What This Chart Shows
These charts show (1) how much each group (e.g. channel, funnel stage, topic) contributed to the KPI in each time bucket, and (2) how the total modeled KPI change moved over that same timeline. The top panel is a heatmap allowing you to see which groups drove or hurt results in which periods. The bottom panel displays total Δ KPI per period, filled to zero, so you can match overall tempo to patterns in the heatmap.
Key Questions This Chart Helps Answer
- Which groups had the largest positive or negative attributed impact in which weeks/months/days?
- How does the shape of total KPI change align with concentrated bands in the heatmap (e.g. a campaign window)?
- Does attribution mix shift when I change grouping (channel vs funnel vs topic)?
Axes, Metrics, and Units
Element | Description |
|---|---|
| X-axis (both panels) | Time period (daily, weekly, monthly, or quarterly). Tick labels show calendar context (week/month/quarter styling). Range slider on the bottom panel to pan/zoom. |
| Top panel Y-axis | Group labels (channel, source type, pattern, topic, or funnel). Each row = one group. |
| Top panel Z (heatmap color) | Contribution for that group in that period. Blue = positive. Red = negative. Intensity = magnitude. |
| Bottom panel Y-axis | Total KPI change (Δ) for that period. Modeled total attributed lift; same units as the selected KPI. |
All values are modeled. They are not raw reported KPI.
Control Options Reference
Control | Meaning |
|---|---|
| KPI classification / hierarchy / drilldown | Which KPI (target) to use for the ledger. |
| Granularity | Daily, Weekly (default), Monthly, or Quarterly. |
| Grouping mode | Channel (default), Source type, Pattern, Topic, or Funnel. Determines how levers are aggregated into heatmap rows. |
| Date range | Filters the ledger/rollup to periods inside the range. |
| Group filter mode | Include or exclude the selected groups. |
How to Interpret the Results
- Heatmap rows: Scan for persistent stripes (steady contributors) vs short bursts (campaign-tied spikes).
- Color sign: One direction = modeled lift; the other = modeled drag (still model output, not proof alone of “bad” media).
- Bottom total: Peaks/troughs show overall attributed KPI change timing; compare to heatmap density in the same columns.
- Reconciliation: If sum of φ and total Δ ever diverge materially, the chart does not surface a user-facing warning; issues are log-only. Treat persistent oddities as analytics follow-up.
- Granularity: Daily = more detail and noise; monthly/quarterly = strategic rhythm.
- Range slider: Use it when the default narrow x-range hides later periods.
Practical Applications for Marketers
| Application | How to use this chart |
|---|---|
| Budget allocation | See which groups earned sustained positive φ vs one-off spikes before shifting spend. |
| Channel / tactic optimization | Compare grouping mode views (e.g. channel vs funnel) for different planning lenses. |
| Campaign evaluation | Align bright bands with known in-market dates and the total Δ trend below. |
| Experiment / validation | Use as hypothesis context; confirm material claims with tests or holdouts where possible. |
| Forecasting / planning | Use quarterly/monthly view for capacity and seasonality narratives tied to attributed impact. |
Common Mistakes & Misinterpretations
| Mistake | Why it is a problem | How to avoid |
|---|---|---|
| Reading φ as revenue or orders | φ is modeled contribution in KPI units, not necessarily a platform counter. | Check KPI definition and axis labels. |
| Treating heatmap cells as experiments | Colors reflect model attribution, not randomization. | Pair with geo/holdout or lift tests for causal claims. |
| Ignoring the total panel | Heatmap slices can look dramatic while total is flat (or vice versa). | Read top and bottom together. |
| