What This Chart Shows
This chart illustrates how your total marketing attribution is built from three components: the Baseline (what would have happened without marketing), the Physics Model (interpretable, equation-based causal effects), and the Model Residual (AI-learned patterns that correct for effects the physics model doesn't capture). It shows how these components add together to produce your final attributed outcome giving you the transparency into the hybrid modeling approach and helps you understand the composition of your total marketing impact.
Key Questions This Chart Helps Answer
- What is my baseline KPI performance without marketing influence?
- How much of my total marketing attribution comes from the interpretable physics model?
- How much comes from the AI-learned model residual correction?
- What is the relative contribution of each model component to total lift?
Axes, Metrics, and Units
Element | Description |
|---|---|
| X-axis | Model Component: Baseline, Physics (Structure), Model Residual, Total. |
| Y-axis | KPI Contribution (Lift Units): The attributed value in your KPI's native units (e.g., revenue, conversions). |
| Baseline bar | Starting point: The predicted KPI outcome without any marketing lift. |
| Physics (Structure) bar | Incremental contribution from the interpretable. |
| Model Residual bar | Incremental contribution from the AI-learned residual, which captures patterns the physics model doesn't explain. |
| Total bar | Final outcome: Baseline + Physics + Model Residual. |
| Connectors | Gray lines connecting bars to show the flow from one component to the next. |
| Bar colors | Green for increasing values, red for decreasing values, distinct color for totals. |
| Hover | Shows the value and, for Physics/Model Residual, the percentage share of total lift. |
Model Components Explained:
Component | Description |
|---|---|
| Baseline | The counterfactual outcome—what your KPI would have been without marketing. |
| Physics (Structure) | Attribution from the causal model. Based on known relationships (e.g., spend → response, lag structures). |
| Model Residual | Attribution from the AI-learned component. Captures real effects not fully explained by the physics model. |
| Total | The sum of all components—your final attributed KPI outcome. |
All values are modeled outputs in KPI units.
Control Options Reference
Control | Meaning |
|---|---|
| KPI Classification | Filter by KPI type (if available). |
| KPI Drilldown | Drill into specific KPIs (if available). |
| KPI Hierarchy | Filter by KPI hierarchy level (if available). |
How to Interpret the Results
- Read left to right: Start at Baseline, add Physics, add Model Residual, arrive at Total.
- Baseline height: This is your counterfactual—what would have happened without marketing. A higher baseline means a larger share of your KPI is non-incremental.
- Physics (Structure) bar: The incremental lift from the interpretable model. A large positive bar means the physics model explains most of the marketing effect.
- Model Residual bar: The AI correction. A small residual suggests the physics model captures most patterns; a large residual means the AI is filling in significant gaps.
- Percentage shares: Hover shows the percentage of total lift attributed to Physics vs Model Residual. This helps you understand how interpretable your attribution is.
- Negative components: Either component can be negative (e.g., if the physics model or residual corrects downward). The waterfall shows the net flow.
- Total bar: Your final attributed outcome. This should reconcile with your overall KPI performance.
Practical Applications for Marketers
Application | How to use this chart |
|---|---|
| Attribution transparency | Explain to stakeholders how the model arrives at its attribution—not a black box. |
| Interpretability assessment | A high Physics share means most attribution is from well-understood causal relationships; high Residual share means more is learned by AI. |
| Model trust calibration | Use this to set expectations: physics-heavy attribution is more explainable; residual-heavy may warrant validation. |
| Baseline understanding | See how much of your KPI is incremental (marketing-driven) vs baseline (organic). |
| Executive reporting | Provide a high-level summary of model composition for leadership reviews. |
Common Mistakes & Misinterpretations
Mistake | Why it is a problem | How to avoid |
|---|---|---|
| Treating Baseline as "wasted spend" | Baseline is what would have happened without marketing—it's not spend, it's the counterfactual outcome. | Understand Baseline as organic performance, not a cost metric. |
| Assuming Model Residual = error | The residual is not error; it's real effects the physics model doesn't capture. A large residual doesn't mean the model is wrong. | Treat residual as "learned effects," not noise. |
| Ignoring the percentage split | The absolute numbers matter, but so does the Physics vs Residual share. A small total with 90% residual is less interpretable than a large total with 90% physics. | Always check the percentage breakdown in hover. |
| Expecting the composition to be static | The Physics/Residual balance can shift with data, model updates, or market changes. | Monitor over time; don't assume a fixed split. |
| Comparing across different KPIs without normalization | Baseline, Physics, and Residual are in KPI units. Comparing revenue attribution to conversion attribution directly is misleading. | Compare within the same KPI type or use normalized metrics. |
Caveats & Considerations
- Aggregate view: This chart shows the total across all channels. For channel-by-channel breakdown, use the Dual-Track Decomposition chart.
- Baseline estimation: The baseline comes from the model's counterfactual prediction, not observed data. It depends on the model's assumptions about what would have happened without marketing.
- Residual distribution: The Model Residual is a single aggregate value from the AI-learned component, not broken down by channel in this view.
- Model version dependence: The Physics/Residual composition depends on the current model configuration. Model updates may shift the balance.
- KPI filters: If KPI filters are applied, the chart reflects only the filtered scope—not the total portfolio.