First, decide which count is missing: the activity's own firings or one product's attributed share
The Floodlight tags guide states the rule that sorts every report of missing conversions into one of two piles: the CM360 activity count is the source, and the attributed counts in DV360Google's enterprise DSP. or Search Ads 360 are each product's share of it. A linked advertiser attributes only the impressions and clicks it served, within the activity's windows, under its own rules, so the same activity can show different counts in three interfaces while the raw count in the ad server stays the one that says whether the tag works. Open the Floodlight configuration and read the activity's status and when it last received a tag, then run a Floodlight report for the period in question. An activity that has received nothing is a tag problem. One that receives firings while a buying platform shows zero is a link, window or delivery problem, and no amount of tag debugging will find it.
The measurement plan asks of every number which objective it serves, and that question narrows the search further. Which decision is starved by the missing count: a bid strategy in DV360 that optimises towards the activity, a conversion column in Search Ads 360, an audience list built from it, or a monthly report? Each reader joins the activity differently. A bid strategy reads it through the linked advertiser; a search column reads it through the click identifier on the landing page; an audience list reads it through the configuration's link; a report reads whichever product it was built in. Naming the reader tells you which join to test first and which campaigns to protect while you do.
Then read the dates. A count that fell to zero on one day points at a change on that day: a container version, a site release, a consent platform update, a link removed, a window changed. A count that was always zero points at the build: an activity created but never deployed, a configuration never linked, a group whose type or tag strings never matched the tag on the site. The tracking outage response procedure sets out who confirms the outage and how to protect campaigns that bid on the missing signal while the cause is found; start it the moment the zero is confirmed rather than after the fix, because an automated strategy fed a false zero learns from it. If the silent tag is a Tag Manager tag of any kind, the diagnosis of a Tag Manager tag that does not fire separates a tag that was never evaluated from one that was held back; if it is the Google Ads conversion tag, the diagnosis of Google Ads conversions not tracking walks that product's own chain. This page stays with the Floodlight-specific layers: the configuration, the activity, the linked advertisers, the windows and the attribution each product applies.
Diagnosis table
Diagnosis
| Symptom | Check | Likely cause | Fix |
|---|---|---|---|
| The activity has never received a tag, or stopped on the day a container version was published | Preview mode and Tag Assistant on the live page: whether the tag fired, on which trigger, with which tag strings and counting parameters | The tag never fires: not published, wrong trigger, mismatched tag strings or a counting-method parameter wrong | Publish to the right environment, point the trigger at the event the site sends, and make the strings and parameters match the activity as configured |
| The activity fires in testing with consent granted and never for most real visitors | The consent tab in Tag Assistant for each branch; whether defaults are set on the consent initialisation trigger | Consent refused or the consent trigger misconfigured, so the tag fires only on the consented branch | Set defaults before the Google tag, re-issue the update from the consent platform, and test the refused and granted branches on every template |
| The ad server counts the activity but the buying platform cannot see it or bids on an activity nobody can explain | The configuration the activity sits in against the one the linked advertiser reads; the register of what depends on each activity | The activity sits in a configuration the buying platform is not linked to, or in an inherited group nobody can explain | Trace the activity and its dependents before changing anything, then build the right activity beside the wrong one and move readers across |
| Every activity in the configuration shows zero in the buying platform while the ad server counts them all | The link between the buying advertiser and the ad server advertiser, checked from both sides | The DV360 or SA360 advertiser not linked to the CM360 advertiser that owns the configuration | Link the advertisers, fire a test conversion and confirm it appears as an activity in the buying platform |
| Firings arrive in the ad server but attributed conversions are few or none, and the search column fell while the engine's own held | The Floodlight report split by click-through and view-through; the click identifier on landing pages; the Conversion Linker on every page | Firings arrive but no media touched that browser inside the windows, or the linker is missing | Deploy the linker on every page, restore the click identifier, and set windows that cover the journey the plan expects |
| The count vanished on the day a new page or template went live, with the container untouched | Preview mode on the changed template: the data layer event name, the variables the tag reads and the confirmation page address | A site release renamed the data layer or moved the confirmation page | Restore the event name or update the trigger and variables, and add the activity to the release checklist for that page |
Fix, verify and prevent each cause
The tag never fires: not published, wrong trigger, mismatched strings or a counting parameter wrong
The Floodlight tags guide describes the chain a firing has to complete. A trigger in the container fires when the page or the data layer says the action happened; the Floodlight tag builds its request from its fields and from the variables that read the data layer; the browser sends it to CM360, where the activity is identified by its tag strings and the counting method decides whether the firing is a new conversion. Each link can be wrong on its own. A container change left unpublished, or published to a staging environment, leaves the live site with the old version. A trigger that listens for an event the site no longer sends never matches. Tag strings that differ from the group and activity strings in the configuration send the firing to an activity that does not exist, and the guide warns that changing live strings breaks every deployed tag at once. And the counting method decides what the order parameter must hold: a standard counter uses it as a random value, a unique counter fixes it and carries the random value elsewhere, and a sales tag uses it for the order identifier so that a refreshed confirmation page is not a second order; a tag built for one method and pointed at an activity set to another counts wrongly or not at all.
