The Attribution That Blames the Wrong Channel
An e-commerce client uses a Last Touch channel report and a Linear attribution model in Analysis Workspace. Last Touch says Paid Search drives 55% of revenue. The client's CFO is skeptical and asks for a second opinion. You build a model comparison using several attribution models. The results show: Paid Search 55% (Last Touch), 22% (First Touch), 31% (Linear), 18% (Position-Based), and 12% (Algorithmic). Email shows 8% (Last Touch), 30% (First Touch), 20% (Linear). The client wants to cut Paid Search budget by half based on the Last Touch number. There is also a data-governance request: the CFO wants a channel-level revenue report shared with an external agency.
1. How do you interpret the model spread, and what would you advise before any budget cut?
Read the worked answer
The spread is the story. Paid Search's share collapses from 55% (Last Touch) to 12–31% under other models, while Email rises from 8% to 20–30%. This is the classic signature of a channel that closes conversions (it appears late in the path) being over-credited by Last Touch, and a channel that opens journeys (Email) being under-credited.
Interpretation:
- Paid Search is a strong closer - it captures the final click. Last Touch flatters it.
- Email is a strong introducer - First Touch and Linear give it far more credit.
- The Algorithmic model (12% for Paid Search) weights by observed conversion probability and is the most defensible for budget decisions, but it is a model, not truth.
Advice before cutting budget:
- Do not cut based on a single model. Cutting Paid Search by half will likely reduce conversions and shift the mix, so the historical attribution will not predict the new state.
- Run a controlled test (geo holdout or budget shift on a subset) rather than a global cut.
- Validate that the models are comparing the same date range, same channel rules, and same conversion definition - model comparison is only meaningful if the underlying data is identical.
- Present the range, not a point estimate: "Paid Search's contribution is somewhere between 12% and 55% depending on model" is the honest statement.
2. For the external agency report, what governance controls apply, and what must you verify about the channel data before sharing?
Read the worked answer
Governance controls:
- Access model. The agency should not receive login access to the parent suite. Use a Virtual Report Suite scoped to the channel data they need, with component curation so only approved dimensions and metrics are visible, and with the agency's users provisioned under a limited permission group.
- Data minimization. Share only channel-level aggregates, not visitor-level or PII-adjacent dimensions. Confirm no custom variables carry email addresses, customer IDs, or other identifiers.
- Contractual/legal review. External sharing requires the client's data-sharing agreement; confirm the agency is approved to receive this data.
Verify before sharing:
- The channel classification rules are correct and stable for the reporting period (a rule change mid-period makes the report misleading).
- The attribution model used is documented and disclosed on the report - an agency reading "Paid Search 55%" without knowing it is Last Touch will draw the wrong conclusion.
- The date range and conversion definition are stated.
- Any VRS curation does not hide a metric the agency needs, which would make the report look better than reality.
Key point: The governance risk here is not just access - it is misleading disclosure. Sharing a Last Touch number without its model label is a governance failure even if access is perfectly controlled.