Why Your Advertising and Store Reports Disagree, and How to Investigate
Explain advertising and store differences through comparable definitions, transaction traces and refund adjustments without forcing incompatible reports to agree.
The Short Answer
Describe one exact discrepancy and make the comparison consistent in dates, timezone, currency, values and order states. Check the source dimension's scope, conversion settings and reporting maturity before assuming transactions are missing. Trace a small authorized set of records where the systems permit it.
Separate definitional differences, collection limits, overlapping platform credit and verified implementation defects. Attribution assigns credit under rules; it does not by itself establish additional sales. Repair confirmed faults at their source, preserve historical definitions and document unresolved uncertainty. The investigation should establish what the business can conclude rather than make every dashboard display the same number.
Key takeaways
- Similar report labels can represent different dates, values and source scopes.
- Check settings and reporting maturity before tracing missing records.
- Keep transaction totals separate from overlapping attribution claims.
- Document confirmed explanations, repaired defects and unresolved limits.
Different reports can assign different credit to the same purchase. A store may record the order when it was placed, an advertising report may organize the result around an earlier interaction, and an analytics report may classify the visit under a different source. A mismatch is worth examining, but it is not automatically a bug.
The useful goal is to understand which differences are expected, which are caused by settings or missing data, and which prevent a decision. Trying to force every dashboard to show one number can remove useful distinctions and hide real defects.
Start with one clearly defined discrepancy. “The systems disagree” is too broad. “For these dates and this purchase definition, one source shows a materially different value” gives the team something it can investigate.
Create a comparison that is genuinely comparable
Write down the exact reports, filters, dates, timezone, currency and extraction time. Identify whether each value represents orders, purchase events, customers, gross sales, net sales or attributed conversion value. Similar labels can hide different definitions.
Check the date basis. A report organized around advertising interaction dates and a report organized around order dates will not necessarily align day by day. Also identify whether later refunds or adjustments change historical periods or appear when the adjustment occurs.
For a fictional comparison, imagine one dashboard showing $12,000 and another $10,000. Before looking for lost transactions, establish whether the first includes shipping and tax while the second excludes them. If the value bases differ, part of the gap may be definitional rather than missing data.
Do not start by applying an arbitrary percentage correction to one source. A convenient multiplier might hide several problems moving in opposite directions.
The acquisition reporting guide explains how commercial and channel views fit together. This investigation focuses on why their numbers differ.
Understand which source each report is describing
Attribution is the method used to assign credit to interactions before an outcome. It is not the same as recording that the transaction happened. A customer can discover a brand through one route, return through another and buy after several visits.
Google Analytics distinguishes traffic-source information at user, session and event scopes in its source-dimension documentation. A first-user source describes a different question from the source of a later session or credit assigned to a key event. Comparing those columns without their scope creates avoidable confusion.
Read the report's actual definition rather than relying on the word “source.” Likewise, a channel group is a classification built from rules; it is not an independent observation of everything that influenced the customer. Check the current channel-group definitions when a classification looks unexpected.
Keep the interpretation narrow. A first recorded source is not proof of the first time a person ever encountered the brand. A session classified as direct does not establish that no advertising influenced the purchase.
Check settings and maturity before investigating individual records
Record the conversion action, attribution settings, counting rules, value source and interaction windows in each advertising system. These choices affect which outcomes can be credited and how they appear.
Google's conversion-window documentation describes the time after an interaction in which a conversion can be recorded. Do not assume that two systems use the same window or interaction types just because their reports cover the same calendar dates.
Check whether the period is mature enough for the intended comparison. Recent purchases may not have been reported everywhere, and later outcomes can still be attached to earlier activity. Refresh an appropriate historical period and record the new extraction time.
If settings changed during the period, separate the before and after sections. A wider window or a new value definition can change reporting without a corresponding improvement in customer demand.
These checks often narrow the investigation. They do not justify dismissing every difference as “attribution.” A missing currency or duplicate purchase still needs to be repaired.
Trace a small set of known transactions where the systems allow it
Use an approved, privacy-conscious test process or a restricted set of authorized records. Start with a normal order and, where relevant, a discounted order, a partial refund and an order spanning a reporting boundary. Compare identifiers and amounts without exporting unnecessary customer details.
Follow the path from the order record to the measurement event and destination. Was the event sent? Did it contain the expected transaction reference, currency, items and value? Was the same event sent by two integrations? Did a confirmation-page refresh create another logical purchase?
Not every platform exposes a transaction-level attribution trail. Where only aggregated reporting is available, state that limit. Do not imply that a modeled or aggregate total can be fully reconstructed from individual customer records you cannot actually inspect.
Use a short investigation table:
| Finding | Likely next check | Possible conclusion |
|---|---|---|
| Same order, different amount | Tax, shipping, discounts and refund definitions | Expected value difference or mapping defect |
| Repeated transaction reference | Multiple senders and destination duplicate handling | Possible duplicate collection |
| Order exists but no browser event | Journey, consent state and collection errors | Expected collection limit or missing event |
| Source changes after a redirect | Actual URL parameters and referral handling | Possible source loss |
| Recent totals increase later | Conversion and processing delay | Period was incomplete |
Those are investigative possibilities, not automatic diagnoses. Confirm the path before making a correction.
Refunds deserve a specific trace because a store action does not prove every measurement destination received an adjustment. Start with the original order reference, each distinct refund reference, its status and processing date. Preserve product value, refunded delivery charges, tax and currency separately where the records support them.
