Skip to content
Faisal Mahmud TusherDigital marketing, SEO & analytics
with Faisal Tusher

Marketing Operations › Troubleshooting

Live Campaign Problems Found Too Late

Why problems on live campaigns are found days late: no named monitor, unclear limits, no cover, no normal range, silent tracking and no route for reports.
Troubleshootingintermediate13 min readUpdated

First, read how the last few problems were found

Start from the record rather than the latest scare: collect the last few problems that ran on live campaigns before anyone acted on them. With so few behind it, a pattern is a pointer to where to look, not a rate, a trend or a benchmark.

Two records already hold much of what the reading needs. The timeline step of the tutorial on reviewing a marketing error records when an error first became visible to customers, when someone noticed it, who noticed it and by what route, and when it was contained. The error log that tutorial builds keeps the date found and the date contained. Together they give two spans for each problem: the days it ran before anyone saw it and the days it stayed live once someone had. This page is about the first span.

Question for each problemWhere the answer sitsWhat it shows
When was it first visible to customers?The review timelineWhere the run of days begins
When was it noticed, and by whom?The review timeline and the log's date foundHow long it ran unseen, and whether the finder sat inside marketing
How was it noticed?The review timeline; the tutorial's log has no field for itThe route the report took, or the lack of one
When was it contained?The log's date containedHow long it stayed live once someone knew

The timeline records how an error was noticed, but the log holds the two dates and nothing for that answer. As general practice, and not as a step the site's guides prescribe, this diagnosis suggests carrying the answer into the log as a field of its own, chosen from a short fixed list of ways a problem is found.

Sort the problems twice: by who found each one (someone in marketing, someone elsewhere in the business, a customer or a partner) and by the days each ran. Two neighbouring diagnoses start from other symptoms: the diagnosis of marketing mistakes that keep happening reads errors that return through the same gap, and the diagnosis of late campaign launches times the stages before launch. This page begins where those end, with the days nobody saw a live problem; containing it once it is known belongs to the procedures linked under the last cause.

Diagnosis table

Match the pattern in your own records to a row; if two rows fit, check both.

Diagnosis

SymptomCheckLikely causeFix
The first look came when someone remembered, or at month endDoes the work item name a monitor, the measures and the review dates?Nobody is named to watch the campaign after launch and no review dates were setComplete the handover: monitor, measures and dates
A problem was seen and left standing while someone asked if they could touch itIs there a written list of what the monitor may pause or change, and did the finder know it?The person watching does not know what they may pause or changeWrite the list before launch, give it to whoever acts and route all else through the review it reopens
The run of days lines up with a leave, a holiday or a day nobody was inWho was named to watch, and did the calendar show the gap?The monitor is away, or the launch lands on a day nobody can watchShow the watch on the calendar, name a deputy and prove the cover with a real run
An odd first day was read as nothing, or an ordinary dip as a crisisIs there a normal range, and where did it come from?Nothing says what normal looks likeGive the measure a range from its own history or a comparable past launch, with signal rules agreed first
A flat or empty report was read as a slow dayThe report against orders or leads from a source outside the tagsTracking fails quietlyPut a test conversion in the launch check and treat a flat line as the outage trigger
The first report came by chance from a customer or a colleague outside marketingWho found each problem, and how the report reached marketingNo route exists for reportsOne route to one named lead, with customer service briefed before the campaign

Fix, verify and prevent each cause

Each fix points to the page that owns it. A step beyond what those pages say is labelled general practice, and so is each Verify step unless it names the guide that sets it.

Nobody is named to watch the campaign after launch and no review dates were set

The campaign workflow guide ends on a handover after launch that names the monitor and what that person is to watch, review and change.

Fix: treat the handover as a stage with its own exit condition, as the guide's table does, and read it from the work item. As general practice, add a named deputy and a first review date there. The paid campaign launch SOP watches only the first day of delivery before handing the campaign on, so what follows depends on this name.

Verify: open the work items of the next few campaigns; each should show a monitor, a deputy, the measures and the review dates.

Prevent: as general practice, count the handover as complete only when a named monitor is on the work item, and rename the monitor there when they change role or leave.

The person watching does not know what they may pause or change

The guide to handling changes after approval sets the monitor's limits as a written list agreed before launch; everything outside the list follows the reviews it reopens.

