What This Chart Shows
This chart shows how marketing impact spreads through your causal network over time and which outcome nodes are hit hardest and when. The top panel is a coverage curve which shows time in "lag steps" (how many steps from your intervention sources), against cumulative impact coverage (0% to 100%). You can see how quickly impact builds—e.g. "50% of impact by step 5" or "80% by step 12."
The bottom panel is a bar chart of the top 20 impacted outcomes: each bar is one target node, bar length is "path strength" (how much impact reaches it), and the label shows arrival time (e.g. t≈7.2). Together you see both how fast impact accumulates overall and which audiences/outcomes get the most impact and when they are reached.
Key Questions This Chart Helps Answer
- How quickly does impact from my interventions spread through the network?
- By what "time" (lag step) do I reach 50% or 80% of total impact?
- Which outcome nodes (audiences/performance targets) receive the most impact?
- When does impact arrive at those top targets?
Axes, Metrics, and Units
Element | Description |
|---|---|
| Top panel X-axis | Time (lag steps). Integer steps from 0 up to the max time used in propagation (e.g. 60). Each step corresponds to cumulative lag from sources; not calendar time. |
| Top panel Y-axis | Cumulative coverage. Proportion of total propagated strength reached by that time step (0 to 1.05). Normalized so the curve rises from 0 to 1. |
| Top panel line and fill | Single curve: cumulative normalized impact over time. Optional reference lines at 50% and 80% with labels. |
| Bottom panel X-axis | Target node. Top 20 impacted nodes (outcomes). |
| Bottom panel Y-axis | Path strength. Sum of absolute edge weights along the path from sources to that node. |
| Bottom panel bars | One bar per top target. Color can indicate positive vs non-positive strength.
Text outside bar = arrival time (e.g. t≈7.2).
Color by strength (e.g. positive vs non-positive). |
Control Options Reference
This chart has no filters, grouping, or date controls in the chart spec. It shows all intervention sources and top 20 targets.
How to Interpret the Results
- Coverage curve (top): Steep rise = impact accumulates quickly; flat later = most impact already reached. Use 50% and 80% lines to see "by when" you get half or most of impact.
- 50% / 80% annotations: "50% by t=5" means by lag step 5 you have half of total impact. If 80% happens late, measurement or attribution windows may need to be longer.
- Bottom bars: Taller bar = more path strength to that outcome (stronger "impact" in the model). Use for prioritization (which audiences/outcomes matter most).
- Arrival time (t≈X): When impact reaches that node in lag steps. Early-arriving, high-strength nodes can be treated as leading indicators; late-arriving ones need longer windows.
- Magnitude: Path strength is from edge weights along paths from sources; it is modeled impact, not raw revenue or spend.
Practical Applications for Marketers
Application | How to use this chart |
|---|---|
| Attribution and measurement windows | Use 50% and 80% coverage and arrival times to set how long after a touch or campaign to measure (e.g. 7 vs 30 days). |
| Audience and target prioritization | Use the bottom panel to see which outcomes get the most impact; focus reporting or optimization on those. |
| Campaign and tactic planning | Use curve shape to see if impact is front-loaded or slow-building; align pacing and creative with that timing. |
| Reporting and storytelling | Use "how quickly impact propagates" and "top impacted outcomes with arrival time" in decks and one-pagers. |
| Model and connectivity check | If the curve is flat or "no reachable targets," use it as a signal to check graph connectivity and edge lags/weights. |
Common Mistakes and Misinterpretations
Mistake | Why it is a problem | How to avoid |
|---|---|---|
| Treating "time" as calendar time | The x-axis is lag steps (propagation steps), not calendar days or dates. | Interpret as "steps from intervention"; map to real time using your lag definitions if needed. |
| Treating path strength as revenue or spend | Path strength is modeled impact from edge weights along paths, not actual revenue or spend. | Use axis label "Path Strength"; treat as relative impact from the model. |
| Comparing across different graphs or pipelines | Results depend on graph topology, edge lags, and weights. Different graphs give different curves and targets. | Compare only within the same model/graph; do not mix with other systems. |
| Using the chart when edge/lag data is missing | If required data is missing or the graph has no sources/reachable nodes, you see an error state, not a fallback. | Ensure the pipeline has written out_csr_*, out_edge_weights, out_edge_lags; check the on-chart message if empty. |
Caveats and Considerations
- Intervention sources: "Sources" are nodes with no incoming edges and at least one outgoing edge. Propagation and coverage are from these nodes only.
- Simplified propagation: Impact is computed with a simplified propagation (e.g. Bellman–Ford–style, limited iterations); path strength is accumulated along “improving” paths. It does not replace full attribution or econometric models.
- Portfolio-level only: No filters; the chart reflects the full causal graph. For segment- or channel-specific impact, use other charts.
- Top 20 only: Bottom panel shows at most 20 targets (by path strength). Other reachable nodes are not shown.
- Lag steps vs days: Time axis is in lag steps (sum of edge lags along paths). Mapping to calendar days depends on how lags are defined in the pipeline (e.g. 1 step = 1 day).
- Assumptions: Results depend on graph structure, edge lags, and edge weights. Use with other diagnostics and business context.
- Uncertainty: The chart shows modeled propagation and strengths; it does not show confidence intervals.