Editorial illustration of separate coloured paths converging on a single solid sales receipt, with translucent overlapping evidence panels.
Marketing Attribution: When Reports Disagree
6 min read
Digital Marketing Strategy
Franky Wong
What should you do when marketing attribution reports disagree?
When marketing attribution reports disagree, align the sales outcome, reporting period and conversion definition before changing budget. Compare tracked interactions with customer recall, but do not add their credits together. Treat the remaining disagreement as a question to investigate through a controlled test, not proof that a channel has failed.
For a Hong Kong business taking enquiries through WhatsApp, telephone and website forms, the path to a sale can cross several records. A customer may discover the business through a private recommendation, search its name and later call. The advertising report, enquiry log and customer's explanation can each describe a different part of that journey.
The useful task is therefore not to force every dashboard into agreement. It is to decide what each record can establish, what remains unknown and which budget decision can safely wait. This is a practical reconciliation method, not a promise of perfect attribution.
Why can different reports all look plausible?
Marketing attribution assigns credit according to the interactions a system can observe and the rules it applies. A sales ledger answers whether a transaction happened. An advertising report assigns credit within its measurement settings. A customer survey captures what a respondent remembers. These records need not produce matching channel totals because they answer different questions.
MarTech's discussion of surveys alongside media mix modelling describes how private sharing and offline exposure can escape click based measurement. It also distinguishes aggregate patterns from customer recollection. That is useful context, not evidence that a particular channel caused your sales.
Before comparing reports, write down the event each calls a conversion, the reporting window and whether it includes repeat purchases or unqualified enquiries. Keep confirmed revenue separate from attributed revenue. If the definitions differ, label the difference before using the figures in a budget meeting. Agreement created by hiding those differences is not better measurement.
What should a customer source question capture?
Ask customers how they first heard about the business, then leave room for an explanation in their own words. Offer useful categories without forcing every answer into paid search or social media. Include referrals, offline discovery and an unknown option. Keep the question optional and separate from the customer's ability to complete a purchase or enquiry.
For a Hong Kong service business, a response might mention a friend's WhatsApp message, an event or a search for the company name. Those are hypothetical examples, not interchangeable channel labels. Preserve the original response alongside any category your team assigns.
Customer recall is directional evidence rather than a receipt for advertising impact. Respondents may remember a recent contact instead of their earliest exposure, and nonrespondents may differ from respondents. Review recurring patterns without treating survey shares as revenue shares. Avoid collecting private conversation content when a simple source description is enough to inform the decision.
How do you build a useful reconciliation record?
Start with a consistent cohort of confirmed customers or qualified opportunities. Do not compare a platform's enquiry total with completed sales and describe the gap as lost attribution. Pick the business outcome that the next spending decision is meant to improve, then document the records that relate to it.
Business outcome: retain a stable internal reference, outcome date and confirmed status. Apply an agreed treatment for cancellations and refunds.
Observed route: record the campaign or source information actually available. Preserve missing values rather than guessing where a customer came from.
Customer explanation: keep the stated discovery source separate from the observed route. Add a consistent category without deleting the original meaning.
Decision note: explain the disagreement, its likely consequence and the evidence needed before changing spend.
Use access controls appropriate to customer records. A reconciliation sheet does not need copied chat histories or unrestricted personal details. An internal reference and a short source description may be sufficient for the reviewer. Keep the working record focused on the decision rather than building a new customer database by accident.
When is disagreement a reason to test rather than cut?
Imagine a hypothetical local retailer whose customers frequently mention recommendations, while its advertising dashboard credits searches for the shop name. Neither record establishes how many purchases would disappear without advertising. Cutting search immediately could remove useful demand capture; crediting search with all discovery could neglect the activity that generated interest.
Write a falsifiable question instead: does reducing this activity change qualified demand or confirmed sales under reasonably comparable conditions? Define the outcome, acceptable business risk and stopping conditions before running a test. Avoid changing the offer, sales process and advertising approach together if you want to understand the result.
A holdout or geographic comparison may help in suitable circumstances, but it requires enough comparable activity and careful design. A business with sparse sales may need a longer observation period or specialist advice. Do not label normal variation as causal proof. Kathleen Schaub's measurement argument is useful here: variable customer behaviour calls for context and ranges, not unwarranted precision.
FAQ
Which report should decide the budget?
Start with confirmed business outcomes, then use attribution reports to investigate how customers arrived. No channel report should independently decide the budget when its conversion definition differs from the sales record. Resolve the definitions before comparing efficiency.
Can customer surveys replace tracking?
No. Surveys describe what respondents remember, while tracking records observable interactions under its own rules. Keep them separate and compare patterns. Missing responses, vague answers and recall bias should remain visible rather than being redistributed across named channels.
Should a small business start with media mix modelling?
Not automatically. Begin with consistent sales records, clear conversion definitions and usable channel history. Consider modelling when the business question and available data justify specialist assessment. An elaborate model cannot repair missing sales records or establish certainty from weak inputs.
What this means for Hong Kong SMEs
Before the next budget review, take a recent set of confirmed outcomes and compare the sales record, available tracking and customer source explanations. Mark what is observed, reported or unknown. Choose the disagreement most likely to change your spending decision and assign an owner to resolve it.
Keep that exercise connected to the wider marketing plan, rather than treating attribution as a separate dashboard project. Bring the disputed conversion definition and a sample of anonymised records to a marketing audit. The next step is a better decision, not another report that appears more certain than the evidence allows.
