An SOP fails when it documents the ideal process rather than the real one. The real one has a step where the system returns an error, a form that is sometimes missing a field, and a customer who replies with the wrong attachment. If the document does not mention those, the reader hits one and abandons the page — and will not come back to it next time.
Every structure below shares four sections. Trigger — what starts this procedure. Owner — the role accountable, not a person's name. Steps, with decision points marked. Done criteria — how you know it finished. Most SOPs include the third and skip the rest, which is why nobody can tell whether the procedure applies to them.
Pick the structure by the shape of the work
| Type | Fits | Decision points | Typical length |
|---|---|---|---|
| Linear | Same steps every time | None | 1 page |
| Branching | Steps depend on input | 2–4 | 2 pages |
| Checklist | High-stakes repetition | Implicit | 1 page |
| Escalation | Exception handling | 1 | 1 page |
| Handover | Work crossing shifts | 1 | 1 page |
| Recurring cycle | Monthly or weekly close | 2–3 | 3 pages |
| Incident | Time-critical response | 3–5 | 2 pages |
1. Linear procedure
Use when: the task runs the same way every time, with no variation worth documenting.
Trigger: [what starts this]
Owner: [role]
Time: [typical duration]
1. [Action] in [system].
2. [Action], checking [field].
3. [Action] and record [reference] in [location].
4. Confirm [condition] before proceeding.
Done when: [observable end state].
If it fails: [first thing to check].
2. Branching procedure
Use when: the route depends on information you only have once you start.
Trigger: [what starts this]
Owner: [role]
1. Check [condition].
2. If [condition A]: [steps].
3. If [condition B]: [steps].
4. If neither applies: escalate to [role] — do not guess.
Done when: [observable end state].
3. Checklist
Use when: the work is repetitive and a missed step is expensive.
Trigger: [what starts this] · Owner: [role]
- ☐ [Item] — verified by [method]
- ☐ [Item] — recorded in [location]
- ☐ [Item] — confirmed by [role]
- ☐ [Item] — [sign-off requirement]
Done when: all items checked and [repository] updated.
4. Escalation path
Use when: the procedure exists mainly to say who decides when the normal route breaks.
Trigger: [the exception]
First response: [action] within [timeframe]
Escalate to: [role] if [condition] · contact [channel]
Information to include: [item], [item], [item]
Done when: [role] has acknowledged and [record] is updated.
5. Handover between shifts or people
Use when: work passes from one person to another and context is easily lost.
Mandatory fields:
| Field | What goes in it |
|---|---|
| Status | Where the work stands right now |
| Open items | What is unfinished and who has it |
| Attempted | What has already been tried and ruled out |
| Risks | What is likely to break next |
| Next action | One specific step, with an owner |
Done when: the receiving person can state back what they are picking up.
6. Recurring cycle (monthly or weekly close)
Use when: the procedure repeats on a calendar and depends on inputs from others.
Cycle: [frequency] · Owner: [role] · Deadline: [day and time]
1. Collect [input] from [source] by [deadline].
2. Reconcile [record] against [record]; investigate any gap above [threshold].
3. Prepare [output] and circulate to [recipients].
4. Archive in [location] with [naming convention].
Done when: [output] circulated and archive updated.
7. Incident response
Use when: minutes matter and the priority is stabilising before understanding.
Severity: [how to classify] · Owner: [role on call]
1. Confirm scope: [what to check].
2. Stabilise: [the action that stops the bleeding].
3. Communicate: [who is told, when, through what channel].
4. Record timeline: [what gets logged and where].
5. Review: [when, who attends, what is produced].
Done when: service restored, [stakeholders] informed, and review scheduled.
The section most SOPs forget
Almost every procedure document ends at the last step, which is precisely where the reader needs help. Always write the "if this does not work" section. It should name the first thing to check, the person who decides, and the point at which to stop and escalate rather than continue. In our experience this is the only part of an SOP that gets read under pressure, and it is the part most often missing.
Keeping them alive
Three habits prevent the usual decay. Date every SOP and set a review interval — quarterly for anything customer-facing, annually for the rest. Link the SOP from where the work happens, not from a shared drive nobody opens. Ask whoever last used it to correct it, since the person who hit the gap is the only one who knows what the document should have said.
Get the skeleton down, then add the failures
A tool can produce the structure in a minute. The "if this does not work" section has to come from whoever has done the job.
Try Rytr →Frequently Asked Questions
How long should an SOP be?
One page for a linear task, two to three for anything with decision points. If it runs longer, it is usually several procedures in one document.
Who should write it?
The person who does the work, reviewed by whoever is accountable for it. Written from a distance, it documents the ideal rather than the real process.
What is the most important section?
The failure section — what to check first and when to escalate. It is the part read under pressure.
How often should SOPs be reviewed?
Quarterly for customer-facing procedures, annually for internal ones. Attach a date and an owner to every document.
Do small teams need SOPs?
For anything repeated and expensive to get wrong, yes. The value is in handing work over without losing context.
Can AI write our SOPs?
It can produce a usable first draft from a description of the steps. The exceptions and escalation rules need your own knowledge.