Why Meta's results and your orders never match
The attribution settings guide explains in one section why Meta and other systems disagree, and the causes it lists are the backbone of this tutorial. Meta's documentation on attribution settings describes the setting as a window after a click or a view in which a conversion can be credited to an ad. A purchase made outside the window is not credited, and a purchase made inside it is credited even if the person was going to buy anyway. Meta reports conversions against the date of the ad interaction, not the date of the purchase, so the most recent days of any period keep growing while the window is open. It matches people across devices through their accounts, which the shop's own source records cannot do. And where it cannot observe a conversion directly, it models part of the total, which the attribution guide asks readers to treat as an estimate and not to slice finely.
The business side has causes of its own. Orders are cancelled and refunded after the purchase event has fired. Test and staff orders reach the live site. Some orders never pass through the website at all, such as phone orders, and are only visible to Meta if they are sent as events. And every other channel credits itself for the same buyers, which the page on channel conversions exceeding sales covers when it adds every platform's claims together.
| Cause | Effect on Meta against orders | Where it is handled |
|---|---|---|
| Events missing or doubled before attribution | Either way; a fault | Steps two and three |
| View-through credit | Meta higher | Step four |
| Reporting by interaction date, open window | Meta lower in recent days | Step one |
| Cross-device matching and modelled conversions | Meta higher than the shop's tagged orders | Step seven |
| Cancellations and refunds | Meta higher | Steps five and seven |
| Orders Meta never received, such as phone orders | Orders higher | Step seven |
| Returning customers credited like new ones | Meta's figure flatters acquisition | Step six |
The guide to why platforms and analytics disagree reconciles a platform against analytics, and the Google Ads tutorial on reconciling with sales records does this job for Google Ads, matching on transaction identifiers. Meta's reporting is aggregated; the attribution guide notes that some reporting is aggregated so that individual-level data is not exposed, so a Meta reconciliation works on totals, rates and a ratio rather than on a list of matched orders.
Before you start
- Reporting access to the ad account and access to Events Manager for the dataset the campaigns use.
- An export of orders or leads from the shop platform or CRM, with order date and time, value, status (completed, cancelled, refunded), customer status (new or returning) and, where the shop records it, the source of the order.
- The account's attribution setting, any changes to it in the period, and the compare attribution settings columns in Ads Manager.
- The measurement plan, so that the purchase or lead event's definition and value are written down; the measurement plan framework covers what it should hold.
- Last month's reconciliation, if there is one, so the ratio can be read as a trend.
- A spreadsheet with one row per line of the reconciliation, kept from month to month.
The steps
Choose a closed period
Reconcile a calendar month that ended long enough ago for the attribution window to have closed on the last interactions in it. Because Meta credits conversions to the date of the ad interaction, the month's figure keeps rising until then, and a reconciliation run in the first days of the next month compares a growing number with a settled one. The same reporting rule shifts the edges: purchases early in the month from interactions in the previous month sit in last month's Meta figure but this month's orders, and the reverse happens at the end. Over a normal month the edges roughly cancel; over a month with a sale at one end they may not, so note it. Use the same time zone for both exports, since a day in Ads Manager follows the ad account's time zone.
Compare events received with orders before attribution
Before looking at attributed results, check that Meta received the purchases at all. In Events Manager, read the purchase event's received count for each week of the month, by connection method, and put it beside the shop's website order count for the same weeks. The Conversions API guide, citing Meta's documentation on the Conversions API, notes that server purchase events over a week should be close to the shop's order count before attribution is applied, and that a large gap means events are missing or doubled. For lead generation, compare lead events received with the CRM's new lead records. A week where received events fell well below orders is a measurement break, and the attributed figures for that week cannot be reconciled until it is explained.
Check deduplication and test orders
Where the pixel and the Conversions API both send purchases, Meta treats a browser and a server event with the same event name and event ID as one. Events Manager shows the deduplicated share in the event's detail; a share that dropped in one week, often after a site release, means some purchases were counted twice. The diagnosis page for duplicate conversions traces the ID, name, timing or sender at fault. Then look for test and staff orders on the live site, which the shop can flag and Meta cannot, and for purchase events that fire on a page reload. Write each fault as its own line with an estimate of the purchases and value it added, and fix it before the next month.
Export results under your setting and under click-only
In Ads Manager, export results and purchase conversion value for the month for every campaign, under the account's attribution setting. Then add the compare attribution settings columns, which the attribution guide describes as showing results under other windows without changing delivery, and read the click-only figures beside them. The gap between the two is view-throughA conversion credited to an ad that was shown but not clicked, within a set view window. credit. Split the totals into prospecting and retargeting, because the attribution guide notes that view credit concentrates where audiences were already close to buying. Keep the account's default setting as the first line of the reconciliation, since that is the figure the team sees every day, and record which setting it was.
Pull the matching orders or leads
From the shop or CRM, export every order or lead in the same period and time zone. Separate completed orders from those cancelled or refunded, and compute the cancellation and refund rate both for all orders and, where the shop records the source, for orders that arrived from a Meta ad click. Check that order value is measured the same way as the purchase event's value parameter, for example with or without tax and delivery, or the value lines will never agree. For lead generation, export the leads with the stage each reached, because a lead counted by Meta and never qualified is part of what the business needs to see.
