Before you start
- Read a working example of the end product. The campaign launch QA checklist shows the kind of short, ordered check that a procedure should end in. Read it first; much of what follows is about writing the document that sits behind a list like that.
- Know where the task sits in the workflow. The campaign workflow guide sets out the stages and gates; a procedure lives inside one stage and should say which.
- Agree where procedures are stored. One library, searchable, with version history, linked from the work management system. Procedures scattered across personal drives are not procedures.
- Find the person who does the task best. You will interview them or watch them do it.
- Book an hour with someone new. A recent starter, or someone from another team, who will test the draft.
- Have the access the task needs, so that you can follow the steps yourself while writing them down.
Pick the task
Start with a task that recurs often, goes wrong in ways that cost something, and is done by more than one person. Setting up a paid social campaign, building and sending a promotional email, publishing a landing page and adding tracking parameters to a campaign's links are typical first candidates. Avoid starting with work that is mostly judgement, such as writing creative concepts, because a procedure can only fix the repeatable parts. Look at the last few launches for evidence: a task that appears in several post-launch fixes is the one to document first. Where nobody is sure which task causes the trouble, mapping the process as it really runs shows the recurring steps, workarounds and loops back that most need a written procedure. Write the task name as a verb and an object, so that the scope is obvious from the title alone.
Define scope and trigger
State in two or three sentences what the procedure covers and what it does not. "Covers setting up a lead generation campaign in the paid social platform from approved creative to ready-for-launch; does not cover audience strategy or creative production." Then name the trigger: the event that tells someone to start. Usually it is a stage change in the work system, such as a brief being signed or creative being approved. A clear trigger matters because procedures that rely on someone remembering to start them are skipped under pressure. Finally, list the inputs that must exist before the procedure begins, so that the person following it can stop and send the work back if something is missing.
Name the owner
Every procedure has two owners, and the document should name both. The process owner is responsible for the document itself: keeping it accurate, reviewing it on schedule and approving changes. The task owner is the role that performs the task each time it runs, written as a role rather than a person's name so that the procedure survives staff changes. Where several roles contribute, say which step each one performs. If nobody will accept being the process owner, stop and resolve that first: an unowned procedure is out of date within months and becomes worse than none, because people follow it and trust it.
List the steps
Watch the best performer do the task, or walk through it with them, and write each action as a numbered instruction beginning with a verb. One action per step. Include the location of each setting, the values that must be used, and the reason where the reason prevents an error ("set the conversion window before publishing, because it cannot be changed afterwards without resetting learning"). Link to the source of values rather than copying them: the naming convention, the UTMURL tags identifying traffic source. builder, the brand template. Copied values go stale. Keep platform screenshots to a minimum, since interfaces change; describe the setting by name and add an image only where the location is genuinely hard to find.
Add checks and evidence
For each step where a mistake is costly or invisible, add a check: what to look at to confirm the step worked, and what evidence to record. "Open the tag preview and confirm the lead event fires once on the thank-you page; paste the preview link into the work item." Checks involving the tag container (GTMTag management system.) or analytics deserve particular care, because tracking errors are silent until the report is wrong; the tracking verification checklist is a good source of checks to borrow. Then collect the critical checks into a short checklist at the end of the procedure. That checklist is what gets used on every run; the full procedure is what people read when learning.
Write exceptions and escalation
List the situations the normal steps do not cover and what to do in each. Common ones are missing inputs, a platform rejecting an ad, a deadline that cannot be met, a request to skip a check, and anything that looks like a compliance question. For each, say whether the person following the procedure can decide alone, must ask the task owner, or must escalate to a named role. Give the escalation route as a role and a channel, not just "speak to your manager". Exceptions are where procedures earn their keep: the steps are what experienced people already know, but the exceptions are what they learned the hard way.
Set versioning
Give the procedure a version, an effective date, the process owner and a short change log at the top or bottom of the document. Every change, however small, gets a new version and a line in the log saying what changed and why. Decide what kind of change needs approval from the process owner and what can be fixed directly, such as a typo or a broken link. Retire old versions rather than deleting them, so that a past campaign can be checked against the procedure that applied at the time. Version history is also the simplest evidence that the procedure is alive, which matters when audits or new leaders ask.
Test with a new starter
Give the draft to someone who has never done the task, with the access they need, and ask them to complete it on a real or test item while you watch without helping. Note every place they hesitate, ask a question, or do something different from what you intended. Each is a gap in the document. Fix the gaps and, if the changes were substantial, test again with someone else. This step is the one most often skipped, and it is the one that separates a procedure that works from a description of how the author does the task. Experts cannot see the knowledge they take for granted; a new starter finds it in minutes.
Roll out and schedule review
Publish the procedure in the agreed library, link it from the relevant stage and work type in the work management system, and tell the people who perform the task what has changed. If the procedure replaces an informal way of working, walk the team through it once rather than relying on an announcement. Then put a review date in the calendar, typically after the first few runs and then on a regular cadence. At each review, look at the evidence recorded on recent runs, the questions people asked and any launch issues that touched the task, and update the procedure. Measure success by fewer repeated mistakes on that task rather than by a KPIA metric tied directly to an objective. attached to the document itself.
