Marketing attribution: models, tools and limitations
Marketing attribution assigns credit for a sale or action to the interactions preceding it. It helps you read customer journeys and compare channels under a defined rule. Budget decisions also require reliable tracking, confirmed sales and margin. Attributed revenue alone does not measure the additional profit a campaign generates.
Start with the action you want to measure
Choose an action relevant to your decision. For an online shop, track purchases and reconcile the amounts with orders, refunds and cancellations. For a service business, distinguish a submitted form, qualified enquiry and signed contract. An advertising platform may observe the form without knowing the sales outcome.
Define these stages with your sales team. A prospect might submit two forms for the same need; a customer might buy several times. Decide how to count these cases and retain the identifiers needed for reconciliation, subject to applicable data rules. Record the definition, person responsible and source that confirms the outcome.
Also distinguish the key events you track in GA4 from conversions used for advertising reports or optimisation. Check your account settings before interpreting their figures as sales.
Models available in GA4
Google describes three models in October 2026: data-driven, paid and organic last click, and Google paid channels last click. The first uses data-specific weights. The second credits the last eligible non-direct interaction; the third prioritises Google Ads, falling back to the second. Entirely direct paths and video interactions have specific rules. Google models and rules.
First-click, linear, time-decay and position-based models remain useful concepts for understanding credit allocation. Google removed them from GA4 in November 2023. An older tutorial asking you to select them no longer describes the current reports. Attribution model comparison report.
A CHF 100 sale example
Fictional example. Someone clicks a social ad, returns through organic search, then clicks a Google ad before buying for CHF 100. Assume these interactions are observed and eligible within the chosen window.
Social advertising
Last non-direct click — CHF 0
Illustration of shared credit — CHF 20
Organic search
Last non-direct click — CHF 0
Illustration of shared credit — CHF 30
Google ad
Last non-direct click — CHF 100
Illustration of shared credit — CHF 50
Total
Last non-direct click — CHF 100
Illustration of shared credit — CHF 100
The final column uses invented amounts to explain the mechanism. They are neither actual Google model output nor weights to apply. In this journey, Google paid channels last click would also credit the Google ad with CHF 100.
The business made one sale, regardless of credit allocation. If several platforms claim this order, adding their attributed revenues would count it several times. Start with confirmed orders when establishing realised revenue, then reconcile advertising reports with that reference.
Margin requires a separate calculation: product or service cost, advertising spend and the other costs included in your analysis. A campaign can receive substantial credit and remain unprofitable. Our marketing KPI guide explains the measures to connect with commercial goals.
Check report scope and the lookback window
In GA4, first-user, session and event dimensions answer different questions. A reporting-model change affects reports using event-scoped traffic dimensions, including relevant historical and future data. It does not change user- or session-scoped traffic dimensions. A lookback-window change applies going forward. GA4 attribution settings.
Before comparing figures, record the report, dimension scope, period and interaction eligibility window. A campaign several weeks before a purchase may fall outside a short window. A longer window may include more interactions without establishing their commercial contribution.
Keep an export and the settings before a change. If you change the model and compare an old export with a new report, you might mistake a reporting-method difference for a marketing change.
Investigate differences between tools
GA4, advertising platforms and your CRM may show different results. Work through these checks to understand a discrepancy.
1. Compare the same action.
Check that you are comparing purchases with purchases, or qualified enquiries with qualified enquiries. Check duplicates and the conversions included in each column.
2. Align the scope.
Record the period, time zone, currency and cancellation handling. Check whether the report groups outcomes by conversion date or advertising-interaction date.
3. Compare the rules.
Record the model, window, eligible channels and interactions considered. A click and an ad exposure represent different observations.
4. Check tracking.
Test events, order identifiers and campaign parameters. Examine breaks during payment, domain changes or transfer to the CRM.
5. Document missing information.
Tools do not receive every interaction or sale. State known limitations rather than forcing totals to agree.
For each discrepancy, record your hypothesis, check and conclusion. An export of ten orders can help identify duplicate reporting; it cannot establish the error rate across your entire business.
Also examine rows without usable attribution. Google distinguishes “(not set)” where information is missing and “Unassigned” where no channel rule matches. Diagnose the causes of these labels rather than removing them from the report. Understanding the attribution report.
Choose tools around the decision
Use platform reports to monitor and optimise their campaigns, GA4 to examine observed journeys on your website, and the order system or CRM to confirm commercial results. Define each tool's responsibility rather than selecting one number without a definition.
A small business can start with consistent campaign parameters, tested event tracking and an enquiry-reconciliation file. A long B2B sales cycle may need CRM stages and commercial outcomes supplied to advertising systems, subject to data protection requirements and tool capabilities.
A data warehouse can bring sources together and apply explicit rules where volume and needs justify it. It requires maintenance and cannot recreate interactions you did not observe. Assess the cost and decision it will support before building it.
Examine attribution and incrementality
Attribution describes credit under a model and its data. To estimate campaign incrementality, you investigate what would have happened without the campaign. Google describes a counterfactual method for its data-driven model. Its conversion credit does not replace your analysis of additional net profit across the business.
A test comparing exposed and control groups can help examine this effect if you can arrange a valid comparison. Define the outcome, duration, required volume and factors that could distort the comparison. Groups with different previous buying behaviour, or periods with different promotions, require cautious interpretation.
If your volume cannot support a conclusive test, use available information to make a limited decision you can revisit. Report decimals do not justify precision your data cannot support.
Keep a record for each review
Record the action measured, definition, sources and reconciliation, model, window, report scope and known limitations. Add the expected margin or commercial value, budget decision and review date. Your team can then explain why it increased, maintained or reduced spending.
For paid advertising or conversion optimisation, bring your reports and definitions of commercial outcomes. We can help identify measurement differences and the decisions they affect.