buyer may assume one hour means the server is back in an hour. Our illustration: the provider's engineer replies in forty minutes with a request for logs, and on the Azure-style definition above the response time has been met.

The same gap appears with severity. If your definition of an emergency is that staff cannot invoice, and the schedule's top level reads as a total outage of the provider's own platform, your invoicing outage may be a lower level than you expected (our reading of how definitions can diverge). Time spent arguing the level during an incident is time not spent fixing it.

What a response time measures

Three vendors state the clock differently, and the differences matter.

Microsoft Azure defines initial response time as above: from submission to an engineer making contact and starting work. Its table for the Standard plan shows under one hour for Severity A, under four hours for Severity B and under eight hours for Severity C, and shorter times on higher plans (Azure support scope and responsiveness).

Amazon Web Services lists first-response times by severity and then qualifies them: AWS Support "makes every reasonable effort to respond to your initial request" within the indicated timeframe. The word effort is worth noticing. Our reading is that it describes a target the vendor works towards, not a guarantee with a stated remedy. The page lists, for example, one hour for the severity called production system down.

Google Cloud says the case is assigned to a specialist and that you will see a response "within the target response time" in its priorities table. Its page defers the actual times to a separate guidelines document, which we did not read, so we quote no Google times here.

Our recommendation: write the clock down in your own words before you sign. For example: starts when the request is logged in the portal or the phone call is answered; stops when a named engineer replies with a substantive action, not an automatic acknowledgement. If a schedule does not say, ask.

Four clocks that get called the same thing

Support schedules and incident reports use several time measures, and the abbreviations blur them. Atlassian warns that MTTR is not a single metric: "it potentially represents four different measurements", because the R can stand for "repair, recovery, respond, or resolve".

For a buyer, it helps to separate:

  • Acknowledgement: the system or a person confirms receipt. Often automatic.
  • Response: an engineer makes contact and begins work. This is what the schedules above commit to.
  • Restoration or workaround: service works again, possibly by a temporary measure.
  • Resolution: the cause is fixed and the case closed.

In the pages we read, we did not find a resolution commitment for any severity. The Azure page does mention resolution, but as something the customer commits to: at its top severity, you "commit to continuous, 24x7 operation, every day with the Microsoft team until resolution". A provider can still offer restoration targets in a contract, so ask. Do not assume they exist.

How severity is decided

The three vendors use similar ideas with different mechanics.

You choose, within limits. AWS advises that although you might want the highest severity your plan allows, the best practice is to choose it "only for cases that can't be worked around" or that directly affect production applications. On Basic Support, AWS says you cannot change the severity after creating the case; on the higher plans listed you can.

You choose and make a commitment. Azure's Severity A means you confirm critical business impact, and Microsoft "may at its discretion decrease the Severity to level B" if you do not stay engaged around the clock. A footnote adds that Microsoft may downgrade the level if the customer cannot provide adequate resources or responses. Severity is therefore partly a statement about your own availability.

Definitions with conditions. Google's highest priority, P1, requires an application or infrastructure that is unusable in production, critical business impact, and that "No workaround is available that can be quickly implemented (less than 30 minutes)". It also requires that the affected component is generally available. A problem in a preview feature may therefore not qualify, on that definition. Google separates priority from escalation: its page says "An escalation is meant to flag issues that have resulted in increased business impact, or a breakdown of the resolution process", and elsewhere that for high-impact issues "escalation doesn't generally make the case proceed faster". Our reading is that the priority you set is the main lever and escalation is for when the process is not working.

Severity is not priority, and your severity is not theirs

Two further distinctions prevent confusion.

First, Atlassian separates the two words: "Severity is a measurement of impact" and priority is "a measurement of urgency". A typo on a home page is low severity but may be high priority for the business. Some schedules use one word for both.

Second, a provider has its own internal severity scale for incidents it declares, which is separate from the severity on your support case. PagerDuty's public guide, for example, defines internal levels where anything above its third level counts as a major incident, and advises that if you are unsure which level applies, "treat it as the higher one". That is internal practice at one company, not a rule for anyone else. The point for a buyer: ask whether your case severity and the provider's declared incident level are linked, and who tells you when an incident affecting you has been declared.

Hours, languages and the clock

Response times often run only in certain hours. On Azure, a Severity B request raised for business-hours support is handled in business hours only, and Severity C is business hours only; its footnote defines business hours for most countries as 9:00 AM to 5:00 PM on weekdays, excluding holidays (Azure page). The footnote also says the country setting you give affects those hours. AWS notes round-the-clock access for technical support on its higher plans.

If your business runs at weekends or across Australian time zones, check which time zone and which public holidays the schedule uses.

What to agree in writing

Our recommendation, not vendor guidance:

  1. Severity definitions in your terms, with at least one example from your own business for each level.
  2. Who may set and change severity, and whether the provider can lower it unilaterally.
  3. What starts and stops the response clock, and in which hours it runs.
  4. Whether any restoration or resolution target exists, and what happens if it is missed.
  5. How you reach the provider out of hours, and whether that route differs from the normal ticket system.
  6. What the provider owes you after a serious incident, such as a written report.

Questions to put to your provider

  • Is the response time measured to first contact or to first useful action?
  • Does the clock run 24 hours a day for the top severity, or in business hours?
  • Do you offer restoration targets, or only response targets?
  • If we choose a severity you think is too high, what happens?
  • When the provider declares an incident that affects us, how do we hear, and how fast?
  • Are status-page updates and support-case updates linked or separate?

A worked example (illustrative)

The following is invented. A firm's schedule has three levels. Level 1: the firm's main application is unavailable to all users, with response in 1 hour, all hours. Level 2: important functions are degraded but a workaround exists, response in 4 hours, business hours. Level 3: general questions and minor problems affecting a single user, response in 1 business day.

The firm tests the wording against three past events. A file server was unreachable for every employee for two hours: Level 1 by definition. The accounting package ran slowly for the finance team, who could still work through a manual process: Level 2. A single user could not print: Level 3.

The test also finds a gap: a serious partial outage that none of the three levels clearly covers. When the remote-desktop server failed for half the staff, with no workaround, the definitions did not say whether half of users counts as all users, so the event fell between Level 1 and Level 2. The firm asks the provider to add the words more than a quarter of users to Level 1 in the next contract revision, and records that the provider replied in writing.

Risks and limits

  • Vendor pages change. The Google page was last updated on 12 November 2026; the Azure and AWS pages show no date.
  • Large cloud vendors' schedules are not what a smaller hosting provider offers. They are used here to show structure.
  • We found no pages stating a resolution commitment. Absence from these pages does not mean no provider offers one.
  • Atlassian and PagerDuty describe their own internal practices; they are examples, not standards.
  • A schedule's wording is only as good as the contract that incorporates it. This is general information, not legal advice.

Next step

Take the last three incidents or support tickets your business raised and classify each one using the severity definitions in your provider's written terms; note where the definition was unclear. If you are reviewing a hosted service, the Managed Services page is a starting point for that conversation; ask for the scope in writing.

Sources

All accessed 15 November 2026.

Learning notes

  • A response time is how long until someone starts work, not until the problem is fixed.
  • Severity describes business impact; priority describes urgency. They are not always the same.
  • Write down which clock a time measures: acknowledgement, response, restoration or resolution.
  • Test the wording against real past incidents before you sign, not during the next one.