Maintenance contracts are usually bought at the worst possible moment: after a project has gone live, while the team is tired and the budget is nearly spent. The scope is copied from another vendor, the service levels are left vague, and the first serious incident reveals what was never covered.
This buying guide is for the manager who has to sign that contract. It sets out what to define before approaching vendors, how to compare proposals on the same evidence, and which terms are worth refusing even when the price looks convenient.
Describe the application before you describe the service
A maintenance vendor cannot price risk it cannot see. Prepare a short technical brief covering the application's purpose, age, stack, hosting model, integrations, user count and change history. Include the parts that are fragile: a custom reporting job, an undocumented API, a payment callback that only one former developer understood.
Then rank each application by business criticality. A payroll system that must run on the 25th of every month needs a different response model from an internal brochure site. Buying one contract for both usually means the critical system subsidises the unimportant one.
The five service lines to scope separately
- Corrective maintenance: diagnosing and fixing defects in existing functionality. Define what counts as a defect versus a change request, because that boundary causes most disputes.
- Preventive maintenance: patching frameworks, dependencies and server packages, checking backups, rotating certificates and reviewing logs on a schedule.
- Adaptive maintenance: changes required by a new browser, payment provider, tax rule or third-party API. These are predictable in principle and often excluded in practice.
- Perfective maintenance: small improvements that keep the system useful — report changes, usability fixes, performance tuning. Cap this in hours or story points rather than promising "unlimited enhancements".
- Security maintenance: vulnerability triage, emergency patching, access reviews and support during an incident. This is not the same as a penetration test and should not be priced as one.
For each line, state the unit of work, the monthly allowance, the rollover rule, the rate for work beyond the allowance, and who approves it. A contract that lists services without units cannot be compared with another that does.
Service levels that mean something
Severity labels are meaningless without definitions and clocks. Write four levels and test each against a real past incident.
- Severity 1: the business cannot operate — checkout down, payroll blocked, data loss in progress. Response and workaround targets apply around the clock.
- Severity 2: a major function is impaired with no practical workaround. Business-hours response with an after-hours escalation path.
- Severity 3: a limited function is affected and a workaround exists.
- Severity 4: a question, cosmetic issue or enhancement request.
Separate response time from resolution time. A vendor can answer in minutes and still take days to fix, and no credible contract promises a fix for an unknown defect. Specify the workaround target, the update cadence, the escalation contacts and the evidence you receive when the ticket closes.
A vendor scorecard
Score each proposal from 1 to 5 against criteria you set before reading the responses. Weight the criteria to reflect your exposure rather than the vendor's marketing.
- Evidence of similar work: named systems, stack and scale, with references you can call. Request a redacted sample of a monthly report.
- Team continuity: who is named, what happens during leave, and how much knowledge sits with one person.
- Security practice: patch process, secret management, access model and incident history. Ask how they would handle a vulnerable dependency discovered on a Friday.
- Documentation: runbooks, architecture notes and a change log you keep at exit.
- Response model: channels, hours, escalation and on-call arrangements.
- Reporting: monthly evidence of work done, not a count of tickets opened.
- Exit assistance: a defined handover period and a price for it.
A low price with no named team, no reporting and no exit clause is not a bargain. It is a transfer of risk back to you at the moment you can least afford it.
Pricing questions to ask in writing
- Is the fee based on hours, systems, users, incidents or a fixed monthly retainer? What happens when usage exceeds the allowance?
- Which rate applies to after-hours work, emergency changes and work requested by a third-party vendor?
- Are third-party licences, hosting, monitoring tools and backup storage included or passed through at cost?
- How is annual indexation calculated, and when can it take effect?
- Is there a minimum term, and what is the notice period?
- What is excluded from the monthly fee, and what would a typical exclusion cost if it happened next month?
Ask for a worked example: three past tickets, how they would be classified, what would be covered and what would be chargeable. Vendors who understand their own contract can answer this in a day.
Red flags in the draft
- "Unlimited support" with no definition of a ticket, a user or a working hour.
- No exclusion list. Every contract has boundaries; a missing list means the boundary will be argued later.
- Response times measured only during business hours when your business runs longer.
- Source code or deployment access held exclusively by the vendor.
- No obligation to document changes or hand over knowledge.
- A termination clause that requires paying the remaining term before you receive your data.
Run a 60-day pilot
Start with a paid pilot covering one application and a defined ticket volume. Give the vendor a mix of routine patching, one small change and one realistic incident drill. Judge them on communication, evidence and whether the system was left in a better documented state. A pilot is also your chance to test the escalation path before an actual outage.
Negotiate the exit at the start
The exit clause should guarantee source code, build and deployment scripts, configuration, database access, documentation and a transition period at a defined rate. Add a knowledge-transfer session with the named team and a final report listing open issues. If the vendor hosts the application, specify how data is exported and how long it remains available after termination.
Maintenance is part of Custom Software & Integrations, not a separate afterthought. The pricing page explains how support hours and project work are separated. For the wider budget picture, read the SME IT budgeting guide.
Sources
- ISO/IEC/IEEE 12207:2017, Software life cycle processes, accessed 17 November 2026.
- ISO/IEC 25010:2023, Product quality model, accessed 17 November 2026.
- NIST SP 800-40 Rev. 4, Guide to Enterprise Patch Management Planning, accessed 17 November 2026.
- NIST SP 800-218, Secure Software Development Framework, accessed 17 November 2026.
Responses (0)
No comments yet. Add the first one below.
Leave a response
Your email address will not be published. Required fields are marked *.