What This Chart Shows
This chart provides a radar view that places several top-performing levers (usually series/touchpoints) on the same polar grid so you can compare their shape across multiple dimensions at once. Each series is represented as a colored polygon on the radar, with each axis representing a different metric (Lift, ROI, Confidence, etc.). The shape of each polygon shows at a glance where a series excels and where it may have weaknesses.
Key Questions This Chart Helps Answer
- How do my top series compare across multiple dimensions (lift, efficiency, model confidence)?
- Which series are strong all-around vs strong on some metrics but weak on others?
- How much of each series' attribution is from the physics model vs the AI residual?
- Are there trade-offs between lift and efficiency among my top performers?
Axes, Metrics, and Units
Element | Description |
|---|---|
| Radar axes (spokes) | Each spoke is a different performance metric: Lift (HCF), Attributed ROI (costs mode only), Physics, Model Residual, Confidence. All values are normalized 0-1 for comparison. |
| Polygons | Each colored polygon represents one series. Larger area = stronger overall performance. |
| Radial position | Distance from center (0) to edge (1) shows normalized performance on that metric. |
| Legend | Shows series names with their corresponding colors. |
Metrics shown:
Metric | Meaning |
|---|---|
| Lift | Attributed lift. |
| Attributed ROI (costs mode only) | Lift divided by cost for that series (efficiency). |
| Physics | Same as Lift; the physics-based component of attribution. |
| Model Residual | The AI-allocated portion of the total model residual, distributed proportionally by series lift. |
| Confidence | Model confidence score: 1 minus the ratio of AI residual to total attribution. Higher = more physics-driven, more interpretable. |
All values are min-max normalized (0-1) across the displayed series for visual comparison. Raw values are shown in hover tooltips.
Control Options Reference
Control | Meaning |
|---|---|
| KPI classification / hierarchy / drilldown | Which outcome (KPI / target) to attribute to. |
| Date range | Optional time window for temporal attribution. |
How to Interpret the Results
- Polygon shape: A larger, more balanced polygon indicates strong all-around performance. A "spiky" polygon (large on some axes, small on others) indicates trade-offs.
- Compare polygons: Overlapping polygons show where series are similar; diverging areas show where one outperforms another.
- Confidence axis: High values (toward the edge) mean more of the attribution comes from the interpretable physics model. Low values mean more reliance on the AI residual.
- ROI vs Lift: A series can have high lift but low ROI (high spend, good results but expensive) or low lift but high ROI (small but efficient).
- Hover: Shows the series name and the actual normalized value for that metric.
Practical Applications for Marketers
Application | How to use this chart |
|---|---|
| Peer benchmarking | Compare top levers on multiple dimensions in one glance. |
| Efficiency vs impact | In costs mode, contrast lift and ROI spokes before scaling spend. |
| Model trust triage | Low confidence or high residual (relative to peers) ⇒ candidate for review or validation. |
| Executive snapshot | Strategic tier—good for discussion, not fine-grained optimization alone. |
Common Mistakes & Misinterpretations
Mistake | Why it is a problem | How to avoid |
|---|---|---|
| Reading radius as dollars or ROAS | Radius is 0–1 within the chart cohort. | Use other charts for absolute lift/ROI; treat radar as profile. |
| Treating “Confidence” as statistical CI | It’s a formula from residual vs lift, not sampling error. | Don’t use it like p-values or confidence intervals. |
| Comparing radars from different filters/dates | Normalization re-scales each chart separately. | Compare shapes cautiously across charts; absolute ranks need raw views. |
| Assuming causal proof from shape | All metrics are model outputs. | Pair with experiments for incrementality claims. |
Caveats & Considerations
- Top-k selection: By default, the chart shows the top 5 series by absolute lift. You can override this with selected_series (up to 8).
- Normalization: All metrics are min-max normalized per axis. This means the scale is relative to the displayed series, not an absolute benchmark.
- No-costs mode: If cost data is unavailable (detected automatically), the ROI axis is hidden and the chart title includes "(No Costs Mode)."
- AI residual allocation: The Model Residual axis distributes the total AI residual proportionally across series by their lift share. This is an allocation, not a direct measurement.
- Confidence calculation: Confidence = 1 - (AI residual / (AI residual + Physics lift)). A series with 0 AI residual has Confidence = 1.0.