What happens to attribution when a job is cancelled or refunded?
The revenue has to be reversed in the same place it was credited, against the original lead and the original source. Most reporting pipelines only ever add. If cancellations are never written back, every channel looks better than it is, and the channels producing the most fragile jobs look best of all because they book fastest and cancel later.
One-directional pipelines are the default bug
Most integrations are built to sync on creation. A job is created, a webhook fires, a row lands in the reporting warehouse, and that row is never revisited. Then the job cancels, the invoice is credited, or the estimate is voided, and none of that reaches the report.
The fix is not exotic but it is deliberate: capture updates as well as creates, or run a periodic re-read of any record whose status can still change. Deciding how far back to re-read is a real design choice — far enough to catch late cancellations, not so far that you rewrite last year every night. That decision belongs in the integration design, not in a spreadsheet at month end.
Restate the original period, or adjust the current one?
Both are defensible and they answer different needs. Restating puts the reversal back in the month the lead was created, which gives you the truest picture of that month's marketing performance and means historical numbers change after people have already seen them. Adjusting books the reversal in the current period, which keeps published reports stable and makes any single month's marketing figures slightly wrong.
For evaluating marketing, restate. For anything that has to tie to accounting, adjust. What you cannot do is switch between them without saying so.
Cancellation rate by source is the real finding
- Price-driven demand. Sources that attract comparison shoppers produce more accepted-then-abandoned estimates.
- Financing-dependent jobs. Cancellations cluster where approval was the deciding factor.
- Urgency-driven storm demand. Jobs booked in a panic get re-shopped once the panic passes.
- Oversold appointments. A source can look excellent on booking rate precisely because the call handling promised more than the business delivers, which is a coaching signal, not a media signal.
Push the reversal back to the ad platform too
If you are uploading offline conversions with values, you are training a bidding algorithm. Uploading only the wins teaches it that everything works. Conversion adjustment mechanisms exist specifically to retract or restate a previously uploaded conversion, and using them is what makes value-based bidding reflect the business rather than the booking desk.
This closes the loop that offline conversion integration is supposed to close. A pipeline that only reports good news is not measurement, it is advertising for itself.
Topics: cancellations · refunds · data hygiene · offline conversions
Have a version of this question about your own business?
The useful answer usually depends on which systems you run and how they're connected. That's a conversation, not a blog post.