Marketing Attribution: Make Credit Useful
Marketing attribution is the process of assigning conversion credit to marketing interactions. An attribution model determines which recorded touchpoints receive that credit and how much. To make the result useful, define the conversion, establish which interactions qualify, and report the credit rule alongside the numbers.
What marketing attribution tells you
Attribution answers which marketing interactions receive credit for a recorded outcome. Incrementality addresses how many additional outcomes occurred because of the marketing. Google’s distinction between attributed and incremental conversions separates conversions counted under tracking and attribution settings from additional conversions measured through an exposed group and a control group.
Treat a conversion count, a credited revenue amount, and an incremental effect as separate measures. For campaign reporting, name the outcome and model explicitly. For a decision about additional advertising spend, assess incremental impact as well as attributed performance.
Common marketing attribution models
First-touch and last-touch rules assign credit to one interaction. Linear, time-decay, and position-based rules distribute it across interactions. The following definitions describe model mechanics; the model names alone do not establish which options a particular platform offers.
| Model | How it allocates credit |
|---|---|
| First touch | Assigns all credit to the earliest eligible interaction. |
| Last touch | Assigns all credit to the latest eligible interaction before conversion. |
| Linear | Shares credit equally across eligible touchpoints. |
| Time decay | Gives greater weight to interactions closer to conversion. |
| Position based, including U shaped | Weights positions in the journey. With at least three touches, Adobe’s U-shaped model gives the first and last 40% each and shares the remaining 20% among intervening touches. |
| Data driven or algorithmic | Uses an algorithm to determine credit from available data. |
Google’s data-driven model evaluates converting and non-converting paths, incorporating interaction timing, device type, and exposure order. Its model is specific to the advertiser and key event.
Google Analytics currently lists data-driven, paid and organic last click, and Google paid channels last click. Its first-click, linear, time-decay, and position-based models were removed in November 2023. Check product availability before using a general model comparison as configuration guidance.
Google Analytics models exclude direct visits from credit unless the entire path is direct. Thus, “last click” in this implementation does not simply mean the final visit regardless of its source.
How to choose an attribution model
Match the model to the reporting question: first touch for earliest recorded sources, last touch for final interactions, and shared credit for the recorded path. These are interpretations of the rules above.
Choose one primary model for recurring reporting and keep an alternative for comparison. Hold the conversion definition, reporting period, eligible channels, and lookback window constant during that comparison. Record whether a proposed campaign or budget change still makes sense when the model changes.
For data-driven attribution, review the platform’s documented inputs and eligibility rules. Do not select it solely because it uses an algorithm. Include data coverage and the intended reporting question in the selection criteria.
Keep the model stable enough to compare reporting periods, and document changes with an effective date. When historical results are restated, retain an export of figures used for earlier decisions.
How to set up marketing attribution
1. Define the outcome and its value
Choose a specific action to measure: a purchase, a completed registration, or a submitted lead form. Record its counting unit and the system that holds the underlying event. Keep purchases, leads, and other actions distinguishable in reporting.
In Google Ads, the Conversions column generally covers primary conversion actions, while All conversions also covers secondary actions and other conversion sources. Check the action list before treating a headline total as sales or new customers.
For revenue reporting, specify the value field, currency, and treatment of refunds or cancellations. Keep actual purchase value separate from an assigned value for a non-purchase action. Label whether a report counts transactions, people, or conversion events.
2. Record marketing sources consistently
For manually tagged links, Google’s campaign URL documentation identifies utm_source as the referrer, utm_medium as the marketing medium, and utm_campaign as the campaign identifier or name. These parameters pass campaign information to Analytics when a referral link is clicked.
Adopt a shared naming convention and check links before distribution. UTM values are case sensitive, so capitalization differences can create separate reporting values. Keep source, medium, and campaign spelling consistent, and verify that each tagged link reaches its intended destination with its parameters intact.
Maintain a documented policy for manual tags and advertising-platform identifiers. Include the source, interaction timestamp, campaign identifier, and conversion identifier in data-quality checks.
3. Specify how interactions are connected
Google Analytics uses User-ID, device identifiers, and modeling as reporting identity spaces. Its blended, observed, and device-based options use different combinations of these methods.
Document the identity method used in each report. For signed-in activity, verify that the same persistent identifier is assigned consistently. For records linked across systems, document the matching key and check unmatched records separately. Preserve an unknown category when the available data does not support a reliable connection.
4. Prevent duplicate outcomes
For web-stream purchase events, Google Analytics deduplicates purchases with the same transaction ID. This mechanism does not apply to app streams, and an order identifier must be unique to its transaction.
