Software maintenance is the line item most likely to be underestimated and least likely to be examined. A system is approved with a build budget, launched, and then kept alive on whatever remains in the operating budget. When a security patch, a payment gateway change or a failed dependency arrives, the work is treated as an emergency rather than a planned cost. The result is a system that is maintained only when it breaks.
A defensible maintenance budget starts with the work the system will require, not a percentage copied from a proposal. This guide sets out the cost lines, a planning model, the distinction between run and change, and the questions that expose an unrealistic support quote.
Why maintenance budgets drift
Three forces push the number off course: scope grows without a budget change, the system ages, and responsibility is split between internal staff and a vendor. A budget that ignores any one will be revised in-year.
Another cause is mixing maintenance with enhancement: small features are approved as projects, then absorbed into a support retainer, which looks expensive while the enhancement pipeline stays invisible.
What the budget must cover
Software maintenance is not one activity. ISO/IEC 14764 distinguishes corrective, adaptive, perfective and preventive maintenance. Each has a different cost driver and should be budgeted separately.
Corrective maintenance
Fixing defects found in production: failed transactions, incorrect calculations, broken integrations, display errors. The cost depends on defect volume, the quality of the test suite and how quickly issues are detected. Track it by severity and by component, not as a single support total.
Adaptive maintenance
Keeping the system compatible with a changing environment: payment gateway API changes, browser updates, operating system and database upgrades, tax or regulatory changes, and third-party service deprecations. This work is predictable in principle and easy to postpone in practice. A budget with no adaptive line will eventually pay for it through emergency work.
Perfective maintenance
Improving performance, usability or maintainability without adding a new business feature. Examples include query tuning, reducing a slow report, refactoring a fragile module or improving logging. Perfective work competes with new features and is often the first item cut. A small allocation protects the system from gradual decline.
Preventive maintenance
Reducing the probability of future failure: patching, dependency updates, backup restore tests, certificate renewal, access reviews and monitoring improvements. Some of this is routine; some is project-like. Both need an owner and a calendar.
Building the annual envelope
Build the budget from four blocks. First, a support baseline for the agreed service level: incident triage, fixed defects, monitoring and regular maintenance. Second, an adaptive and security reserve for upgrades and patches. Third, a change pool for small enhancements, with a defined approval process. Fourth, a contingency held by the business, not the vendor.
A simple formula:
Annual maintenance envelope = support baseline + adaptive and security reserve + change pool + contingency
Price each block from evidence. Use last year's ticket volume by category, the dependency upgrade calendar, the number of integrations and the expected change requests. Where evidence is missing, state the assumption and review it after one quarter.
A percentage model with a floor and a ceiling
Many teams use a share of the original build cost as a starting hypothesis. Treat it as a model, not a market rate. For a business-critical system with custom integrations, plan toward the upper end. For a simple, stable internal tool with few integrations, the lower end may be adequate.
- Lower band: a stable system with a small user base, few integrations, good test coverage and a low change rate.
- Middle band: a system with several integrations, regular regulatory or payment changes and a moderate enhancement pipeline.
- Upper band: a system with high availability requirements, complex integrations, legacy dependencies or a large security surface.
As a planning hypothesis, use 15-20% of the original build cost per year for the lower band, 20-25% for the middle and 25-30% for the upper. These are model inputs, not a market rate. Set a floor below which the system cannot be responsibly supported, and a ceiling above which replacement deserves evaluation.
Cost drivers by system age
A new system usually costs less to maintain in its first year because the team remembers the design and the dependency set is current. As it ages, more effort goes to adaptive work, security patching, knowledge loss and testing. A system without documentation becomes dependent on specific people, and the cost of their availability rises.
Replacement is not automatically cheaper. A rewrite carries delivery risk, data migration and retraining. Compare the three-year maintenance envelope of the current system with the build plus run cost of the replacement. Sometimes the honest answer is to invest in maintainability rather than replace.
Run versus change
Separate the budget into run and change, and publish both. Run covers keeping the current system healthy and available. Change covers new behaviour, new reports and new integrations. Mixing them hides the true cost of the enhancement pipeline and makes the support retainer look inefficient.
Give the change pool a lightweight approval path with a threshold. Small requests should not require a steering committee, but they should be recorded, estimated and counted against the pool. Each quarter, report how much of the pool was consumed by unplanned work; that figure is the best argument for fixing the underlying cause.
What to ask a maintenance vendor
- What is included in the baseline, and what is billed separately?
- How are severity levels defined, and what response and restoration times apply?
- Are security patches and dependency upgrades included, or treated as projects?
- How is knowledge transferred, and what do we own if we leave?
A low retainer with vague inclusions is an option to pay more later. Ask for a worked example of the last month, with tickets, hours and categories.
A 90-day budgeting sequence
- Days 1-15: inventory the system, integrations, dependencies, environments and named owners.
- Days 16-30: classify last year's work into corrective, adaptive, perfective and preventive; calculate the actual split.
- Days 31-45: define service levels and the boundary between run and change.
- Days 46-60: obtain two maintenance quotes against the same scope and service level.
- Days 61-75: build the four-block envelope and test it against the percentage bands.
- Days 76-90: approve the budget, publish the change process and set the quarterly review.
Budget checks
- Share of maintenance effort spent on unplanned work.
- Median time to restore service by severity.
- Patch and dependency update latency.
Planning maintenance support
If you need a maintenance plan that matches the system you actually run, start with Custom Software & Integrations. The pricing page explains how support, enhancements and security work are separated. For the wider budget context, read how SME IT budgeting changes year by year.
Sources
- ISO/IEC 14764, Software engineering — Software life cycle processes — Maintenance, accessed 17 November 2026.
- NIST SP 800-40 Rev. 4, Guide to Enterprise Patch Management Planning, accessed 17 November 2026.
- CISA, Known Exploited Vulnerabilities Catalog, accessed 17 November 2026.
- OWASP, Top 10 Web Application Security Risks, accessed 17 November 2026.
- ISO/IEC 25010, Systems and software quality models, 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 *.