Fix: the changes guide has the list written before launch; as general practice, give it to the person who will act on it. For pausing, the pausing underperforming ads procedure rests the decision on a threshold document that the account lead owns, so as general practice the list points at that document. For money moves, the media planning tutorial for a plan in flight writes down before launch what the planner may move without the budget holder; as general practice, the list says which of those moves the monitor holds.

Verify: ask the monitor and the deputy, separately, what each may pause, move or leave for agreement, and read the answers against the written list.

Prevent: as general practice, revisit the list when the threshold document or the media plan's rules change, and keep it beside the monitor's name.

The monitor is away, or the launch lands on a day nobody can watch

The key-person guide counts cover only after a second person has done the task for real, and the work management and calendars guide uses the calendar to spot a launch on a day nobody can monitor.

Fix: treat the watch as a recurring task in the guide's coverage matrix, with a deputy who has run it, and show both names beside the launch date on the work item so that the calendar can reveal a launch that falls when neither can watch. Adding the watch to the matrix is general practice, and so is moving the launch date or naming cover before launch.

Verify: have the deputy run a real first review while the monitor is available but not involved, the key-person guide's own test of cover, and record what it found.

Prevent: as general practice, read planned leave against the calendar when launch dates are set and when someone joins, leaves or changes role.

Nothing says what normal looks like

The control charts framework gives a metric a normal range from its own history, with signal rules agreed before anyone reads the chart; this page quotes none, because any figure would have to come from the team's own history.

Fix: give each measure the monitor watches a range from its own history, with the signal rules agreed before reading. For a launch with no history of its own, the framework borrows a baseline from a comparable past launch. The first days can rest on few events, and reading small numbers in marketing reports shows why a range says more than one figure. The marketing reporting cadence playbook assigns the flagging of anomalies to the weekly view, which is where the range gets read.

Verify: take a recent problem and its report and ask which rule would have fired, and on which day.

Prevent: as general practice, keep the range and where it came from on the work item, and revisit it when the framework's own rules call for it.

Tracking fails quietly

The tracking outage response procedure treats a flat line on a dashboard as a trigger, and has the analytics owner confirm it against a source outside the tags. The tracking verification checklist adds that tracking can decay quietly until a report looks wrong, which is how a tracking outage passes for a slow stretch.

Fix: include a test conversion in the launch check, as the paid launch SOP asks, with the tracking verification checklist as its source; the campaign launch QA checklist also confirms a conversion on a test click. Raise a suspected outage through the outage procedure.

Verify: make a recognisable test conversion on the live page, confirm it appears in the report the monitor reads, then compare that report with orders or leads from a source outside the tags.

Prevent: as general practice, keep that comparison as a standing line in the weekly view, and run the verification checklist again after a release that touches the pages the tags depend on, as it asks.

No route exists for reports

The marketing incident management playbook and the live ad and page error response procedure take over once a report reaches a named lead; where no route exists, the first report reaches marketing by chance.

Fix: write an escalation route for campaign problems: who a colleague or customer service contacts, by what means and who covers when that person is away. The customer service briefing tutorial puts it in a one-page support brief beside the offer, its terms, the questions to expect and the end date; as general practice, give the same route to anyone else who may see the campaign. Where a report concerns an incident involving personal data, or one that reached people who had opted out, the route also names whoever owns data protection: the playbook says to tell that owner straight away, and any duty under the GDPR that follows is for that owner to decide.

Verify: the next report from outside marketing should reach the named lead through the route and show as one log entry; ask customer service whom they would contact about an offer that looks wrong.

Prevent: as general practice, send the brief with each campaign, say so when the route or its owner changes, and add the how-it-was-found field to the log.

Prevention: a named monitor, written limits, a normal range and one route for reports

Four things should stand for each campaign: a monitor with a deputy, a list of what the monitor may do, a normal range for the measures and one route for reports.

HabitWho holds itWhere it is written
A named monitor, a deputy and the review datesThe campaign ownerThe work item, as the handover's exit condition
What the monitor may pause or moveThe campaign owner, with the account lead for the pausing thresholdThe work item, beside the monitor's name
A normal range for each measure and its sourceThe analyst who prepares the weekly viewThe work item, beside the measures
The route for reports and its ownerThe campaign owner, with the customer service leadThe support brief

As general practice, keep that block on the work item so that whoever finds a problem knows where to look; who holds each habit is general practice too.