Fix: in Tag Manager, open the live page in preview mode, which the Floodlight guide, citing the Tag Manager Help Centre's page on preview and debug, says shows whether the tag fired on the right trigger and with which values, and use Tag Assistant to read the request leaving the browser with its strings, counting parameters and custom variables before CM360 sees them; then publish the right version to the live environment, point the trigger at the event the site actually sends, and make the strings and the counting parameters match the activity as it is configured.
Verify: the configuration shows the activity's last received tag updating after the test, and a Floodlight report over the test period shows the firing counted once, with the GTMTag management system. version number noted beside it.
Prevent: keep the activity list beside the container with each activity's strings, counting method and trigger named, so a tag is built from the register rather than from memory, and prove every Floodlight tag in preview on the real page before a version goes live rather than reading the change list.
Consent refused or the consent trigger misconfigured
The Floodlight guide states that Floodlight tags deployed through the Google tag respect consent mode, and the consent mode guide explains the mechanism that the consent mode documentation describes: a default command issued before any Google tag loads declares the starting state, an update command changes it when the visitor chooses, and Google's tags check the state before writing storage or sending data for advertising. In the basic implementation nothing is sent until the visitor interacts and nothing at all if they decline; in the advanced one the tags load with denied defaults and send cookieless pings until consent is granted. Two faults follow from that. A site where most visitors decline, or where the banner never reaches them, records only the consented share, which can look like a broken tag when it is a working one with little to count. And a consent trigger built by hand, or a platform template that sets defaults after the Google tag has loaded, fires the tag on one branch only: the guide's commonest mistake is defaults issued after the tag, when the denial arrives too late, and its testing rule is to try all three states, before the choice, after a refusal and after a grant, on every template.
Fix: put the default command on the consent initialisation trigger, or hard-code it above a hard-coded Google tag; confirm the consent platform re-issues the update on every page rather than only where the banner was answered; and give any non-Google tag that depends on the same choice an additional consent requirement rather than a blocking trigger of its own.
Verify: in Tag Assistant, the consent tab shows the defaults issued before the Google tag's own event, the Floodlight tag held back after a refusal, and the tag fired with the granted state after an acceptance, on the enrolment or checkout template as well as the home page.
Prevent: make the three-state test a step whenever a tool, tag or banner setting changes, and read the share of sessions that grant consent beside the activity count so a fall in consent is not misread as a fall in conversions; if a clean first visit shows tags or cookies before any choice, the page on tags firing before consent audits that first page load.
An inherited configuration, or an activity the buying platform cannot see
The CM360 hub places Floodlight at the advertiser rather than in any campaign, which is why its mistakes outlive every campaign and agency that made them, and the tutorial on cleaning up an inherited Floodlight configuration describes what that inheritance usually looks like: activities named only by a number, duplicates that count the same action twice, a purchase in a counter group with no revenue, and activities that look dead while still feeding a bid strategy or an audience list. A buying platform reads only the configuration of the advertiser it is linked to, so an activity created in a second configuration, or under a different CM360 advertiser, counts in the ad server and never appears where the trader is looking. The tutorial's method is to understand before changing: export every group and activity into a register, record when each last received a tag, find what depends on each one in DV360, Search Ads 360 and the audience lists, and treat silence as a question rather than proof of retirement, because the page may have moved or the action may be seasonal.
Fix: confirm which configuration the linked advertiser reads and which one the activity sits in; where they differ, create the activity in the configuration the platform can see and move its readers across one at a time, with bid strategy goals last; where the activity is an inherited one nobody can explain, trace it through the container and the register before anything is renamed or retired, and build a correct activity beside it rather than retyping the group or changing its strings.
Verify: the new activity receives tags and the buying platform shows it as available to line items and strategies, while every report and strategy that read the old one carries on unaffected.
Prevent: keep the register and the change log current, with each activity's purpose, owner and dependents named, so the next person inherits a list rather than a mystery, and give every activity a reader before it is created.
