Careers
Part of The honest guide to marketing attribution
Nine attribution question patterns across ecommerce, B2B, and lead generation
Nine attribution question patterns across ecommerce, B2B, lead generation, retail, apps, content, and email, with named tools and provider lookback rules.
What to take away
- Start each example with an outcome and decision, not a favored model.
- The correct unit may be a person, account, household, device, or location.
- Model comparison shows sensitivity; it does not identify causal truth.
Marketing attribution examples are useful when they expose how the unit, path, outcome, and decision change. These nine scenarios are templates for questions, not promises that a particular allocation will be accurate.
Named tools make the comparison concrete. Ecommerce examples use Google Ads and Adobe Customer Journey Analytics. B2B examples use HubSpot and Salesforce. Lead-generation examples use CallRail and Invoca.
Google Ads documents a default 30-day click and 1-day view conversion window. Adobe Attribution IQ supports lookback windows from 1 to 90 days. HubSpot attribution reports offer 1, 7, 30, 90, 180, and 365 day lookback windows. These are provider rules, not universal recommendations.
Where no public figure exists, label local windows as typical ranges, such as a 90 to 180 day B2B cycle or a 30 day lead-generation call window.
Nine distinct attribution patterns
| Business setting | Outcome | Useful attribution question |
|---|---|---|
| Ecommerce | Mature paid order | How does credit move after refunds? |
| B2B | Accepted opportunity | Which account contacts enter the path? |
| Subscription | Retained paid cohort | How does maturity change source value? |
| Retail | Matched in-store sale | What share lacks digital identity? |
| Lead generation | Qualified appointment | How are duplicate and reopened leads handled? |
| Mobile app | Post-install value | Which install and reengagement rules apply? |
| Content | Sales-assisted inquiry | How does content persistence alter paths? |
| Renewal | Does suppression change eligible exposure? | |
| Local service | Completed job | How are calls, forms, and walk-ins deduplicated? |
Adobe documents attribution component settings for model, container, and lookback choices in Customer Journey Analytics, including product examples where different settings redistribute the same value. These are software mechanics, not universal recommendations. A different way to divide the same value is marketing mix modeling, which works on aggregate spend rather than recorded paths.
Example: compare path views without mixing totals
Google Ads describes its attribution reports and notes differences involving path metrics, event timing, networks, sources, and report constraints. The documented coverage explains its own reports. It does not represent every customer contact.
Compare Path Views Safely
- Freeze conversion action and population
- Extract source-system count and value
- Build path on stated identity unit
- Apply two models to same records
- Explain channel movements via paths
- Keep causal question separate
A good example ends with a decision limit. A B2B account report may help sales see covered engagement but cannot tell which contact caused the opportunity. A retail match may reveal known digital paths but cannot silently assign unmatched sales to the same distribution. Copying an example past its stated limit is one of the standard marketing attribution mistakes.
Adapt the unit and maturity before the model.
Turn examples into test cases
Treat every example as a bounded pattern. Record its audience, source data, identity rule, business constraint, decision, and outcome window before transfer. Replace the original figures with local inputs and test a failure case. If the example depends on unavailable rights, staff, or product functions, label it unsuitable instead of copying only its visible report.
NIST's Privacy Framework starting guide describes an organizational process for identifying and managing privacy risk. Use it to assign owners, describe data processing, and record responses. The framework is voluntary and does not establish legal compliance or approve a particular marketing use.
The W3C information and relationships explanation describes how visual structure and relationships should also be available programmatically. Apply that test to the tables, controls, and summaries an example produces. Passing one criterion does not establish full accessibility, so test the real output with the users and technology in scope.
Keep the evidence record beside the decision so a reviewer can reproduce the reasoning without relying on memory. Rebuilding an example locally follows the same steps as measuring marketing attribution from scratch. Set the next review date and name the change that would trigger an earlier check.
Common questions
Should examples use real percentages?
Use verified local results with methods and limits, or label numbers as hypothetical. Never invent performance.
Can a model split one conversion?
Yes. Fractional channel values should still reconcile to the eligible source outcome under the same rules.
Which example fits a mixed online and offline business?
Use one that makes match coverage and unmatched outcomes visible, then validate identity, timing, privacy, and reconciliation locally.
