Refunds and cancelled sales in offline conversion tracking

Why refunds matter for your ads

Offline conversions teach ad platforms which clicks became customers. That only helps if the customers are real. A sale that is refunded a week later is still counted, unless something tells the platform.

For most businesses the refund rate is a few percent, and it does not undo the value of sending sales back. For some it is much higher: subscriptions with trial periods, high-ticket services, online courses. There, counting refunded sales inflates your return on ad spend, and the platform keeps finding more people who buy and then ask for their money back.

DataCops handles this in three ways: it corrects refunds on Google Ads, it never sends stages that did not happen, and it lets you choose a later stage to count. The wider picture is on our offline conversions page.

Google Ads: refunds are corrected

When a refund reaches DataCops, it looks up the original sale by its order ID and sends Google Ads a conversion adjustment:

• Full refund, or an unknown amount: a retraction. The conversion is taken back.
• Partial refund: a restatement. The conversion stays, at the lower value that was kept.

Refunds arrive two ways. From HighLevel, the payments module reports them on its own. From anything else, your store, billing tool or CRM posts the refund to the DataCops conversions API, directly or through Zapier, Make or n8n, with the same order ID as the sale.

Each refund is applied once. If the same refund webhook fires twice, the sale is adjusted a single time. If Google rejects an adjustment, for example because the upload is too recent, a retry worker tries again. Google asks for a gap of a few hours between the conversion and the change, so a refund on the same afternoon is applied a little later.

The result: Smart Bidding learns from the revenue you kept, not the revenue you first booked. See Google Ads offline conversions for the full setup.

Meta and TikTok: what to do instead

Meta and TikTok offer no call to correct a sale once sent. Nothing is sent to them for a refund, and the sale stays counted on the platform. You still have good levers:

• Send a later stage. Count paid, not signed. Or count a stage you only reach once the refund period is over.
• Hold the lead until it qualifies. DataCops can hold a lead and send it only when your CRM marks it qualified. HubSpot is checked every 5 minutes; any other CRM releases it with one API call. Held leads expire after 14 days.
• Read the platform number as a ceiling. Compare it with net revenue in your monthly report.

The trade-off is speed. A later stage means fewer events, each arriving later, so the platform learns a bit slower. Keep the platforms' own time limits in mind: Meta rejects events sent more than 7 days after they happened, and Google Ads ignores uploads more than 90 days after the click. For businesses with a heavy refund rate, a later stage is usually worth it.

What is never sent

From HighLevel, the stages sent are lead, booked, showed, won with value, and paid. Cancelled, no-show and lost are never sent. A cancelled booking or a lost deal is never turned into a conversion, so you are not counting things that did not happen.

From other CRMs, you decide which events to post to your private webhook. Post only the stages you trust. Every post is logged in plain words: sent, not sent, duplicate, ignored or rejected.

One sale, one conversion

Double counting looks a lot like a refund problem. The platform shows more revenue than you made, and it is hard to tell why. Often the same order was reported by more than one tool.

DataCops keeps a sale ledger. A purchase with an order ID is sent once per ad platform, and remembered for 400 days, no matter whether Shopify, a webhook and the pixel all report it. A lead with an email is sent once per platform within 30 minutes. Duplicates are logged as duplicate_sale or duplicate_lead, with the first message ID, so you can see exactly what was held back. If a send fails, the claim is released so a retry can go.

The conversions API is also safe to call twice: the same messageId counts once. And event IDs are passed to Meta so the browser pixel and server events dedupe.

Check what each platform received

The DataCops delivery report shows, per platform, how many conversions were sent, skipped or failed, with a plain-English reason and the value sent, for up to 400 days. When finance asks why a number looks high, you can show exactly what went out and when.

Sale refunds versus click fraud refunds

These two are easy to mix up. A sale refund is money you gave back to a customer, and DataCops corrects the conversion. A click fraud refund is money Google may give back to you for invalid clicks.

For the second, DataCops has a refund evidence export on the Organization plan. It lists every suspicious Google Ads click from the last 60 days with the reason (bot, automated browser, datacenter IP, VPN, proxy, Tor or repeat clicks), the IP and the click ID, as a CSV for Google's Click Quality Form. You file it yourself, and Google decides. The Organization plan can also block bad traffic at the Cloudflare edge before it reaches your site.

Judge platform revenue against net revenue

Ad platform reports count what they were told. Your accounts are net of refunds. Put both side by side, and never treat one as a mistake in the other.

A useful habit: keep one line in your monthly report called refunded revenue, taken from your accounts, and show platform revenue minus that figure next to the platform number. On Google Ads the gap shrinks on its own, because refunds are corrected. On Meta and TikTok it stays, and everyone can see where it comes from.

A made-up example

This is an example, not a real result. Say you sent 100 won sales at 500 each, so each platform shows 50,000. Later, 12 are refunded in full. Your accounts show 88 sales, or 44,000 net.

With DataCops, Google Ads gets 12 retractions and shows 44,000. Meta still shows 50,000, and its return on ad spend looks about 13.6 percent higher than the real one. Knowing that, you read the Meta number as a ceiling and plan budget against the net figure.

A simple refund policy for reporting

Write this down once so nobody argues about it monthly.

• Define the counted stage. Decide which stage is a sale for reporting, ideally one reached after the refund window.
• Send refunds with the order ID. So Google Ads is corrected automatically.
• Report both figures. Platform revenue and net revenue, side by side, with the difference labelled.
• Review the refund rate quarterly. If it climbs, move the counted stage later.

How to talk to finance

Finance and marketing often fight over which revenue is right. Both are, for different jobs. Marketing needs a signal fast enough to steer ads. Finance needs money that will not be reversed.

Say it in one sentence: the ad platforms tell us which clicks became buyers, Google Ads is corrected for refunds, our accounts tell us what we kept, and we show both. Agree the counted stage together. That turns an argument into a rule.

Good to know before you start

• Refunds are matched by order ID, so send an order ID with every sale.
• Google Ads applies a correction a few hours after the original upload, and DataCops retries until it lands.
• The refund evidence export for click fraud is part of the Organization plan ($359 a month, $299 yearly).
• The Free plan includes 2,000 sessions. See pricing.