Skip to content
All field notes
Measurement
Published
Nov 6, 2026
Reading time
14 min read

CM360 Offline Conversions Accepted but Missing From Reports?

Trace missing CM360 offline conversions through the actual upload route, identifiers, processing diagnostics, deduplication and attributed reporting.

AP
AdProof Research
Independent measurement desk
A server pixel-GET path ends at a request receipt, while a separate offline-upload sheet passes through row diagnostics toward a blank report.
A pixel request and a structured offline upload need different diagnostic evidence. Conceptual illustration without account or attribution results.
In this guide (12 sections)

If an offline purchase is missing from CM360 or DV360, first identify the interface that actually received it. A server requesting an image-tag URL is not evidence of a completed offline-import workflow. A successful API response is also different from finished processing and advertising credit. Preserve the request, its destination and its receiving evidence before changing tags or resending orders.

For new API integrations, Google's current developer guidance recommends the Data Manager API rather than starting a new Campaign Manager 360 API workflow. Existing CM360 integrations still need their own response parser and reporting checks. This guide explains both paths without treating their identifiers, deduplication rules or diagnostic responses as interchangeable.

Two original questions reveal the missing checks#

An older affiliate-tracking question describes capturing a dclid, receiving a purchase notification and requesting a Floodlight image endpoint from a server. The author reports missing DV360 conversions and asks whether a browser tag is indispensable. Replies point toward an offline API and checking the identifier, but do not verify a working implementation.

A separate CM360 API question reports HTTP 200 and hasFailures: false, while conversions and Floodlight impressions remain absent. The visible exchange does not establish the cause. These are verified historical questions, not evidence of a new platform outage. Their common task is establishing which transition is supported by evidence.

Related guide

Check the count behind the receiving report.

Separate a collected event, the applicable Floodlight counting rule and attributed conversion credit.

Review Floodlight counting

Opens a practical guide. No email required.

1. Name the actual upload route#

Ask for the endpoint, API version or connector, authentication context and exact response body. Capture what the production job sends, rather than a simplified example copied into an API console. Keep secrets and raw advertising identifiers in restricted operational records.

Observed interfaceEvidence needed nextAvoid this assumption
Server requests an image-tag URLDocumented integration contract and the intended receiving recordAn HTTP response proves an offline import
CM360 conversions.batchinsertPer-conversion status plus the relevant receiving reportHTTP 200 proves every record succeeded
Data Manager API events.ingestReturned request reference and asynchronous destination diagnosticsAcceptance means processing is complete
A product screen or connector called Data managerIts actual downstream interface and documented behaviorSimilar labels imply the same API schema

Google's CM360 overview documents supported identifier-based offline uploads and explicitly directs new implementations toward Data Manager API. It does not establish that an arbitrary server GET to an image tag is equivalent. For a legacy route, retain the necessary profile permissions and actual target activity.

There is also a naming trap. The Search Ads 360 upload help page discusses a point-and-click Data manager workflow and its backend constraints. Do not apply that screen's description to every request made to the separate developer API. Resolve the actual interface before interpreting apparently conflicting instructions.

2. Freeze the destination and the business action#

Define one outcome: a completed purchase, a qualified lead or another agreed event. Record whether the source is an accepted business record or only an intermediate notification. A checkout-start message is not a completed sale simply because an integration labels it conversion.

Build a destination inventory with configuration ID, activity ID, account ownership, environment, integration version and effective dates. Similar activity names can conceal different destinations. Compare the actual serialized values between a working test and the affected job, including string formatting and environment-specific configuration.

Google's Floodlight offline-conversion workflow separates destination preparation, event preparation, ingestion and diagnostics. Use those stages as review gates. A correctly authenticated request to the wrong legitimate activity can be operationally successful while answering the wrong reporting question.

For counting-policy questions, continue with the Floodlight counting-method guide. Its tag-counting definitions do not replace an audit of the import route or destination.

3. Check identifier lineage before adding another identifier#

Record the identifier type and how it reached the business event. For example: captured at an authorized redirect, persisted in an approved store, associated with an order under a documented rule, then serialized into the intended API field. A field's presence alone does not validate that lineage.

The CM360 overview makes an important qualification: an uploaded dclid or gclid does not force credit to the specific click encoded in that value. The conversion is associated with the user behind the click, and another eligible interaction can receive credit. Do not promise deterministic affiliate-click attribution from successful identifier submission.

Preserve original values in the approved system and compare restricted fingerprints or internal references in the troubleshooting packet. Do not invent missing click IDs, reuse an unrelated identifier or copy personal information into public screenshots. Different identity paths have different prerequisites; a first-party matchId path should not be substituted merely because a click-ID field is empty.

4. Interpret the response for the interface you have#

The CM360 upload guide returns a status for each submitted conversion. Inspect that collection, retain the association with your input rows and examine hasFailures. For its documented verification workflow, the Floodlight Impressions report includes attributed and unattributed outcomes, which makes it more useful than an attributed campaign total for an initial receiving check. Google generally describes availability within 24 hours; that is guidance, not a deadline proving all historical rows arrived.

