First, read the error log for repeats and sort them by the check that missed them
Begin with the log, not with the latest incident. Mark the recent entries that describe the same kind of error reaching customers, whatever the channel or person: offers shown after they ended, prices that did not match the checkout, sends to a segment that should have been excluded. If entries were closed as one-offs with no link to the earlier error they repeat, that is the first finding: entries never linked never show a pattern.
Next, sort each group by how the error got past the team. As the tutorial on reviewing a marketing error finds, the answer usually takes one of three forms: no check existed, a check existed and was not run, or a check ran but could not see this kind of error. Each needs a different fix: a skipped check points at the gate and the pressure around it, a blind one at the procedure or the matrix.
For each group, follow one representative repeat back with the five whys, resting every answer on a record and continuing past any answer that names a person. Where several chains end in the same place, that is the gap to close. Check the dates too: repeats bunched around absences or a sale season have their own causes below.
An error that is live right now is contained first with the live ad and page error response SOP. Repeats in names and tracking values that will not group in reports belong to the diagnosis for inconsistent campaign names.
Diagnosis table
Diagnosis
| Symptom | Check | Likely cause | Fix |
|---|---|---|---|
| The same kind of error returns, and past entries end with a reminder or a name | Whether earlier reviews reached a process cause | Errors fixed and forgotten, or reviews that stop at the person who slipped | Review every repeat and carry each chain to a process, rule or check |
| A repeat comes through a gap an earlier review had named | Whether the earlier actions are confirmed and in place today | Actions agreed after an earlier error never confirmed as done | One owner, a date and a proof per action; confirm on the date |
| Different people make the same slip on the same task | Whether a procedure exists with the check, an owner and a review date | No procedure, or one that lacks the check and has drifted | Write or update the procedure, with an owner and a review date |
| The error carried a price, claim or data risk, but no reviewer for it saw the work | Which matrix rows the work should have triggered | Approval matrix with no row for the risk | Add a row triggered by something intake can ask about |
| The launch check passed, or left no record, on work that went out wrong | Who ran the check, and whether a result is recorded | Launch check run by the builder, or skipped under deadline | A second person runs the check and records the result |
| Repeats cluster when the regular owner is away or someone new has joined | Who did the task, and whether the onboarding pack names the rule | Cover staff and new joiners working without the unwritten rules | Write the rule down, train a second person and test the cover |
| Repeats cluster just before or during a sale, launch or seasonal peak | Repeat dates against the promotion calendar and recent changes | Risky changes made close to or during a peak | Plan a change freeze; finish and test changes before it starts |
Fix, verify and prevent each cause
Errors are fixed and forgotten, or the review stops at the person who slipped
A fix removes the error; only a review changes the route it took. Among the reviewing tutorial's written triggers for a review is an error of a kind that has happened before, so every repeat qualifies. Reviews mostly fail by ending too soon. A chain that stops at "the coordinator missed it" leaves the process untouched, and an action list built on context, such as taking more care in busy weeks, is the tutorial's own sign that the review stopped early.
Fix: review each group of repeats with a facilitator who was not involved, carry every chain until it reaches a process, rule or check the team can change, and separate causes from what made the error last longer and from plain context.
Verify: each action from the last few reviews names a field, a step, a check, a matrix row or an owner, and none asks people to be more careful.
Prevent: when an answer names a person, ask next why the work relied on their memory.
Actions agreed after an earlier error were never confirmed as done
Many repeats were predicted: the earlier review named the gap, and the agreed change never landed. The reviewing tutorial calls the confirmed field in its log the commonest gap in any review process, and an action that never happened is the usual reason an error returns.
Fix: assign each action a single owner, a due date and the evidence that will prove it landed, and have whoever runs the log go back on that date to confirm the change is in place; the marketing incident management playbook gives that follow-up to its operations owner.
Verify: check each earlier action against what exists today, such as the field on the build sheet or the line in the procedure.
Prevent: treat an action past its date as an open risk, chased at the monthly read rather than left to drift.
The task has no written procedure, or the procedure lacks the check
When different people make the same slip on the same task, the task usually runs from memory or from a procedure that no longer matches practice. The tutorial on writing a marketing SOP keeps a standard operating procedureThe written way of doing one recurring task: who owns it, what starts it, the steps in order, checks and exceptions, so anyone qualified gets the same result. alive with three habits: an owner plus a set date for its next review, a link from the place the work happens, and an update within days when a mistake gets through.
Fix: write the procedure for the task the repeats came through, or add the missing step and its check to the existing one, with the evidence to record and an owner and review date at the top.
Verify: someone who has not done the task follows the procedure on a real item while you watch; the check that missed the repeats is now a step they cannot pass without doing.
Prevent: report overdue procedure reviews the way the team reports overdue campaigns.
The approval matrix has no row for the risk
Some repeats never reached the reviewer who would have caught them. The approvals and compliance guide routes work to reviewers by the risk it carries through an approval matrixA table that says which reviews each type of marketing work needs, such as legal, brand or budget, and who gives each, matched to the risk of the work., and says that incidents which got through point to a missing row. Each row is triggered by something observable that the intake form can ask about, such as a price or a new claim. The claims and pricing rows exist because marketing must not mislead and claims must be capable of substantiation before they are made, which the approvals guide traces to the UK's Advertising Standards Authority; whether a piece breaks a rule is for the reviewer named in that row to decide.
Fix: add a row whose trigger is what the repeats had in common, in words intake can ask, with a reviewer and a standard.
Verify: the next pieces carrying the trigger show the new review on their approval record.
Prevent: run the approval audit, which samples live work and follows each gap back to its source in the matrix, the workflow or the records.
