Search the guides

Filtering & analytics

S2S Postback: how a purchase finds its advertising click

Follow the click identifier from the first visit to the conversion message, then diagnose missing, duplicated and delayed events.

Think of a postback as a receipt sent to another system

A customer has paid, but the tracker shows no conversion. Follow the message about that order before changing the campaign. An S2S Postback is a server-to-server notification: the system that knows the order’s status sends that result to the tracker. The customer’s browser does not need to stay open for this later message.

The message must identify the earlier advertising click. A Click ID is the reference assigned to a click so later records can be linked back to it. The sender and receiver must agree on the field carrying that reference and on the meaning of the event status.

Keitaro provides one concrete example: its postback processing uses a click reference and a recognised status. Other integrations may name or handle fields differently. A delivered request is not enough; the receiving system must understand it and attach it to the intended record.

Keep the identifier intact

Event flowFrom click to confirmed outcome
  1. TrackerCreates a click identifier
  2. DestinationStores it with the visit or enquiry
  3. Order systemAssociates it with the outcome
  4. PostbackReturns the event and identifier
  5. ReportMatches the event to the original click

The same click reference must remain traceable at every handoff.

Keep the click reference through each handoff: the landing page, the enquiry and the order record. Store it where the integration expects it. Do not replace it with a newly generated reference halfway through the journey. A valid-looking value is useless if it no longer points to the original click.

In a hypothetical funnel, a form captures the reference, but an import into the sales system omits that field. The order is real and the postback is delivered, yet the tracker cannot find its click. Check where the reference was last present. Repairing the delivery address will not fix a lost data field.

Finding the click and avoiding a duplicate are different jobs

Linking the result to a click answers “Which visit led to this outcome?” This is often called correlation. Think of the click reference as the number written on a claim ticket: returning it lets the receiving system find the earlier record. Similar timing alone is not a dependable replacement.

Deduplication answers another question: “Have we already recorded this event?” It prevents a repeated delivery from creating an extra result. The event’s identity and the receiver’s rules matter here. A click reference alone may not distinguish all cases because one click can lead to more than one valid purchase.

Two approachesFind the click. Then check for a repeat.

Which click?

  • Match the order to its visit.
  • Keep the original click reference.

Already counted?

  • Recognise the same event sent again.
  • Do not reject a different real order.

A correct click reference does not, by itself, prevent duplicate events.

For example, click C leads to order A. The payment message is sent twice after a timeout; the revenue should not double. Later, a separate order B may be a genuine new result from that click. This is an illustrative example. Ask how your receiver distinguishes those cases before choosing its supported identifiers.

A lead, an approval and a payment are different events

Agree what each event means before comparing reports. A submitted form is not necessarily an accepted lead, and neither automatically means paid revenue. If one system calls submission a conversion while another waits for payment, their totals will differ even when the integration works.

Also agree how changes are represented. An order may be refunded or its amount corrected. A later message should update or add a record according to the receiver’s documented rules. Decide this before launch rather than trying to repair inflated totals after several different statuses have arrived.

  • Name the business event and the system that is authoritative for it.
  • Specify the identifier and the supported status values.
  • Define amount, currency and whether revenue is estimated or final.
  • Decide how retries, duplicates, refunds and status updates behave.
  • Choose the reporting date and time zone for reconciliation.
Step by stepFrom enquiry to recorded revenue
  1. Submitted

    The enquiry exists.

  2. Accepted

    The business accepts the lead.

  3. Paid

    Revenue is confirmed.

  4. Adjusted

    A refund or correction changes the result.

Name the state accurately; an enquiry and a payment are not the same result.

Plan for retries without double counting

A sender may retry when it does not receive an answer quickly enough. The receiver might already have saved the first attempt. Idempotency means that repeating the same operation does not create another business result. For a payment notification, a retry must not become an extra payment in the report.

Test both a repeated message and a separate valid event. They are not the same case. Use the receiver’s supported event references and update rules; do not assume that “one click, one conversion” fits every offer. Record the expected outcome of each test before sending it.

Keep a restricted delivery log with the event reference, attempt time, response and retry state. After an apparent success, find the actual event record. Do not expose keys or customer details in shared screenshots. The useful evidence is that the intended event was recorded once and linked correctly.

Worked example: a paid order disappears from the tracker

Take one paid order that is absent from the tracker. The following is a teaching example, not a reported client incident. Keep that order in view throughout the investigation; repeatedly creating fresh clicks makes it harder to understand which records belong together.

Start in the order record. Is the click reference present? If not, look back at the enquiry, then the form handoff. Find the last place where the value existed. A missing import field and a failed network delivery need different fixes, so do not change both at once.

Diagnostic routeFour checkpoints for one missing event
  1. Order record

    Is the click reference present?

  2. Sender log

    Was the correct status sent?

  3. Receiver log

    Was the event accepted or rejected?

  4. Report

    Do date and status filters include it?

Stop at the first missing piece instead of changing every integration setting.

If the reference is present, inspect the sender’s log. Was a notification triggered by the correct status, such as payment rather than form submission? No attempt means you should inspect the trigger or sending queue. A rejected attempt means you should read the receiver’s response and check the fields it expects.

If the receiver accepted the event, locate its individual record before studying the chart. Check the report’s campaign, status, time zone and date basis. A payment recorded later can be grouped differently from the original click. Verify the chosen view before sending another message just because a summary is empty.

Close the case after one controlled event completes the full path and a supported retry leaves the result correct. Write the cause plainly: “the import dropped the click reference” is useful; “postback fixed” is not. Keep the note free of credentials and unnecessary personal data.

Debug in the order information is created

For a new setup, make one labelled test click and keep its reference. Complete the normal page action, inspect the destination record, then trigger the supported test outcome. Follow the outgoing attempt into the receiving log and finally the report. This gives you a known path to reuse after future changes.

Change the component where the evidence stops. No outgoing message points to the sender; a rejection points to the receiver’s requirements; an accepted event missing from one view points to reporting. After the fix, repeat the same test instead of accepting a different successful journey as proof.

A reliable message does not solve every attribution question

S2S describes how an event travels between systems. It does not recover a click reference that was never stored, recognise a person across devices or decide which advertisement deserves credit. That last decision is attribution: the rule used to assign a business result to a marketing interaction.

Server-side delivery also does not remove the need to respect applicable privacy choices. Document the fields being sent and their purpose. Use the supported integration and the minimum necessary data rather than copying a whole customer record into a marketing message.

You can trust the workflow when a test event has a traceable click reference, repeat delivery behaves predictably and results agree with the business system. Build that foundation before using the report to judge traffic quality or move budget between campaigns.

Questions answered

Does S2S mean cookies are never involved?

No. A server-to-server notification describes the delivery of an event. The earlier capture and storage of the click reference may still involve browser state or other mechanisms.

Why can a delivered postback be missing from the report?

Possible causes include an unknown click identifier, an unsupported status, deduplication or a report filtered by another date. Trace one test event through each stage.

Should I resend an event when the dashboard is empty?

First inspect the receiver’s event record and reporting filters. Resending without understanding the original attempt can create a duplicate; use the platform’s supported retry process.

ADSTRAFFIC / ENAll guides →