Why platform conversions and business records differ
Google documents conversion tracking as a record of what people did after interacting with an ad, counted under the settings on each conversion action. The measurement hub explains the rules that matter here: each action has its own counting, value, window and attribution model; the standard columns place a conversion on the date of the click, not the date of the sale; and a conversion outside the window, or one that cannot be joined to an ad interaction, is not counted, even if it happened.
The business system follows different rules. It records every order or lead when it arrives, whatever brought the customer, and it knows things the platform cannot: which orders were paid, refunded or cancelled, which leads were duplicates, which became customers. So the platform counts some things the business cannot attribute, the business counts some things the platform never sees, and both count some things differently.
| Difference | Platform side | Business side |
|---|---|---|
| View-through and engaged-view conversions | Counted for display and video, in their own columns | No click, so no order can be tied to an ad |
| Modelled and cross-device conversions | Estimated or matched by Google | An order with no ad identifier on it |
| Returns, cancellations, failed payments | Value recorded at the time of conversion | Netted off later |
| Duplicates | Possible where a tag fires twice or two actions record one event | Removed when orders are de-duplicated |
| Dates | Placed on the click date in standard columns | Placed on the order date |
| Windows | Conversions after the window are not counted | Counted whenever they arrive |
None of these is an error on its own. The reconciliation's job is to size each one, so that what is left over is small and stable, and anything that moves can be investigated. The academy exercise on reconciling conversions with sales records for a piano retailer practises sizing these differences for a closed month of conversions against the retailer's orders and showroom bookings.
Before you start
- Read access to the Google Ads account, its conversion settings and its change history.
- An export of orders or leads from the order system or CRM for the same period, with order or lead identifiers, dates, values and status, including refunds and cancellations.
- Any ad identifiers the business stores: a transaction ID the purchase tag passes, or a click identifier saved with each lead, as the offline conversion import entry describes.
- The primary conversion actions and their windows, from the action register if one exists; the conversion action clean-up tutorial builds one.
- A sheet with one row per difference, kept from month to month.
The steps
Choose a closed period
Pick a calendar month that ended long enough ago for the longest click window on the primary actions to have closed, so the Google Ads figure has stopped growing. Lead accounts with long windows reconcile a month that ended some time ago; shops with short windows can reconcile sooner. Avoid a month with a large settings change in it, such as a new window or attribution model, unless explaining that change is the point. Use the same period length each month so that the ratio is comparable over time.
Export conversions by conversion time
The standard Conversions and value columns place each conversion on the date of its click, which the hub notes is the default. Business systems date orders by when they happened, so export the conversion-time versions of the columns for the primary actions, for the period, by day. Export the click-through, view-through and engaged-view figures separately, and the All conversions column for the same actions, so that secondary actions recording the same event are visible. Record the account's settings for those actions at the time, because a reconciliation is only comparable with others made under the same settings.
Pull the matching orders or leads
From the business system, export every order or lead for the same days: identifier, date, value, status and any stored ad identifier or source field. For a shop, that is orders with payment status and refunds; for a lead business, leads with their stage and outcome. Remove the business's own test orders and internal entries now, and note how many there were, since some of them may also have fired the conversion tag.
Match on identifiers where they exist
Where the purchase tag passes a transaction ID and the order system stores the same ID, or where the CRM stores the click identifier with each lead, count the orders or leads that carry an ad identifier: these are the business's view of conversions the platform could match by click. Compare that count and value with the platform's click-through figures. Where no identifier exists on the business side, match at the level of daily totals instead, and note that the reconciliation is coarser. The setup tutorial uses the same transaction ID to stop a reloaded confirmation page counting twice, so its absence is also a reason to expect duplicates in a later step. Where qualified leads or won deals are sent back by file, the upload log kept by the offline conversion upload SOP shows which outcomes were sent and which were rejected, so the month's gap can be traced upload by upload.
Set aside conversions the business system cannot see
Subtract from the platform figure the conversions that no business record could be tied to an ad. View-through and engaged-view conversions come first: there was no click, so no order carries one; the view-through conversion entry explains the kind of credit involved. Then cross-device conversions and conversions matched through enhanced conversions, which Google documents as using hashed first-party data to recover matches the click identifier lost. Then modelled conversions, estimated for activity consent or browsers prevented the platform from observing. These may be real sales; they are set aside because the business system cannot confirm their source, not because they are false.