For Data Manager API, the ingestion reference defines validateOnly. A validation-only request does not execute an upload. A passing validation is useful evidence about that request, not a recorded purchase. In REST JSON, preserve the returned requestId; documentation and client libraries may present the equivalent field as request_id.

The diagnostics workflow requires a separate status check after acceptance. Inspect each destination until it reaches SUCCESS, PARTIAL_SUCCESS or FAILED; PROCESSING remains unresolved. Check warnings even for success. Its received-record count includes records that can later fail and is not an attributed-conversion count. Validation-only requests do not provide this production-processing evidence.

5. Keep fast-fail validation separate from later processing#

Google's upgrade comparison contrasts CM360's partial-failure response with the Data Manager API's fast-fail model. Do not read fast-fail as a promise that all accepted events will later receive campaign credit. Request validation, asynchronous processing and attribution are separate boundaries.

A useful operational log preserves the ingestion response and a later diagnostic snapshot under the same request reference. Record when each was retrieved. If the integration logs only its initial success badge, improve evidence capture before concluding that the platform lost a conversion.

Use a bounded inspection policy suited to the actual API. Save the last state and next check rather than polling indefinitely or calling every still-processing request a failure. Where diagnostic totals cannot identify individual business rows, retain that limitation instead of manufacturing an event-level join.

6. Preserve the event's identity across retries#

The CM360 FAQ explains that HTTP 200 can coexist with row errors. It distinguishes retryable transient or not-yet-found cases from invalid data and permission failures. Its safe-retry guidance preserves the original ordinal, event timestamp and identity fields. Its stated age limits also belong to that interface; do not move an old purchase's timestamp forward to make it eligible.

For Data Manager API migration, consult the field mapping. CM360's ordinal has no equivalent field there. A stable event-specific transactionId is central to the documented destination deduplication model. Do not copy a constant legacy ordinal into the transaction-ID field for all purchases.

Keep one internal event reference linked to its original attempt, retry reason and subsequent outcome. A second attempt is not a second business purchase. Conversely, two genuinely distinct purchases need distinct event identities. A deduplication rule cannot repair a selector that assigned the same identity to unrelated orders.

7. Align the receiving report before investigating attribution#

Save the report type, activities, date range, timezone, exact metrics and extraction time. Check whether the selected view represents event receipt, attributed conversions, quantities or monetary value. An upload-time day is not automatically the event-time day.

For the CM360 API, inspect quantity and value with their intended units. The Conversion resource defines the event timestamp in epoch microseconds and interprets value in the owning advertiser's currency. A connector sending seconds where microseconds are expected has not preserved the business event's time. Record the serialized field and the human-readable conversion together in the restricted worksheet. The upload guide also specifies a quantity of at least one for inclusion in certain measures, including Total Conversions. A zero-quantity record and an absent record are different investigations.

After receiving evidence exists, examine attribution under the relevant campaign and activity definitions. Keep inventory-specific support and conversion windows in scope. Do not resend already processed purchases simply because an attributed campaign report is smaller. Also do not label every difference harmless: preserve an unexplained residual and give it an owner.

8. If records disappear after an edit, audit the update contract#

An unanswered historical question about conversion updates reports a successful batchupdate response followed by a missing conversion in a report. The author also describes using that method for creation. The original does not prove deletion, deduplication or a platform defect. Its useful next step is a controlled comparison of the existing record, the update and the receiving report.

Google's current CM360 editing guide describes editing an existing conversion, not a general create-or-update operation. The identifying fields must match the existing event: its configuration, activity, identifier, ordinal and timestamp. encryptedUserIdCandidates cannot be used for edits. Retain the identity returned by a successful insert; after an edit, keep the returned identity for subsequent edits.

Treat that identity as an operational reference, not a label to reconstruct from memory. A matching order ID in your CRM does not establish that every platform field is identical. Before another attempt, compare the restricted original response and serialized update field by field. Preserve data types, timestamp precision and destination. Do not generate a new timestamp or ordinal merely because the record is being changed today.

The editing guide requires both quantity and value, and supplied values replace the existing values. Optional customVariables are updated or inserted when supplied and remain unchanged when omitted. Other unsupported fields cannot be modified through this workflow. Those rules are specific to CM360's editing contract; the recommended Data Manager route for new imports does not establish identical editing behavior.

A hypothetical replacement-value check

Assume two invented accepted purchase records, one currency of USD and one unchanged reporting definition. Order A originally has quantity 1 and value 120; Order B has quantity 1 and value 80. An authorized business correction reduces A's recognized event value to 90 while preserving its quantity. This is a teaching scenario, not a customer refund or an executed API request.