As general practice, read the same dates again after the next few problems and set them beside the last reading: if the finders now sit inside marketing and the spans are shorter, that suggests the watch and the route are working; if not, return to the diagnosis table with that problem. A few problems cannot show a trend, so treat a reading as a pointer, not a verdict.

As general practice, give the risk its own line in the register. The marketing risk register tutorial gives each risk an owner, an early warning sign and a response prepared in advance; here the monitor owns the risk, the weekly view's anomalies are the warning sign and the route for reports is the response.

A broom maker's autumn offer runs wrong for days (illustrative)

Illustrative example — not a real result

Lerigias Brushworks

Lerigias Brushworks, an invented maker of hand-bound brooms and garden besoms for households and garden centres, ran an autumn offer by search ads and one email, both carrying an order cut-off date. Two days after launch the campaign owner went on leave for 9 days; no deputy was named and the work item held no review date. On the third day an order-form change stopped the confirmation page recording orders, so the weekend report read as quiet. On the fourth day a dispatch clerk saw last year's cut-off date in an ad but did not know whether she could pause it and had nowhere to send the observation. The first report to reach marketing was a customer's call on the sixth day.

The head of the workshop pulled the last 5 entries from the error log. The timelines recorded who noticed each problem and how, but the log held only the dates. Each problem had run unseen for between 3 and 8 days and none had been found by marketing: 2 by customers, 1 by a garden centre buyer, 1 by the dispatch clerk and 1 by the finance clerk comparing the order book with the report. The autumn offer matched each row of the diagnosis table: no review dates set, no written limits, no cover, no normal range against which a flat report could be read, tracking that failed quietly and no route for reports.

The business named a monitor and a deputy with a first review date, wrote a short list of what a monitor may pause without approval, borrowed a normal range for orders from the previous autumn's launch, added a test conversion to the launch check, gave customer service one route for reports and added the how-it-was-found answer to the log as a field of its own. On the next offer the deputy ran the first review on its date while the owner was away and paused one ad with a mistyped date under the written list.

When a live campaign problem turns up alongside other symptoms in the function, the troubleshooting hub points to the remaining diagnoses, and the process hub covers the workflow and procedures behind them.

Common mistakes

  • Waiting for the month-end report. The days before the first look are the cost, so give the first look a date.
  • Naming a monitor without naming what they may do. A name without limits leaves a visible problem standing while someone asks for permission.
  • Treating the launch check as the watch. The check ends at launch; the watch needs its own owner, measures and dates.
  • Reading a flat report as a quiet day. Compare it with orders or leads from a source outside the tags first.
  • Acting on a dip before checking the data. A pause made on broken tracking pauses nothing that was wrong.
  • Judging a campaign from a few early counts. A range, read against the team's own history, says more than one figure.
  • Leaving the how-it-was-found answer in the timeline alone. As general practice, a field in the log lets several problems be read together.
  • Stopping at the person who was away or who found it late. Trace the gap to a missing name, limit, range, test or route, as the review tutorial asks.

Frequently asked questions

Should the person who built the campaign be the one who watches it?

The name matters more than the role. The handover after launch names the monitor, and that person may or may not be the builder. As general practice, give the monitor a named deputy and a written list of what they may pause, and write all of it on the work item, where a colleague who spots a problem can find it.

A report looks quiet. Is that good news?

Not on its own. A flat line, or a platform and the business's own records disagreeing far more than usual, is what the tracking outage procedure treats as a trigger, and a quiet report can mean that tracking has stopped while the business carries on. Check the figures against a source outside the tags before reading them as a slow day, and read them against the metric's normal range rather than against the day before.

How long is too long to find a problem on a live campaign?

The site cites no figure for this, and a figure from elsewhere would not describe your campaigns. Read your own error log instead: the days each problem ran, who found it and how. With few problems behind the reading, a range says more than a single figure, so compare the spread of days before and after a change rather than one average, and treat the exercise as a pointer to where to look, not as a score.

Terms used on this page

Frameworks

Sources

  1. Regulation (EU) 2016/679 (GDPR) — EUR-Lex

Related