The launch check is run by the builder, or skipped under deadline pressure
A launch check that passed faulty work was usually run by its builder, or skipped when the date was tight. The campaign workflow guide gives the check to a second person, since whoever built the work tends to see what they intended instead of what is on the page, and records the result against the campaign.
Fix: make a second person's recorded check the condition for going live, starting from the campaign launch QA checklist, and add an item for whatever the repeats slipped past, such as when an offer ends as well as when it starts.
Verify: every campaign launched since has a recorded check by someone other than its builder.
Prevent: when a launch is tight, keep the check and move the date; if dates keep forcing that choice, the diagnosis for campaigns that keep launching late finds where the days go.
Cover staff and new joiners work without the rules in the regular owner's head
Repeats around holidays, illness or a new hire come from rules nobody wrote down. The regular owner knows that a certain price goes to the commercial owner; the colleague covering for a fortnight does not. The tutorial on building a marketing onboarding pack lists the rules a new starter could break in their first days before anyone mentions them, each linked to its owner's page, and starts them on a low-risk real task. The guide to reducing key-person risk relies on a coverage matrix to spot tasks only one person can do, and asks that person what they check that no procedure records.
Fix: write the missing rule into the procedure and the pack, and train a second person with real runs.
Verify: the second person runs the task for real from the documents while the usual owner is available but not involved.
Prevent: update the pack after every joiner, and use each planned absence to test cover before an unplanned one does.
Risky changes are made just before or during a peak
When repeats bunch around a sale or a seasonal high point, the cause is often timing. The tutorial on planning a change freeze traces many peak-period breaks to edits made shortly before or inside the busy weeks, when there is little time to test or put things right, and uses a change freezeA set period around a peak when changes that are hard to reverse are stopped while routine, reversible work continues through an agreed route. to hold back hard-to-undo changes while everyday reversible work continues.
Fix: name the period and the kinds of change it covers, finish and test what the peak needs before it starts, and send any change during it through a written exceptions route with a named decision-maker.
Verify: the change record for the next peak shows every covered change made during it approved as an exception.
Prevent: review the freeze afterwards, including what broke and whether a change caused it, and set the next dates from that history.
Prevention: a repeat field in the log, confirmed actions and a monthly read beside the risk register
Errors handled one at a time never show the gap they share. Give the error log a repeat field that points each new entry to any earlier one that shares its cause, and a confirmed field that records when each agreed action was checked as done. Fill the repeat field when the entry is written.
Then read the log once a month, as the incident playbook does, beside the marketing risk register. Confirm the actions due, chase the open ones, and look for the patterns one review cannot show: one gate that keeps letting different errors past, one error type turning up in several channels, or slips that gather at a single handover or season. For each repeat, look up its risk on the risk registerThe single table of what could knock a marketing plan off course, each risk with a warning sign, an owner and a response prepared in advance.: did the early warning show, and did the planned response hold? A repeat that matches no risk is a new risk to add.
The reviewing tutorial warns that a falling error count can mean errors are going unreported. Two other figures tell you more about learning: how many reviews ended with every action checked off, and how many repeats now have a named cause. Report them as measures of the system, never of individuals.
Take extra care when a repeat touches people's data. Where an error exposed personal dataAny information relating to an identified or identifiable living person, such as a name, email address, location data or an online identifier. or contacted anyone who had opted out, the person who owns data protection hears about it straight away and decides whether the GDPR places any duty on the business; the review records the facts and does not decide. When opted-out people are contacted more than once, the usual reason is a suppression list that some sending tool never receives; the UK regulator's direct marketing guidance expects an objection to be respected, so treat the gap as a process cause, not bad luck.
Keep the other cadences too: procedure reviews on their dates, the approval audit on schedule, the coverage matrix current and a freeze before each peak. If the log also shows late launches or reporting delays, the marketing operations troubleshooting hub lists the other diagnoses so you can pick the right starting point.
A rail holiday operator stops repeating an expired fare in its ads (illustrative)
Drovisk Rail Holidays
Drovisk Rail Holidays, an invented operator of escorted rail holidays with a marketing team of 6, found that a paid social ad showing an early-booking fare that had already ended had reached customers for the third time in 11 months. Each time the ad was paused within hours, the old fare was honoured for the 20 to 40 customers who asked, and the error was logged as a one-off.
Read by the check that missed each error, the log showed a pattern. All 3 ads had been built by colleagues covering the paid social specialist's leave. The build procedure, last reviewed 2 years earlier, had no step for a fare's end date, and the launch check confirmed fares when they started, not when they ended. The matrix sent prices to the commercial owner only when an offer was new, and all 3 fares were extensions of existing offers. Of the 4 actions agreed after the first error, 2, an end-date field on the build sheet and a named monitor for always-on ads, had never been confirmed as done.
The team added an end-date step and check to the procedure, owned by the paid social lead with a review date 3 months out; added a matrix row sending any dated price to the commercial owner; trained a second person, who ran the build for real twice before the specialist's next leave; confirmed the 2 open actions; and began each monthly read with the log's repeat field. In the following 8 months no fare ad ran past its end date, and the launch check caught 1 ad whose end date had been left blank.
Common mistakes
- Logging each repeat as a one-off. Without a link to the earlier entry, the pattern never appears.
- Ending the review at a name. The person changes; the gap stays open for the next one.
- Agreeing actions and never checking them. The earlier review often named the gap the repeat came through.
- Fixing habits instead of documents. A change that lives in people's heads leaves with them.
- Letting the builder check their own work. They see what they meant to do, not what they did.
- Assuming cover staff know the rules. Write the rule down and test the cover before the absence.
- Changing tags or templates in the week of a sale. Finish and test before the freeze, or wait.
- Celebrating a falling error count. It may mean errors are going unreported.