Field or totalBeforeIntended after correction
A: quantity11
A: value, USD12090
B: quantity11
B: value, USD8080
Combined quantity22
Combined value, USD200170

The intended replacement value for A is 90, not the difference of −30. The combined value changes by 30 / 200 = 15%, while combined quantity remains two. A monetary decrease therefore does not imply a missing purchase record. A connector must not fill an unchanged required quantity with zero because only the value changed.

Also keep row presence separate from metric inclusion. As noted above, a zero quantity can affect inclusion in some conversion measures. A record absent from one filtered or attributed report is not proof that its underlying event was deleted. This does not establish that zero quantity caused the historical question: its visible sample explicitly supplies quantity 1.

Verify the correction with three saved states

Before an authorized edit, save the existing identity and values, the business reason for the change and a reproducible report snapshot. Keep the candidate payload separate from the approved payload. This makes an incorrect default, delta value or wrong destination visible before it becomes a reporting dispute.

After the edit, preserve the response for each requested record and any errors. Re-run the receiving check under the same activity and event-time scope, then inspect the attributed report separately. Record the new extraction time and any other settings changes. A smaller report can follow the intended value correction, a metric filter or another unresolved condition; the response alone does not choose the explanation.

If the result is still unexplained, provide support with the original identity, before/after payloads, per-record responses, request times and both report definitions. Keep raw identifiers private. Do not restore the old state by blindly replaying a creation request, and do not keep applying compensating values until an aggregate total looks right. Preserve the unresolved event and give the next check a named owner.

A hypothetical acceptance worksheet#

Hypothetical example only: a fashion-retail operation reviews one purchase event, one destination activity, one currency and a fixed event-time interval. It has 50 accepted business orders. Its selector deliberately holds four for review and sends the remaining 46 distinct records.

CheckpointTeaching countWhat the evidence establishes
Accepted business orders50The defined business population
Selected and attempted records46Four were held before the destination received a request
Processing failures established by diagnostics3These records need their specific corrective action
Records processed without errors43Processing evidence exists for this population
Attributed report credits27The result of the saved attribution/report contract

The pre-upload hold is 4 / 50 = 8% of business orders. The processing failure share is 3 / 46 ≈ 6.52% of attempted records, using a different denominator. Do not add those percentages.

The numerical difference between 43 processed records and 27 report credits is 16. It is a comparison residual, not proof of sixteen failed uploads or sixteen organic orders. The report might apply different attribution and counting units. Investigate its contract and available linkage before assigning a cause.

This example deliberately separates stages. It does not claim that a real report exposes a transaction-level mapping for every credit or that AdProof has executed these uploads.

Send a packet that the next owner can use#

Before escalating, preserve:

  1. One action definition and the source selection rule.
  2. Actual interface, version, destination and permission scope.
  3. Restricted input references and serialized time/quantity/value definitions.
  4. Request response, per-row statuses or asynchronous request diagnostics, as applicable.
  5. Stable event identities and a bounded retry history.
  6. Receiving report and attributed report, with their distinct scopes.
  7. The first unsupported transition and the next concrete check.

For a scoped delivery comparison, ask AdProof to review the campaign-evidence boundary. Bring the platform import diagnostics and business-event ledger from their owning systems alongside the campaign, dates, timezone and delivery exports. AdProof's supported campaign-event and aggregate export workflow can support the delivery side of a reconciliation; it does not supply a CM360 upload integration or determine Google's attribution decisions. The anonymized fashion-retail delivery case illustrates that bounded measurement workflow.

Frequently asked questions#

Is server-side Floodlight conversion measurement impossible without a browser purchase tag?

No. Google documents offline conversion-upload paths. Their destination, identity and permission requirements still apply. A server requesting an image-tag URL does not by itself demonstrate that those requirements were met or that the receiving system processed an offline record.

Does hasFailures false prove the conversion received campaign credit?

It describes the CM360 API batch response. Preserve the returned statuses, verify receiving evidence and then inspect the report's attribution contract. An import result does not promise credit to the campaign or specific affiliate click you expected.

Should I use the same parser for CM360 and Data Manager API responses?

No. Their request schemas, response structures and processing checks differ. Retain per-conversion results for the CM360 route and request-linked asynchronous diagnostics for Data Manager API. Validation-only success is not an executed upload.

Should I resend every purchase absent from the attributed report?

First separate held, rejected, pending and already processed events. Correct the actual error and preserve event identity on an eligible retry. Inspect attribution and report scope for processed records rather than generating fresh identities to force another count.

Can batchupdate replace a creation request or accept a value delta?

The documented CM360 workflow edits an existing conversion identified by its exact fields. Required quantity and value replace the old values; they are not described as adjustments to add. Preserve the existing identity and carry forward any required value that should remain unchanged. Do not infer a general upsert or transfer this behavior to another API.

Your next step

Check the count behind the receiving report.

Separate a collected event, the applicable Floodlight counting rule and attributed conversion credit.

Review Floodlight counting

Opens a practical guide. No email required.

Back to all field notes