Change Freeze

What is a Change Freeze?

A change freeze is a set period when changes to production systems are restricted or banned, except for approved emergencies. It reduces the risk of outages at critical times such as peak trading or year-end.

Updated November 2026 · 2 min read


Why organizations freeze changes

Most outages are caused by changes: new code, configuration updates or infrastructure work. During periods when an outage would be especially costly, many organizations decide the safest change is no change. Freezes are common around retail peaks such as Black Friday, financial year-end close, major product launches, regulatory deadlines and holiday periods when fewer staff are available to respond to incidents.

TypeWhat is restrictedTypical use
Change freezeAny change to production systemsPeak trading, year-end, holidays
Code freezeMerging new code to a release branchStabilizing a release before launch
Soft freezeOnly high-risk changes; low-risk changes allowed with approvalLead-up to a hard freeze
Hard freezeAll changes except emergenciesThe critical window itself

How a change freeze works

  1. Announce the freeze dates well in advance so teams can plan work around them
  2. Define what is in scope: which systems, environments and change types
  3. Set an exception process, usually approval by a Change Advisory Board (CAB) or an on-call lead, for emergency fixes
  4. Increase monitoring and on-call coverage during the freeze
  5. Plan the thaw: a controlled return to normal, rather than releasing everything at once

An example

An online retailer freezes changes to its checkout, payments and inventory systems from mid-November to early December. Teams finish and release their peak-season features by late October. During the freeze, one emergency fix is needed when a payment provider changes an API; the on-call lead and CAB chair approve it within an hour, with a rollback plan. After the freeze, changes are released in small batches over a week, rather than all on the first day, to avoid a spike in incidents.

The trade-offs

Freezes reduce risk during the freeze window, but they have costs. Work piles up, so the first releases afterwards are larger and riskier. Teams lose practice deploying. And a freeze can hide underlying weaknesses in testing and release practices. High-performing teams, as measured by DORA metrics, tend to rely less on freezes because small, frequent changes with automated testing are lower risk to begin with. Practices such as feature flags, canary deployments and a strong CI/CD pipeline make it possible to keep releasing safely even at busy times.

Keeping freezes useful

  • Keep freezes short and limited to the systems that really matter
  • Prefer targeted freezes to blanket bans across the whole estate
  • Track error budget use and incidents to decide whether a freeze is needed
  • Review each freeze afterwards: did it prevent incidents, and what did it cost in delay?
  • Use the time for work that does not touch production, such as planning and technical debt

Frequently asked questions

What is the difference between a change freeze and a code freeze?

A code freeze stops new code from being merged into a release, usually so it can be stabilized and tested before launch. A change freeze is broader: it restricts any change to production systems, including configuration and infrastructure, for a period of time. A code freeze is about a release; a change freeze is about operational risk.

Can you make changes during a change freeze?

Usually only emergency changes, such as fixes for security vulnerabilities or production incidents, and only through an approved exception process. That often means sign-off from a Change Advisory Board, a senior engineer or the on-call lead, plus a tested rollback plan. Routine and planned changes wait until the freeze ends.

Are change freezes good practice?

They can be a sensible safeguard for short, critical periods. But frequent or long freezes are often a sign of fragile release processes. Teams that deploy small changes often, with automated testing, monitoring and easy rollback, can usually keep releasing safely and need freezes less, or only for a narrow set of systems.