The buying advertiser not linked to the ad server advertiser that owns the configuration
The tutorial on setting up a new advertiser in DV360 and CM360 explains what the link does: it brings the Floodlight configuration across to DV360, so that line items can count and bid towards the same activities the ad server records, audiences can be built from them, and creatives trafficked in the ad server can be synced rather than pasted as tags. The SA360 Floodlight and reporting guide describes the same dependency for search: a client account is linked to the CM360 advertiser that owns the configuration, and from then on every activity is available as a conversion column and a conversion source for bid strategies. Without the link, or with a link to the wrong advertiser, the ad server counts everything and the buying platform shows nothing, for every activity at once, which is the signature that separates this cause from a single broken tag. The tutorial's rule is to check the link from both sides and to fire a test conversion that shows up as an activity before any campaign relies on it; it also asks for two people employed by the business to hold full access to the ad server advertiser, because a link that only a departed agency could see is a link nobody can repair.
Fix: confirm which CM360 advertiser owns the configuration the tags carry, link the DV360 advertiser or the Search Ads 360 client account to that advertiser, and re-select the activities on the line items, strategies and columns that were reading nothing.
Verify: a test conversion fired on the site appears as an activity in the buying platform, and the attributed count for the following days rises from zero as served media is matched against new firings.
Prevent: record the two advertisers, their link and the people who can reach each in the setup record the tutorial ends with, and make the link a line on the pre-flight for every new advertiser, deal or agency handover rather than something the first campaign discovers.
Firings arrive but no media touched that browser inside the windows, or the linker is missing
The Floodlight guide describes the attributionAssigning conversion credit to touchpoints. step: each activity carries a click-through window and a view-through window, the ad server attributes a conversion to a click that happened within the click-through window before the firing, and to an impression within the view-through window where no eligible click exists, which is what makes a view-through conversionA conversion credited to an ad that was shown but not clicked, within a set view window., and a Conversion Linker tag deployed on every page stores the click identifiers that arrive on landing pages so that later firings can be matched in browsers that restrict cookies set by the ad server directly. The guide's own testing table names this row plainly: firings arrive but no conversions are attributed, because no served media reached that browser inside the windows, or the linker is missing. The SA360 Floodlight and reporting guide adds the search version. The join between a conversion and a search click works through the click identifier each engine appends to the landing page address; if the identifier is missing, because auto-tagging is off, a redirect strips parameters or a tracking template does not produce the engine's parameter, the tag still records the activity for CM360 but Search Ads 360 cannot credit it to a click, and the commonest finding is a Floodlight column falling while the engine's own column holds steady. Two further behaviours the same guide documents look like a fault and are not: Floodlight conversions appear after a delay, so the most recent days are always incomplete, and a change to a window or a model applies from the change forward, so a report that spans it compares two definitions. The Campaign Manager 360 Help Centre documents the windows and the tag formats; this page names no default lengths.
Fix: split the Floodlight report by click-through and view-through conversions and compare the firing count with the attributed count; deploy the linker on every page, restore the click identifier on every landing path, and set the windows per activity to cover the journey the plan expects rather than accepting whatever was there.
Verify: a test journey from a served ad to the activity is attributed within the expected delay, and the search column and the engine's column move together again over the following week.
Prevent: add the click identifier and the linker to the release checklist for any change to landing pages, redirects or tracking templates, write each activity's windows into the measurement plan as a decision, and record the expected gap between the ad server's count and each product's attributed share so that a moving gap is read as a finding.
A site release renamed the data layer or moved the confirmation page
The Floodlight guide's last symptom row, counts that vanish after a site release, has the causes the tags, triggers and variables guide explains from the container side. A custom event trigger matches the event key of a data layer push, and a tag's fields are filled from data layer variables at the instant the trigger matches, so a release that renames the event, moves a value to a different key or pushes it after the trigger has fired leaves the trigger unmatched or the tag sending empty values. A confirmation page that moves to a new address breaks any trigger that was conditioned on the old path, and the guide's warning about values scraped from the page applies with full force: a trigger built on a button's class or a value read from page text works until a designer renames the class, and then silently sends nothing. The container was untouched, which is why the release is the last place anyone looks.
Fix: open the new template in preview mode, read the event names and variables the page now pushes against the ones the trigger and tag expect, and either restore the event and key names in the site's data layer, which is the contract the site keeps with the container, or update the trigger and the variables to the new names and the new address in the same version.
Verify: a real journey through the new page fires the tag with the right values in preview, the activity receives tags again in the configuration, and the Floodlight report counts the test firing once under its counting method.
Prevent: list, for every page the measurement plan depends on, the Floodlight activities that listen to it, and make that list part of the release checklist for the page, so whoever ships a new enrolment or checkout template re-tests those activities before the release rather than after the count falls.