Validate conversion collection against the underlying transaction or event records before analyzing channel shares. Check repeated event delivery, missing identifiers, incorrect values, and reused transaction IDs. Record how corrections enter the reporting dataset.
5. Set the lookback window and eligible interactions
The Google Analytics lookback window determines how far before a key event a touchpoint can qualify for credit. The window is distinct from the date range used to select outcomes for a report.
Choose the window with reference to the recorded time between interactions and conversion. Document which channels and interaction types qualify, including clicks and eligible ad views. Show the window next to the model name rather than leaving it as an undocumented setting.
6. Reconcile the report before using channel credit
Compare the recorded outcome total with the total admitted to attribution. Review unmatched records, exclusions, duplicates, and differences in value or date definitions. Keep missing or unassigned records visible instead of redistributing them to known channels without a documented rule.
For a model that distributes one outcome’s credit across eligible touches, check that the shares reconcile to that outcome. Distinguish this from participation reporting: Adobe’s participation model gives full credit to every unique touchpoint, so its summed credit can exceed the outcome total.
How to read attribution metrics
Google Analytics can split key-event and revenue credit across interactions. Fractional conversion counts represent allocated credit, rather than fractions of people or orders. Read the amounts together with the model, eligible population, and value definition.
| Metric | What to examine |
|---|---|
| Attributed conversions | The key-event credit allocated to a channel, including fractional shares where the model distributes credit. |
| Attributed revenue | The revenue credit assigned under the selected model. Verify that the underlying value represents revenue. |
| Cost per conversion | In Google Ads, eligible cost divided by reported conversions. Check which actions and interactions enter the calculation. |
| Attributed return on ad spend | Attributed conversion value divided by ad spend. Keep it distinct from incremental ROAS, whose numerator is incremental conversion value. |
Match the cost and conversion populations before comparing campaigns. Use consistent currencies and periods, and keep non-revenue conversion values clearly labeled. Retain the underlying event count alongside allocated credit when reconciling transactions.
Include the data extraction date and the attribution settings with report exports. In Google Analytics, changing the reporting model applies to historical and future data, whereas a lookback-window change applies going forward. Those settings therefore have different implications for comparisons over time.
Why attribution reports disagree
Different traffic-source scopes
Google Analytics distinguishes user, session, and event traffic-source scopes:
- First user dimensions describe the source that initially acquired the user.
- Session dimensions describe where sessions originated.
- Event dimensions support credit allocation for key events.
The selected reporting attribution model affects event-scoped dimensions, while user and session dimensions are unaffected. A user-acquisition report and an event-attribution report therefore answer different questions. Compare reports at the same scope before investigating a channel discrepancy.
Different dates, windows, and counting settings
Google Ads’ primary conversion columns report by ad-click time. Its conversion-time columns report by when the conversion occurred. Align these date definitions when reconciling outcomes across systems.
Google’s discrepancy guidance identifies other causes, including conversion windows, counting settings, tagging, cross-device measurement, view-through conversions, and processing delays. Check those settings before interpreting a mismatch as a change in marketing performance.
A Google Ads attribution-report lookback window and a conversion window also serve different purposes: the former controls eligible prior interactions, while the latter controls when a conversion can be recorded after an interaction. Matching only the displayed number of days does not establish equivalent measurement.
Incomplete identity and modeled data
When users decline Analytics identifiers such as cookies, Google Analytics’ reporting identity documentation describes behavioral modeling based on similar consenting users. Modeling is an estimate rather than an observed interaction record for each affected user.
In practical terms, document whether a report includes modeled data and which identity options it uses. Review missing identifiers and unmatched outcomes alongside the channel results. Interpret credit within that stated coverage rather than assuming the report contains every interaction.
Attribution, incrementality, and marketing mix modeling
Use attribution to analyze recorded paths and allocated credit. For the additional outcomes caused by advertising, a lift study asks a different question. Google’s user-based Conversion Lift documentation describes comparing outcomes between users exposed to ads and users held back from exposure. Report the resulting incremental estimate with its uncertainty and tested population.
Marketing mix modeling works at an aggregate level. Google’s Meridian data specification uses media and spending data, a target outcome, and control variables organized over time and, ideally, geography. This specification implies that individual touchpoint paths are unnecessary model inputs.
The Meridian framework supports estimating channel contribution, modeling response to spending, and planning allocation. Its causal estimates depend on the framework’s assumptions and model inputs.
For ongoing campaign reporting, keep attribution definitions and data collection consistent. For causal questions, use suitable lift evidence. For broader channel allocation, assess aggregate modeling together with experimental results where available. State which question each result answers before using the numbers in a budget decision.