The phrases “24/7 support” and “rapid response” in a maintenance contract are sentences that, on the day the site goes down, turn out to commit to nothing. A measurable SLA (Service Level Agreement) states which event falls into which priority, when the clock starts and stops, and how the result is reported. The framework below covers how to set up the SLA in a maintenance and support contract for corporate websites, from the perspective of both the IT team receiving the service and the party providing it; it doesn’t go into pricing. Define the scope of server maintenance and technical support separately for updates and capacity tracking of the server hosting the application.
What the scope of a maintenance contract should include
The SLA is built on the scope. If the scope is vague, every request starts with a “is this maintenance or development?” debate and time measurement loses its meaning. While the clock runs, the parties are interpreting clauses. The scope section should clearly list at least these headings:
| Item | In-scope examples | Usually out of scope |
|---|---|---|
| Updates | Core, plugin and theme updates, trial in the test environment | Redevelopment resulting from a major version migration |
| Security | Patching, vulnerability tracking, suspicious file review | Comprehensive penetration testing |
| Backup | Regular backups, periodic restore testing | Setting up off-server archive infrastructure |
| Monitoring | Tracking uptime, SSL expiry and domain expiry | Server hardware monitoring (if hosting is separate) |
| Bug fixing | Fixing a broken existing function | Adding new functions |
| Content support | A limited number of content and image changes | New page design, content production |
| Reporting | Monthly SLA and maintenance report | Marketing and analytics reports |
Out-of-scope work isn’t forbidden; it’s handled as a separate request type with its own timeframe and approval process. This separation protects the measurement: a new feature request doesn’t fall within the resolution time defined for a critical bug.
The second issue is the boundary of responsibility. If the site runs on hosting the service provider doesn’t manage, the provider can’t commit to availability for server outages; it can only commit to detection and escalation to the hosting company. You can’t guarantee infrastructure you don’t manage. By the same logic, the contract should have a “dependent system” clause for failures in third-party systems such as the ERP, the payment provider or the CRM. Who has access to the domain and DNS panel is also written in this clause; some of the longest outages in the field come from not being able to reach the person who can change the DNS record.
Priority classes and an example SLA matrix
Priority is determined by criteria written in advance, not by the urgency felt by whoever opens the request. Two axes are used: impact (how many users, which business process) and urgency (is there a workaround, does the damage grow over time). The matrix below is a starting point; times are negotiated according to the organization’s needs and the scope of the service.
| Priority | Definition (example) | Response time (example) | Resolution target (example) |
|---|---|---|---|
| P1 – Critical | Site unreachable, payment or form completely broken, active security breach | 1 hour, 24/7 | Temporary or permanent fix within 4 hours |
| P2 – High | An important function is broken, a workaround exists or the impact is limited | 4 business hours | 2 business days |
| P3 – Normal | Visual error, problem on a single page, an error that doesn’t block a function | 1 business day | 5 business days or the next maintenance window |
| P4 – Request | Information request, small content change, improvement suggestion | 2 business days | As scheduled |
The definition column is what makes the matrix work. Is a “the site is slow” complaint P1 or P3? If the definitions are written with concrete examples, this debate happens at the contract table, not when the request comes in. Who can raise or lower the priority, and on what grounds, should also be written down.
Also clarify the distinction between “24/7” and “business hours”: which time zone are business hours in, which days do they cover, how are public holidays counted? If a 24/7 response is committed for P1, an out-of-hours notification channel (phone, on-call line) is defined separately. A P1 arriving by email at night may not be read until morning. Without a channel, there’s no commitment.
The difference between response time and resolution time
The ambiguity we encounter most often in contracts is what “intervention time” measures. In a measurable SLA, three separate timestamps are defined:
- Response (first intervention) time: The time from the request being logged until a competent person examines it and gives the first meaningful reply. An automatic “your request has been received” email doesn’t count as a response.
- Workaround time: The time until the business process is working again; for example, rolling back a broken plugin to its previous version.
- Permanent resolution time: Removing the root cause. In P1 incidents, the permanent resolution can be tracked separately as P2 or P3 after the workaround.
Clock-stop conditions should also be in writing. Typical reasons: waiting for information or access from the customer, a change the customer must approve, a response from a third-party provider. If these reasons aren’t kept as a separate status in the ticketing system, at month-end it turns into a debate about which time was caused by whom. Without a record, there’s no measurement.
Planned maintenance, urgent intervention and development requests
Maintenance work moves in three separate streams, each with different rules.
Planned maintenance: Updates, backup tests, performance checks. Done in a maintenance window announced in advance. Planned downtime in this window is excluded from the availability calculation; for this, the window’s duration, frequency and notice period must be defined in the contract. Which test workflow updates follow can be defined separately in a WordPress update policy document and attached to the contract as an annex.