Split new and returning customers
Mark each order as from a new or returning customer using the order system, which the guide to new and returning customers explains is where customer status belongs. Compare the share of returning customers among orders that arrived from Meta clicks with the share among all orders. Where the campaign type offers it, Ads Manager's audience segment breakdown is a second view, and the page on Advantage+ audience reaching existing customers explains how to read the two together. A Meta figure that reconciles well on purchases can still be mostly repeat buyers, and cost per new customer from the shop's data is the number acquisition should be judged on.
Set aside what each side cannot see
Some of the gap is structural and should be named, not fixed. Meta can see what the shop cannot: purchases by people who saw or clicked an ad on one device and bought on another, matched through their accounts, and conversions it modelled where identifiers were missing, which the attribution guide describes as estimates. The shop can see what Meta cannot: orders that never passed through the website, orders from people who declined consent, for whom neither the pixel nor the server should send an event, and orders that came after the window closed. Write each as a line with its size where you can measure it, such as phone orders, and as a note where you cannot, such as modelled conversionsConversions a platform estimates statistically for activity it could not observe directly, such as unconsented or cross-device journeys.. Apply the cancellation and refund rate from step five to the click-only figure, because the purchase event fired before the order was cancelled.
Calculate the ratio and its trend
Walk from Meta's reported figure to the reconciled one: the account's setting, less view-through credit, less the faults found, less cancellations and refunds. The result is the Meta figure the business accepts. Divide it by the reported figure, for purchases and for value.
Reconciliation ratio = Reconciled Meta value ÷ Meta-reported value(ratio)
As a cross-check, the shop's own count of orders that arrived from a Meta click is a floor: it misses cross-device purchasesA journey in which the ad interaction and the conversion happen on different devices, so that only a signed-in identity or modelling can connect them. and anything the browser did not tag, so the reconciled figure should sit above it. Put the month's ratios beside the previous months'. A stable ratio is structure; a jump is a fault, a change of setting or a new campaign type, which the change history usually explains.
Set targets on the reconciled number
The optimisation hub argues that targets should be set on the number the business trusts, with a known relationship to the platform's number. The ratio is that relationship.
Meta target ROAS = Required business ROAS ÷ Value reconciliation ratio(ratio)
For cost per result, multiply the business's required cost per purchase by the purchase ratio to get the figure Meta should report. The ROAS calculator and the CPA calculator do the arithmetic. If the account uses a cost per result goal or a ROAS goal, remember that the learning phase guide lists changes to a cost goal among the edits that restart learning, so change it once, deliberately, rather than every month.
Record the result and repeat monthly
Write one row per month: the period, the attribution setting, received events and orders, each line of the walk, the reconciled figures, the ratios, the new-customer share, and any fault found with the date it was fixed. The monthly paid media review checklist puts the reconciliation on the agenda before results are discussed. Share the sheet with finance rather than only its conclusion; the trace is what turns a disputed number into an agreed one. When a ratio moves, record the cause before adjusting any target.
A month reconciled for a workwear shop (illustrative)
Rivetgull Workwear reconciles a closed month
Rivetgull Workwear, an invented workwear retailer, reconciles a month that closed long enough ago for its attribution window to have finished. Meta spent £15,600. The shop recorded 2,380 orders, of which 150 were trade orders taken by phone and never sent to Meta. Events Manager received 2,214 server purchase events against 2,230 website orders, close enough in every week. In the third week the deduplicated share fell after a theme update dropped the event ID from the browser tag, and the developer restored it.
| Line | Purchases | Value |
|---|---|---|
| Meta-reported, account's default setting | 760 | £68,400 |
| Less view-through credit (click-only compare column) | −170 | −£15,300 |
| Less purchases counted twice in the week deduplication failed | −38 | −£3,400 |
| Less cancellations and refunds at the rate for Meta-click orders | −22 | −£4,000 |
| Reconciled Meta purchases | 530 | £45,700 |
| Shop orders carrying a Meta click source (floor) | 455 | £39,200 |
The value ratio is £45,700 ÷ £68,400, about 0.67, against 0.68, 0.66 and 0.69 in the previous three months, so the gap is structural apart from the duplicate line, which is a fault now fixed. Reported ROAS was about 4.4; on the reconciled value it was about 2.9, below the 3.2 the business needs from its margin. Of the 455 orders with a Meta click source, 61% came from new customers, against 39% across all orders, so the team reports cost per new customer as a range between £48, using the reconciled purchases, and £56, using the floor. Finance adopts the 0.67 ratio, the reporting target for Meta becomes a ROAS of about 4.8 on the default setting, and the shortfall goes to the monthly review as a question about retargeting's share of the spend.
Common mistakes
- Reconciling an open month. Meta's figure is still growing; the gap looks smaller than it is.
- Skipping the events-against-orders check. A measurement break then hides inside an attribution argument.
- Ignoring deduplication. One week of failed deduplication inflates the month and every target set on it.
- Mixing attribution settings between months. A ratio means nothing if the setting under it changed.
- Comparing gross purchase value with net revenue. The purchase event fires before cancellations and refunds; add the line.
- Treating every difference as an error. View-through, cross-device and modelled credit are structural; size them and move on.
- Judging acquisition on all purchases. Split new and returning customers in the shop's data.
- Setting targets on Meta's figure alone. A target met in Ads Manager can miss the return the business needs.
- One reconciliation, never repeated. The trend in the ratio is the useful part.
