Zero downtime is a promise about customer experience, but it is bought with redundancy, engineering discipline and operating attention. The visible cost is often a second environment or a load balancer. The larger cost is usually the work that makes releases reversible: database changes that run alongside old code, health checks that mean something, observability that shows which version is failing and people who can diagnose a problem at night.
This guide is for a manager who has been quoted for "zero downtime" and needs to know what the quote includes. It explains the cost components, gives formulas you can apply to your own infrastructure and provides a worksheet for comparing proposals.
Where the money goes
- Architecture review: deciding which services can run in parallel and which need a different release pattern.
- Environment duplication: a second application tier, database approach, storage and network for blue-green or canary.
- Routing and health checks: load balancer rules, readiness probes and a switch that works under pressure.
- Database migration design: expand-and-contract changes, backfills and compatibility between two versions.
- Observability: per-version error rates, latency, traces and business metrics, with enough retention to investigate.
- Delivery pipeline: automated build, test, deploy and rollback, plus the permissions that keep it safe.
- Feature flags and rehearsals: the controls and practice that make a release reversible.
- On-call and maintenance: the recurring people cost nobody puts in the project quote.
Cost by strategy
| Strategy | Main cost driver | Capacity implication | People implication |
|---|---|---|---|
| Rolling | Orchestration and version compatibility | Moderate | Platform skill |
| Blue-green | Duplicate environment | High for the application tier | Environment discipline |
| Canary | Traffic routing and metrics | Moderate, depends on share | Observability skill |
| Feature flags | Flag management and cleanup | Low infrastructure | Product discipline |
| Hybrid | Several controls at once | Highest | Strongest team |
Capacity and environment formula
incremental capacity = peak capacity x headroom factor x environment multiplier
For blue-green, the multiplier is roughly two for the application tier. For canary, the extra capacity depends on the traffic share and how long both versions run. Add storage, data transfer and observability retention. Size from the peak you must survive during a release, not from average load. A release that works at noon on a quiet Tuesday may fail at nine in the evening.
People and release formula
implementation effort = services x days per service + database migrations x days per pattern + observability + rehearsal
run cost = releases per month x minutes of human attention per release + on-call cover + platform maintenance
Measure the minutes, not the ceremony. A team that spends two hours on every release because the rollback is untested has not bought zero downtime; it has bought a longer release. The implementation estimate should separate one-off work from the recurring attention the system will demand.
Calculator logic
Inputs: number of services, releases per month, peak concurrency, database migration complexity, rollback target, current downtime cost, team size and whether the platform is managed or self-managed. Outputs: one-off implementation effort, incremental monthly infrastructure, recurring people cost and a risk-adjusted comparison with the current release method.
The calculator should expose its assumptions. A single total is not useful for a decision, because the difference between a one-off project and a permanent operating cost is the part that surprises management. Show both, and show which strategy produced them.
Downtime cost from your own data
Build the cost from your own numbers: revenue per hour multiplied by hours of lost trading, plus staff hours spent recovering, make-good or service credits and the value of orders abandoned during the outage. Add a separate line for irreversible damage, such as a payment file that cannot be replayed or a stock movement that has to be reconstructed by hand.
Do not use an industry average as your own. If you cannot measure revenue per hour, use a range and state the assumption. The purpose is to compare the cost of the release method with the cost of the failures it prevents, not to produce a precise number for a slide.
Budget worksheet
| Line | Type | Basis | One-off | Monthly |
|---|---|---|---|---|
| Architecture review | People | Days | Yes | No |
| Pipeline and rollback | People and tools | Services | Yes | Maintenance |
| Second environment | Infrastructure | Capacity | Setup | Yes |
| Database migration work | People | Migrations | Yes | Review |
| Observability | Tools and people | Data volume | Yes | Yes |
| Rehearsal and training | People | Days | Yes | Refresher |
| On-call cover | People | Hours | No | Yes |
Build or buy
A managed platform reduces implementation effort but adds subscription cost and limits control. A self-managed setup gives control but requires platform skills that must be hired or trained. The decision depends on team capacity and on how much of the release path is genuinely unique to the business. Most SMEs should buy the platform and keep the release discipline in-house.
Procurement questions
- What counts as downtime in the contract, and which services are covered?
- What is the rollback target, and how often is it rehearsed?
- Who owns the pipeline, the infrastructure and the on-call rota?
- What happens when a database migration fails halfway through?
- Is observability included, and for how long are traces retained?
- What is the exit plan if we move to another platform?
A 90-day plan
Weeks one and two measure current releases and outages. Weeks three and four choose a strategy for one service. Weeks five to eight implement the pipeline, health checks and rollback for that service. Weeks nine to twelve rehearse, measure and review. Expand only after one service has a measured, reversible release and the team can explain what it would do at 2am.
We design release pipelines, monitoring and recovery procedures under Managed Technology & Cybersecurity. The pricing page explains how assessment and ongoing operations are scoped. For the strategy trade-offs behind these costs, read zero downtime deployment strategy comparison.
Sources
- Google, Site Reliability Workbook: Canarying Releases, accessed 17 November 2026.
- Kubernetes, Deployments, accessed 17 November 2026.
- Amazon Web Services, Reliability Pillar, AWS Well-Architected Framework, accessed 17 November 2026.
- Martin Fowler, Feature Toggles (aka Feature Flags), 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 *.