Urgent intervention: P1 and P2 incidents, critical security patches. They don’t wait for the maintenance window. But urgent changes are also recorded, and a short root cause note is written after the incident; a note that isn’t written means the same failure repeats three months later.
Development requests: A new page, a new function, an integration change. These are subject to scope and schedule agreement, not SLA times. Many contracts cover small developments within maintenance by defining a monthly hours pool; use of the pool is shown in the monthly report. Part of the pool can be set aside for reducing accumulated technical debt.
SLA measurement and monthly reporting
A commitment that can’t be measured can’t be reported. Measurement sources are named in the contract: a ticketing system (helpdesk) for requests, an external monitoring tool for availability. Both parties should be able to access these sources; a panel only the provider sees isn’t a measurement from the customer’s point of view.
Availability measurement
If an availability commitment is given, the measurement method should answer at least these questions: Which URLs are monitored (only the home page, or the form and payment steps too)? What is the check frequency? How many consecutive failed checks count as an outage? From how many different locations is the check made? How is planned maintenance excluded? If the home page returns 200 while the payment step returns 500, a tool monitoring only the home page says “the site is up”.
Writing the minute equivalent of percentage targets in the contract makes expectations clear. An example calculation: a 30-day month is 43,200 minutes. 99.9% availability allows about 43 minutes of downtime in that month, and 99.5% about 216 minutes (3 hours 36 minutes). A monitoring tool with a 5-minute check interval may not see outages of a few minutes at all, or measures them in 5-minute steps. The target and the measurement precision must match.
Request measurement and report content
The monthly report should show the outcome of every commitment in the contract. Recommended content:
- Number of requests opened and closed by priority
- For each priority, requests that met and missed the response and resolution time targets, and the reasons for misses
- Requests with the clock stopped and the reasons for stopping
- Measured availability, outage events and root cause notes
- Updates made (component, old and new version), the result of the backup restore test
- Use of the development hours pool
- Recommended improvements and open risks for the next month
The consequence of missed targets (service credit, additional working hours or an improvement plan) is defined in the contract. Writing a penalty for an item that can’t be measured produces an empty debate for both parties; penalties should only be tied to items whose measurement source has been agreed. How to monitor silent integration errors, for the technical infrastructure of the report, is explained in the integration monitoring and reconciliation guide.
For teams who want to place service management processes in a corporate framework, ISO/IEC 20000-1 defines the requirements for service level management. A small maintenance contract doesn’t need to apply the whole standard; it’s a good reference for putting the terms into a common language.
Pre-contract checklist
When evaluating a proposal or reviewing a draft, there should be written answers to the following questions:
- Are in-scope and out-of-scope work listed with examples?
- Do the priority definitions include concrete examples; who determines the priority?
- Are response, workaround and permanent resolution times defined separately?
- Are business hours, time zone and holidays specified; is there an out-of-hours P1 channel?
- Are the clock-stop conditions in writing?
- Are the planned maintenance window, its frequency and notice period defined?
- If there’s an availability target, are the measurement tool, check frequency and monitored URLs specified?
- Are the responsibility boundaries for hosting, domain, DNS and third-party systems clear?
- Is there a commitment on backup frequency, retention period and restore testing?
- Are the content and delivery date of the monthly report specified?
- Are ownership of admin accounts, licenses and access credentials, and their handover at the end of the contract, defined?
The last item is the one most often forgotten. If license accounts and admin access are registered in the provider’s name, the site owner may struggle to access their own site when the contract ends. Access without a clear owner is a risk. For organizations keeping request records on Microsoft 365, the helpdesk setup with Microsoft 365 guide can serve as an example measurement infrastructure. For a maintenance service with a defined scope and SLA, see the website maintenance and support contracts page; to set up availability measurement, see the web and uptime monitoring page.
Frequently asked questions
What’s the difference between response time and resolution time?
Response time is the time until the request is examined by a competent person and the first meaningful reply is given. Resolution time is the time until the business process is working again or the root cause is removed. The contract should define the two separately.
If hosting is with another company, can the maintenance provider commit to availability?
Not for server and network outages; it doesn’t manage that infrastructure. In that case, the provider can commit to times for detecting the outage, notifying the hosting company and resolving problems on the application side.
When is the clock stopped in SLA measurement?
Typically when waiting for information, access or approval from the customer, and when waiting for a response from a third-party provider. Clock-stop conditions should be written in the contract and recorded as a separate status in the ticketing system.
What should a monthly SLA report include?
Request counts by priority, requests that met and missed the target with the reasons for misses, requests with the clock stopped, measured availability and outage events, updates made, the backup test result and use of the development hours pool.