Account for returns, cancellations and duplicates
The platform records value at the moment of conversion, before returns, cancellations and chargebacks, as the conversion value entry notes. Net the business figure for refunds and cancellations on the matched orders, and record the difference as its own line. Then look for platform conversions with no completed order behind them: a tag firing on a reload, on a failed payment, or on a page reached without completing the purchase. For leads, de-duplicate repeat submissions and junk; the deduplication entry covers how double counting arises. Each fault found here is a tracking fix, not just a reconciling line.
Allow for windows and lag
Some orders in the period followed clicks from before it, and some clicks in the period will produce orders after it. Exporting by conversion time aligns the dates, but the windows still decide which orders the platform can credit at all: an order that came longer after the click than the window allows is in the business system and absent from the platform. Use the account's Conversion LagThe time between an ad interaction and the conversion it leads to, which leaves the most recent reporting days incomplete. to estimate how many orders fell outside the window, and record it as a line. The conversion windows and attribution guide covers setting windows from that lag.
Calculate the ratio and its trend
What remains is the figure the business can confirm. Divide it by the figure the platform reported.
Reconciliation ratio = Business-confirmed value ÷ Platform-reported value(ratio)
Calculate it for conversions and for value, and put this month's ratios beside the previous months'. A stable ratio is structure: it describes how this platform counts for this business. A ratio that jumps is a fault, a settings change or a new campaign type, and the change history usually shows which. The residual, the part of the gap no line explains, should be small and stable; if it is not, a cause is missing.
Set targets on reconciled numbers
Targets in Google Ads are set on the platform's figures, but the business's required return is set on its own. The ratio converts one into the other.
Platform target ROAS = Required business ROAS ÷ Reconciliation ratio(ratio)
The same logic applies to a target CPA, dividing by the conversion ratio instead. A platform return that looks comfortably above target can hide a business return below it, which the high ROAS but not profitable diagnosis traces from the finance side. The ROAS calculator does the arithmetic. Move the target to the reconciled level in steps, as the Smart Bidding strategies guide advises for any target change.
Record the result and repeat monthly
Write one row per month: the period, the platform figures, each reconciling line, the business figure, the ratios and any fault found and fixed. The monthly paid media review checklist puts the reconciliation on the agenda, and the monthly reporting pack SOP places it in the pack's order of work. Sharing the sheet with finance, rather than only its conclusion, is what turns a disputed number into an agreed one; the diagnosis for finance distrusting marketing numbers explains why the trace matters as much as the total.
A month reconciled (illustrative)
Whiskferry Bakeware reconciles a closed month
Whiskferry Bakeware, an invented online bakeware shop, reconciles a month that closed long enough ago for its click window to have finished. Google Ads spent £16,800 and reported a purchase value of £58,900, a ROAS of about 351% against a platform target of 350%. Its order system stores the transaction ID the purchase tag passes.
| Line | Purchases | Value |
|---|---|---|
| Google Ads purchases, by conversion time | 1,240 | £58,900 |
| Less view-through and engaged-view purchases | −55 | −£2,500 |
| Less cross-device, enhanced-match and modelled purchases with no order identifier | −140 | −£6,600 |
| Platform purchases the order system could match | 1,045 | £49,800 |
| Tag fired on a failed payment, no completed order | −21 | −£740 |
| Unexplained residual | −14 | −£460 |
| Orders carrying a Google Ads transaction ID | 1,010 | £48,600 |
| Less refunds and cancellations on those orders | Unchanged | −£4,100 |
| Business-confirmed net value | 1,010 | £44,500 |
The value ratio is £44,500 ÷ £58,900, about 0.76, and the last three months ran at 0.78, 0.77 and 0.75, so the gap is structural. The failed-payment line is a fault: the tag fires on the payment page before payment is confirmed, and the developer moves it. On confirmed net value the month returned about 265%, below the 300% the business needs from its margin. The shop therefore moves its platform target ROAS towards 300% ÷ 0.76, about 395%, in two steps, and records the month as the first line of a running reconciliation sheet shared with its accountant.
Common mistakes
- Reconciling an open period. The platform figure is still growing; the gap looks smaller than it is.
- Comparing click-date figures with order dates. Export by conversion time.
- Treating every difference as an error. View-through, modelled and cross-device credit are structural.
- Treating every difference as structure. Tags firing on failed payments or reloads are faults to fix.
- Comparing gross platform value with net revenue without a returns line. The platform records value before returns.
- One reconciliation, never repeated. The trend in the ratio is the useful part.
- Setting targets on platform numbers alone. A target met in the platform can miss the business's return.
- Keeping the sheet from finance. The trace, not the total, is what builds trust.