Prevention: a release check and a consent-refused test for every activity the plan depends on
Every cause above was a change nobody connected to the activity: a container version, a banner rebuilt, an agency handover, a redirect added, a template shipped. The measurement plan is the document that makes the connection visible, and its row for each Floodlight activity should name the business question it answers, the group and activity strings, the counting method and the windows, the page and the data layer event it listens to, the consent branch it fires on, the configuration and advertiser it belongs to, every product and strategy that reads it, and the person who owns it. A plan with that row is what lets whoever changes the enrolment page, the consent platform or the advertiser link see which activities to re-test before the change ships.
Two tests do most of the protecting. The release check: before any page the plan names goes live, walk the real journey in preview mode and Tag Assistant, confirm the activity's tag fires on the right trigger with the right strings and values, then confirm in the configuration that the activity received it and in a Floodlight report that it was counted once. The consent-refused test: on the same template, repeat the journey before any choice, after a refusal and after a grant, and confirm that the tag is held back on the refused branch and fires on the granted one, as the consent mode guide's three-state test describes; a consent implementation that works only on the home page is the classic way one template breaks a correct setup. Both tests belong on the release checklist for the page, not on the trader's list after launch.
Then watch the count. Compare each activity's daily firings with its recent average and with the business's own record of orders, enrolments or leads for the same days, and read the share of consented sessions beside it, so that a fall in the count can be sorted on the day into a tag that stopped, consent that changed or a business that sold less. Read the ad server's count against each product's attributed share and note the usual gap, because a gap that stays inside its range is a definition and a gap that moves is a finding. When a zero is confirmed, the tracking outage procedure protects the strategies that bid on the signal, annotates the dates and reports the cause, so that the gap is understood rather than forgotten. A line item that spends little while its activity is broken may have a second problem of its own, which the diagnosis of a DV360 line item that is not spending reads from the bid funnel, and a strategy that kept spending on a false zero may have drifted onto inventory the plan never chose, which the page on programmatic spend going to low-quality sites reads from the placement report by spend.
Finally, keep the ownership honest. The register of activities and the change log the cleaning-up tutorial ends with, the setup record the advertiser tutorial ends with, and the measurement plan should name the same owner for each activity, and that owner should be told about releases, banner changes and link changes before they happen. The programmatic troubleshooting hub lists the neighbouring symptoms that a campaign can show once its conversions are counted again.
A language school finds its sign-up activity went quiet after a release (illustrative)
Idlewickshaw Languages
Idlewickshaw Languages, an invented online language school, sees the enrolment Floodlight activity its display line items bid on fall to zero in DV360 on a Tuesday, while the school's own system records 14 to 20 enrolments a day as usual. The analyst opens the Floodlight configuration first: the activity last received a tag 9 days ago, the day a redesigned enrolment page went live, so the missing count is the activity's own and not an attributed share.
Preview mode on the new page shows the data layer event renamed from the name the trigger listens for to a new one the developers chose, so the trigger never matches; the container had not been touched. A second finding appears while testing both consent branches: the school's enrolment subdomain uses a different consent platform from the main site, and the consent trigger had been rebuilt the month before to fire only on the main site's update event, so even with the event restored the tag would fire for enrolments only on the consented branch of one of the two tools.
The developers restore the event name in the next deploy, the analyst points the consent trigger at both platforms' update events, and the three-state test is run on the enrolment template as well as the home page. Over the following 7 days the activity receives 118 tags against 124 enrolments in the school's system, and DV360 attributes 31 of them to its line items. The bid strategy, held at its budget and target with no structural edits from the day the zero was confirmed, resumes normal learning after 2 weeks of stable counts, and the school's release checklist gains a line naming every Floodlight activity that listens to the enrolment page.
Common mistakes
- Debugging the tag when the ad server already counts the activity. The missing figure is an attributed share; test the link, the windows and the delivery instead.
- Testing only the consent-granted branch. The refused branch is where a hand-built consent trigger fails for most real visitors.
- Renaming tag strings to tidy an inherited configuration. Every deployed tag that uses them breaks at once; build the right activity beside the wrong one.
- Assuming the link between advertisers exists because campaigns ran. Delivery does not need the link; conversions do.
- Reading the most recent days as complete. Floodlight conversions arrive after a delay, and a window or model change applies from the change forward.
- Shipping a new confirmation page without re-testing the activities that listen to it. The container was untouched, and the count still fell.
- Letting a bid strategy learn from a false zero. Hold its budget and target steady, apply a data exclusion for the window where the platform offers one, and annotate the dates, as the tracking outage procedure sets out.
