Mapping ITIL change types to merge requests
ITIL sorts changes into three types: standard, normal and emergency. Each type has its own path to approval, and most change management processes in larger organisations are built on that split. Merge requests rarely are. This guide shows how to carry the three change types over to GitLab and GitHub merge requests with change policies, so a documentation fix, a production release and a hotfix each follow the path meant for it.
The three change types in brief
In general terms, ITIL describes the three types like this:
- A standard change is low risk, well understood and pre-authorized. It follows a known procedure, so it does not need a fresh approval each time.
- A normal change is assessed and authorized before it is made, by whoever the organisation names as the change authority.
- An emergency change must be made as soon as possible, often to restore a service. Its authorization is faster, and it is reviewed afterwards.
The value of the split is proportion: approval effort goes where risk is, and routine work is not held up by the same process as a database migration.
How a change policy expresses a change type
In Merge Pilot, a change policy decides which merge requests it covers, what change type they are, and which controls must pass before the merge. Policies belong to a VCS connection and are tried from top to bottom; the first active policy that matches decides. A catch-all at the bottom covers everything the policies above it do not. See How a policy is chosen.
Each policy has one of four outcomes, set on its Change type tab:
| Change type | Change approval | Merge |
|---|---|---|
| Standard change | None, unless Open a change approval as a record is on (then nobody has to approve it) | Merges once its enforced controls pass |
| Normal change | Opened in the tracker before the merge; must be approved by the people the policy asks for | Held until the change approval is approved and every enforced control passes |
| Emergency change | Opened for a review after the merge, within the Review deadline | Merges once its enforced controls pass; the approval does not hold it |
| Refuse | None | Nothing it covers can merge; the policy's description tells the author what to do instead |
Controls apply to every change type. A standard change still waits for its enforced CI or secret scan; "pre-approved" means no person has to sign off, not that nothing is checked. See Change types.
Deciding which merge requests are which
The change type is only useful if the right merge requests land in it. A policy's conditions, on its Scope tab, can match on:
- target and source branches, such as
main, release/*orhotfix/**; - repositories, such as
team/payments-*; - the author: people only, bots only, or named accounts;
- labels the merge request has, or must not have;
- changed files: only these files changed, touches any of, touches none of;
- the size of the change, in files or lines;
- draft state.
A merge request must meet every condition a policy sets. Patterns are globs, not regular expressions; a pattern that looks like a regular expression is refused when the policy is saved. See Matching.
A worked example
A typical ordered list for one VCS connection:
- Emergency changes: emergency change; labels Has any of
emergency; authors People only; CI and secret scan enforced; review within 72 hours. - Documentation: standard change; Only these files changed
**.md, docs/**; secret scan enforced. - Dependency updates: standard change; Bots only, listing
dependabot[bot], renovate[bot]; CI, secret scan and dependency scan enforced. - Normal changes (the catch-all): normal change; work item required; CI and secret scan enforced; one approval from the project's approver roles.
The built-in ITIL change management template sets up the first, second and fourth of these; the Dependency update bots template adds the third. Templates are a starting point and every policy stays editable. See Templates.
Things to watch when matching
- Order matters. A policy with no conditions above others makes everything below it unreachable. The policy list warns about this, and about policies that can never be reached because one above always matches first.
- Renamed files count twice. A file moved into
docs/counts with both its old and new path, so moving code intodocs/does not make it a documentation change. - Unknown diffs fall through. When the VCS cannot send the list of changed files (a very large change), a policy with a file or size condition does not match, and a later policy decides. A narrow standard policy cannot be proven to apply to a change nobody can read.
- Bots only means bots. With Bots only and no bot listed, every bot account matches, including CI tokens and GitLab project or group access tokens. List the bots you mean.
Before saving, Which policy applies? takes a described change through the saved policies and shows, for each one above the deciding policy, the first condition it fails. Test run tries unsaved edits on a real merge request. See Trying policies out.
Normal changes: the change authority
ITIL leaves the choice of change authority to the organisation. A normal policy expresses it in Who must approve: one or more ways to approve, each with requirements such as 1 of Security and 2 of Operations, or 1 of the change advisory board as an approver pool. Any one way met approves the change.
Two refinements bring this closer to how change authorities usually work:
- Approvals by risk level lets a requirement ask for a different number of approvals at low, medium, high and critical risk, so a larger or riskier change needs more sign-off under the same policy.
- Code owners of the changed files asks the people the CODEOWNERS file names for the files that changed.
The change approval is opened in Jira or the built-in tracker, assigned as the policy says (never to the author), and its decision holds the merge gate. See Who must approve.
Standard changes: pre-approved, still checked
A standard policy is for changes your organisation has agreed need no individual approval. Typical candidates are documentation-only changes and dependency updates from known bots, where the checks carry the weight: CI must pass, no secrets, no new high-severity vulnerabilities.
If your process wants a ticket for every change, even pre-approved ones, turn on Open a change approval as a record. A change approval is then opened for the record, and nobody is asked to approve it. See Standard changes.
Emergency changes: merge now, review after
An emergency policy lets a change merge as soon as its enforced controls pass, without waiting for approval. A change approval is opened for the review afterwards, due within the policy's Review deadline (72 hours by default). The review uses the same Who must approve and the same checks as any approval: the author never decides it.
Who may declare an emergency
The person who declares an emergency is whoever added the label the policy matches on; for a policy that matches on something else, the author. By default anyone who may label the change may declare one, so a hotfix never waits. Every declaration is recorded on the merge request, written to the audit log as emergency_declared and announced.
Because an emergency skips the approval before the merge, it is the path most worth limiting. The company's Emergency changes settings, which an emergency policy can override with This policy sets its own rules, offer:
| Rule | Effect |
|---|---|
| Who may declare one | Anyone who may label the change, or only people of one approver pool |
| The author may not declare their own change an emergency | Someone other than the author must add the label |
| Limit emergencies per repository | Once a repository has had this many emergencies in the window, further ones are not emergencies |
When a rule is not met, or who added the label cannot be read from the VCS, the emergency policy does not match and the next matching policy (usually the normal one) covers the change. See Who may declare an emergency and Emergency changes.
When the review is late
When a quarter of the review time is left (at most a day before the deadline), Change approval due soon is sent. A review still undecided at its deadline is written as emergency_review_overdue, sent as Emergency review overdue and commented on the change approval. If an emergency change merged before its review change approval could be opened, for example while the tracker was unreachable, the review is opened afterwards for changes merged within the last 30 days.
Change freezes and the three types
A company's change calendar can hold merges during a freeze. The freeze mode decides how each type is treated:
- Hold merges: every merge is held, emergencies included.
- Emergency changes only: changes under an emergency policy pass; every other change is held.
- Raise risk only: nothing is held by the freeze; assessed changes score critical risk.
Approvals go on during a freeze, so normal changes can be approved and merge once it ends. See Change freezes.
Rolling it out
- Start from the ITIL change management template on a VCS connection, then adjust conditions and approvers.
- Add the Dependency update bots template if bots open merge requests in your repositories.
- Set the company's emergency rules before anyone relies on the
emergencylabel. - Check the order with Which policy applies? for a docs change, a bot update, a labelled hotfix and an ordinary feature.
- Watch repositories first if you want to see what the policies would do before they block anything; the monitoring report shows what change management would have held.
Next steps
The Change policies feature page sums up matching, change types and controls. If your change advisory board works in an ITSM tool today, the ITSM CAB and merge gate comparison sets the two approaches side by side. To map your own change types to policies with us, ask for a demo.