What This Chart Shows
This chart shows how per-lever attribution rows were treated by the model’s contribution controls. It shows whether each lever’s attributed lift was accepted as-is, capped, or suppressed. Every piece of attribution goes through a governance process that validates whether the modeled contribution is trustworthy. The chart displays three governance states: PASS (attribution passed all checks), CLAMPED (attribution was adjusted to stay within reasonable bounds), and KILLED (attribution was suppressed due to quality concerns). This shows how much of your total attributed lift is high-confidence vs how much was flagged or adjusted by the model's internal quality controls.
Key Questions This Chart Helps Answer
- How much of my attribution passed all validation checks?
- What percentage of levers had their attribution clamped or suppressed?
- How much total lift was affected by governance controls?
- Is the majority of my attributed lift coming from validated, high-confidence sources?
- Are there data quality issues I should investigate?
Axes, Metrics, and Units
Element | Description |
|---|---|
| Left panel (Donut chart) | Item Count by State: Number of levers (series/channels) in each governance category. Shows count and percentage. |
| Right panel (Bar chart) | Lift Sum by State: Total attributed lift (in KPI units) for levers in each governance category. |
| X-axis (Bar chart) | Governance State: PASS, CLAMPED, or KILLED. |
| Y-axis (Bar chart) | Total Lift (KPI Units): Sum of attributed lift for all levers in that state. |
| Colors | Green = PASS, Yellow/Orange = CLAMPED, Red = KILLED. |
Governance States:
State | Meaning |
|---|---|
| PASS | Attribution passed all validation checks. High confidence in this lift. |
| CLAMPED | Attribution was adjusted (clamped) to stay within reasonable bounds. The model detected potentially extreme values and moderated them. |
| KILLED | Attribution was suppressed (set to zero or near-zero) due to quality concerns. The model flagged these as unreliable. |
All values are modeled outputs.
Control Options Reference
Control | Meaning |
|---|---|
| KPI classification / hierarchy / drilldown | Which outcome (KPI / target) to filter by. When a specific KPI is selected, governance counts reflect only that outcome. |
How to Interpret the Results
- Pie chart (left): Shows how your levers are distributed across governance states. Ideally, the majority should be PASS (green).
- Bar chart (right): Shows the total lift in each state. Even if few levers are KILLED, they may represent significant lift (or vice versa).
- High PASS percentage: Good—most attribution is validated and trustworthy.
- High CLAMPED percentage: Moderate concern—the model found extreme values and adjusted them. May indicate noisy data or unusual channel behavior.
- High KILLED percentage: Investigate—significant attribution was suppressed. This could indicate data quality issues, model fit problems, or channels with unreliable signals.
- Compare count vs lift: A few CLAMPED levers with large lift may be more impactful than many KILLED levers with small lift.
Practical Applications for Marketers
Application | How to use this chart |
|---|---|
| Trust assessment | Use the PASS percentage to gauge overall confidence in your attribution. |
| Data quality audit | High CLAMPED or KILLED rates signal potential data issues worth investigating. |
| Model validation | Compare governance distribution before/after model updates to assess improvement. |
| Stakeholder communication | Use this chart to explain that attribution is not blindly accepted—it goes through quality controls. |
| Prioritization | Focus optimization efforts on PASS levers; investigate CLAMPED/KILLED levers separately. |
Common Mistakes & Misinterpretations
Mistake | Why it is a problem | How to avoid |
|---|---|---|
| Assuming CLAMPED or KILLED means "bad channel" | Governance flags indicate modeling uncertainty, not necessarily that the channel is ineffective. Flags can arise from data sparsity or unusual patterns. | Investigate flagged levers before making cut decisions; the underlying channel may still be valuable. |
| Ignoring the lift panel | A small count of CLAMPED levers can represent a large share of lift (or vice versa). Looking only at count percentages can be misleading. | Always compare both panels—count distribution and lift distribution. |
| Expecting 100% PASS | Some level of CLAMPED or KILLED is normal, especially with sparse or noisy data. Zero flags may indicate the governance is too lenient. | Treat a healthy mix as expected; focus on trends over time rather than absolute perfection. |
| Using this chart to evaluate campaign performance | This chart measures model trust, not campaign effectiveness. A PASS lever can still have low lift; a CLAMPED lever can still be valuable. | Use this for data quality and model trust; use other charts (ROI, Lift Bars) for performance evaluation. |
| Comparing governance across different KPIs without context | Different KPIs may have different data quality characteristics, leading to different governance distributions. | Compare governance within the same KPI or understand why distributions differ across KPIs. |
Caveats & Considerations
- Lift sums can be negative: If a lever has negative attribution (e.g., cannibalization), its lift contributes negatively to the bar chart. A PASS lever can have negative lift.
- KPI filtering: When a specific KPI is selected, the chart reflects only levers attributed to that outcome. Aggregate view shows all outcomes combined.
- Model version dependence: Governance thresholds and rules may change across model versions. Compare distributions within the same model version.