Distinguish a completed full refund from a partial refund, an unpaid cancellation, an exchange and a disputed payment. They do not necessarily imply the same money movement or reporting treatment. Multiple partial refunds also need separate records so a repeated import does not deduct the same amount twice.
For an explicitly fictional example, an order has $120 of product value, $10 delivery and $13 tax. A completed refund returns $30 of product value and $3 of tax. Retained product revenue is $90, while the net customer payment under these simplified assumptions is $110. Those totals answer different questions; neither should be copied blindly into every conversion-value field.
Build a destination check alongside the original transaction:
| Destination | What to verify |
|---|---|
| Store and payment records | Refund completed, identifiers match, and component amounts explain the actual movement |
| Analytics | The configured refund event arrived with the intended original transaction and item/value mapping |
| Advertising conversion action | The specific source supports an adjustment, identifiers match, and the permitted method was accepted |
| Business report | Gross and retained values remain distinguishable, with the date basis and extraction time recorded |
Google Analytics documents a refund event tied to the transaction ID, with item information supporting item-level reporting. That describes an implementation path; it does not mean your store's integration sends it automatically. Verify the integration and resulting records.
Google Ads separately distinguishes restating conversion value and retracting a conversion. Their effects on value and counts differ, and retraction has lasting consequences. Check the current action's eligibility, identifiers, timing and supported adjustment route before an authorized change. Do not assume a refund in analytics automatically corrects a separately tracked advertising purchase.
Decide how the business review handles time. A refund processed this week can relate to an order acquired last month. Show the processing-period view for current operations and, where needed, a clearly labeled revised order-cohort view for acquisition economics. A newer cohort has had less opportunity to generate returns, so compare similarly mature cohorts or disclose the difference.
Keep adjustment attempts and outcomes visible: accepted, rejected, not found, unsupported or still pending. Investigate the reason rather than repeatedly uploading changed amounts until a dashboard matches. Preserve prior exports and document restatements. The finished explanation should identify which difference is timing, which is value definition, and which is a missing or incorrect adjustment.
Examine missing and overlapping information honestly
Some visits cannot be observed or joined under the site's approved measurement setup. Consent choices, browser behavior, device changes and incomplete identifiers can limit the available journey. Do not fill every gap with a confident source label.
At the same time, more than one platform can claim credit for one purchase. Adding those platform claims together can exceed the store's sales. Keep them as separate attributed views rather than treating the sum as a transaction ledger.
Check campaign naming and tracking parameters for consistency. Test the actual route through redirects and landing pages. A parameter in the original link is not proof that it survives to the measurement point.
Avoid using internal navigation links to create a new external campaign story. The website should preserve a clear distinction between how someone arrived and where they clicked inside the site. Follow the destination's documented campaign behavior and test the effect of any change.
The event-taxonomy guide is the next step when the investigation reveals an unclear trigger, identifier or value definition.
Separate attribution from the question of additional sales
An attributed purchase means the system assigned credit under its rules. The business may also want to know how many purchases would not have happened without the campaign. That is a causal question, and the attribution report alone does not answer it.
A suitable experiment can help investigate additional effect, but the design matters. Audience overlap, geography, timing, contamination and sample size can all affect interpretation. Do not relabel a simple before-and-after comparison as a controlled experiment.
Use attribution as one source of evidence for decisions inside and across channels. Compare it with total orders, new-customer patterns, contribution and credible experimental evidence when available. None of those views needs to become a magical single number.
The creative-testing guide discusses the difference between observations and stronger comparisons in a practical campaign context. Preserve that distinction when explaining a result to the business owner.
Close the investigation with a documented explanation
Classify the gap into confirmed definitional differences, verified implementation faults and unresolved parts. State the size and period where possible. Keep the evidence behind the explanation so the next analyst does not repeat the same investigation.
Repair verified faults at their source. If two integrations send the same purchase, clarify ownership and test the correction. If values use different definitions, label the reports or adjust the intended mapping through a controlled change. Do not rewrite historical numbers merely to create agreement.
Recheck the original cases after the fix and monitor a subsequent comparable period. Document any backfill, exclusion or restatement, including what remains unavailable. A report with a visible limitation is more trustworthy than one that silently erases it.
Then return to the decision that prompted the work. Perhaps the evidence is now sufficient to compare campaigns. Perhaps an affected period should be excluded from that decision. The investigation is complete when the team understands what it can responsibly conclude, not when every dashboard happens to look alike.
Questions about attribution discrepancies
Should advertising revenue add up to the store's revenue?
Not necessarily. Different platforms can claim credit for the same purchase, and their date and value definitions can differ from the store. Keep the transaction total separate from attributed views and reconcile their meanings before comparing them.
Can changing the attribution model fix poor campaign performance?
It changes how credit is assigned in the affected reporting; it does not itself create better customer demand or order economics. Evaluate the model for the question it supports, record the change and avoid presenting a reporting shift as a commercial improvement.
What if we cannot explain the entire gap?
Document the parts you verified and the limits that remain. Identify whether the uncertainty is large enough to change the intended decision. Continue a focused investigation where it matters instead of inventing a precise reconciliation from incomplete information.
Explore the analytics and measurement guides for related event, reporting and business-economics work.
Sources
Still have a question?Ask the DTCUP Community ↗

