First, match reported events to order IDs for a few days
Choose three to five recent days. From Events Manager, note the purchase events received and how many came from the browser and how many from the server, and whether they were marked as deduplicated. From the shop, export the orders for the same days with their order numbers, in the same time zone. Then, in a test session, place an order and watch which events arrive in the test events view, with which identifiers, from which route.
Three results are possible. If the platform received more purchase events than the shop took orders, events are being duplicated at source, and the first three rows of the table apply. If received events roughly equal orders but Ads Manager reports more attributed purchases than another tool credits to TikTok, the difference is attribution, and the last two rows apply. If both are true, fix the duplicates first, because attribution comparisons mean nothing on doubled data. When events are missing rather than extra, the TikTok pixel not firing page is the right starting point.
Diagnosis table
Diagnosis
| Symptom | Check | Likely cause | Fix |
|---|---|---|---|
| Browser and server counts together approach twice the orders; little shown as deduplicated | Event identifier on the browser event against the server event for one test order | Pixel and Events API without a shared event identifier | Send the same identifier from both routes |
| Extra purchases cluster on confirmation-page refreshes and order-history visits | Trigger of the purchase tag; repeat events for one order number | Purchase triggered on page view | Fire once on the order-complete event |
| Helper reports the pixel loading twice; every event doubled, including page views | Hard-coded snippet in the theme alongside the tag manager install | Pixel installed twice | Remove one install; keep a single owner |
| Event count matches orders, attributed purchases exceed what analytics credits | Attribution settings; view-through share in reports | View-through or long attribution windows crediting more | Report view-through separately; record the setting |
| TikTok count compared with analytics or another platform, not with the shop | Which systems the comparison uses | Comparing an attributed count with an order count from another system | Compare like with like; reconcile with a written method |
Fix, verify and prevent each cause
Two routes, no common identifier
When the pixel and the Events API (Conversions APIServer-side event sharing with Meta.) both report a purchase, TikTok keeps one copy only if they share the event name and identifier. If the browser sends the order number and the server sends a generated key, or one of them sends nothing, both copies count. The pixel and Events API tutorial holds the setup step for this, which is not repeated here.
Fix: make both routes read the same value, usually the order number, from one source at the moment the order completes.
Verify: a test order shows one browser and one server event with the same identifier, marked as deduplicated, and received purchases fall to roughly the order count.
Prevent: whenever the shop platform, server integration or tag changes, place a test order and inspect the identifiers on both routes.
A purchase that fires whenever a page loads
A purchase tag triggered by visits to the confirmation page counts every refresh, every return to that page from an email link, and every visit to an order history page that shares its address. The tell is repeat events for the same order number, spread over hours or days.
Fix: fire the purchase once, on the order-complete event the shop pushes when payment confirms, as the pixel and Events API tutorial recommends.
Verify: refreshing the confirmation page in a test session sends no further purchase event.
Prevent: ban page-view triggers for purchase events in the container's naming and review rules.
The same pixel loaded twice
Pixels end up installed twice when a hard-coded snippet in the site theme survives a move to Tag Manager, or when a shop platform's integration and a manual tag both load it. Every event, from page view to purchase, then doubles. The pixel helper usually warns about it.
Fix: choose one install route, remove the other, and check that the remaining route sends every planned event, as the tutorial's verification section describes.
Verify: the helper shows one pixel load per page and page views fall to match analytics sessions in proportion.
Prevent: record which route owns the pixel, and check the theme and platform integrations whenever the site changes hands.
Attribution doing what it was told
TikTok credits a conversion to an ad that was viewed or tapped within the attribution windows chosen. A view-through window credits people who saw the ad and bought later without tapping; a long click window credits purchases days after the tap. These are not duplicates: each maps to a real order. The TikTok measurement hub explains the settings and the reports that compare windows.
Fix: decide which windows the team reports, record the choice and the date, and show view-through conversions as a separate line rather than inside one total.
Verify: with view-through shown separately, the click-through count sits within the shop's orders.
Prevent: keep the AttributionAssigning conversion credit to touchpoints. setting fixed across comparison periods and annotate any change.
Two systems counting different things
The order system counts every order. Analytics credits orders to the session that converted, usually by its own last-click rules. TikTok credits orders to its ads within its windows. Comparing TikTok's attributed purchases with analytics' TikTok figure, or with another platform's attributed purchases, always shows a gap, and none of those systems is wrong. The platform versus analytics discrepancies guide sets out the method.
Fix: compare TikTok's click-through purchases with analytics sessions from TikTok, compare received events with shop orders, and agree in writing which system owns which decision.
Verify: each comparison pairs the same unit, and the remaining gap has a named explanation.
Prevent: keep the reconciliation method in the reporting pack so the argument is not repeated each month.
Prevention: one event identifier per order
Every reliable purchase count rests on one rule: an order produces one purchase event, carrying one identifier, sent by each route that reports it. Build the identifier from the order number at the moment payment confirms, pass it to both the pixel and the server, and never trigger a purchase on a page view. Keep a single owner for each event source. After any change to the shop, the container or the server integration, place a test order and check the test events view for one browser event and one server event, deduplicated. Weekly, compare received purchases with shop orders for closed days; a ratio well above one is a duplicate, not a good week. Show return on ad spend (ROASRevenue generated per unit of ad spend.) with view-through credit as a separate line so scaling decisions rest on the part of the figure the business can see. The TikTok sales setup explains why a deduplicated, valued purchase event shapes what delivery learns.
A fashion retailer whose return on spend looked too good
A fashion retailer reports a TikTok return on ad spend of 6.1 for March while the finance team sees little change in revenue. Events Manager shows 1,412 purchase events received that month, 690 from the browser and 722 from the server, of which 118 were marked deduplicated. The shop took 702 orders from all channels. A test order reveals that the browser event uses the order number while the server integration sends its own transaction key, so the copies never match. The same check finds a hard-coded pixel in the theme's checkout template beside the Tag Manager install. The team removes the theme snippet and changes the server integration to send the order number. In the first clean fortnight received purchases fall to 352 against 361 orders, and reported return falls to 2.4, with view-through conversions now shown on their own line.
Common mistakes
Treating every gap as a duplicate, or every gap as attribution. Comparing TikTok with analytics before comparing received events with orders. Leaving a theme snippet behind after moving to Tag Manager. Triggering purchases on the thank-you page view. Changing attribution windows mid-quarter without annotating the reports.