SOP template
The table below is a template for the document. Copy the sections in order; each should be short.
| Section | What to write | Example entry |
|---|---|---|
| Title | Verb and object | Set up a lead generation campaign in paid social |
| Scope | What is covered and what is not | Covers set-up from approved creative to ready-for-launch; not audience strategy |
| Trigger | The event that starts the procedure | Creative status changes to approved in the work system |
| Inputs | What must exist before starting | Signed brief, approved creative, campaign name, tracking parameters |
| Owners | Process owner and task owner, as roles | Process owner: operations lead; task owner: paid social specialist |
| Steps | Numbered actions, one per line, starting with a verb | Create the campaign using the agreed name from the brief |
| Checks and evidence | What to verify and where to record it | Preview the form; paste the test lead ID into the work item |
| Exceptions | Situations outside the steps and who decides | Ad rejected: fix if policy is clear, otherwise escalate to channel lead |
| Checklist | The critical checks for every run | Name, budget, dates, audience, tracking, form test |
| Version and change log | Version, effective date, what changed and why | New version after platform settings were reorganised |
| Review date | When the procedure is next reviewed | After the next few runs, then on the agreed cadence |
A charity documents its email send
Brightwater Trust, a medium-sized environmental charity, chose its fundraising email send as the first task to document because three of its last twelve sends had gone out with broken donation links. The email lead wrote a procedure of 22 steps with six checks, including a test send to a seeded list and a click-through of every link. A volunteer coordinator who had never built an email tested it and stopped at five points; the fixes added a note about which template to use for appeals and an escalation route when the donation platform was down. In the next twenty sends, no broken links went out, and two new team members ran sends unaided within their first month.
Keeping procedures alive
Most procedure libraries decay in the same way. They are written in a burst of enthusiasm, they are accurate for a few months, platforms and people change, and the documents stop matching practice. People notice the mismatch, stop reading them and go back to asking colleagues. The library then sits there looking like evidence of good process while providing none.
Three habits prevent this. First, every procedure has a named process owner and a review date, and operations reports on overdue reviews the same way it reports on overdue campaigns. Second, the procedure is linked from the place where the work happens, so the person doing the task meets it at the moment of need rather than having to remember it exists. Third, launch problems are traced to the procedure that should have prevented them: when a mistake gets through, the post-launch review asks which step or check was missing, and the procedure is updated within days. A procedure is also the usual first fix for a task only one person can do, and the guide to reducing key-person risk finds those tasks across the team with a coverage matrix. Once someone new has run the task from the finished procedure, the coverage matrixA table of a team's recurring tasks and the accounts behind them, recording who can do each one unaided today, so work only one person can do becomes visible. can show a second name against it. If the same kind of mistake keeps reaching customers even after the procedure is updated, the diagnosis for marketing mistakes that keep happening checks whether the gap sits in the procedure, the approval matrix, the launch check or the cover for absences.
Procedures for content have their own considerations, including editorial standards and archiving, and the content governance guide covers those. For where procedures fit alongside workflow, handovers and improvement, the process hub gives the overview. Where a step in a procedure hands work to an AI tool, the instructions given to the tool are kept the same way, with named inputs, test cases, versions and owners, as building a team prompt library shows.
For a finished example of a procedure written this way, the marketing tool onboarding SOP shows steps, owners and checks for bringing a new tool into the stack.
Verify it worked
Test the finished document against its own steps.
- Read the title alone. It should be a verb and an object, and the scope should be obvious from it.
- Check that scope, trigger and inputs sit at the top. The trigger should be an event in the work system, not someone's memory.
- Look for both owners, named as roles. No personal names should appear.
- Open the closing checklist. Every costly or invisible step should have a check behind it and a place to record evidence.
- Ask a new starter to finish the task without your help. Each hesitation should become a fix in the change log. In the illustrative example above, the volunteer coordinator's stopping points went back into the document as fixes.
If it goes wrong
- The new starter hesitates or keeps asking questions. The document describes how the author works, not the task. Note each gap, fix it and test again with someone else.
- People skip the procedure under pressure. The trigger relies on memory, or the document is hard to find. Tie the trigger to a stage change and link the document from the work system.
- A mistake reaches launch although the procedure was followed. A check or an exception is missing. Trace the mistake to the missing step, add it and record a new version.
- Values in the document have gone stale. They were copied, not linked. Replace each one with a link to its source.
Next steps
- Read the campaign workflow guide to confirm the stage and gate your procedure lives in.
- Use the tracking verification checklist to test whether the checks in a tag or analytics step would catch a silent failure.
Common mistakes
- Documenting everything at once. A library of thirty half-finished procedures is worse than five good ones. Start with the tasks that cause the most repeated mistakes.
- Writing for experts. If the procedure only works for someone who already knows the task, it is a reminder, not a procedure. The new-starter test catches this.
- Copying values into the document. Naming rules, parameter values and brand settings belong in their source; link to them so that a change in one place does not leave procedures wrong.
- No exceptions section. The steps cover the normal case; people need help with the abnormal one.
- Naming a person instead of a role. The procedure breaks the day that person changes job.
- No owner or review date. The document drifts out of date and people quietly stop trusting it.
- Storing procedures away from the work. If finding the procedure takes longer than asking a colleague, people will ask a colleague.
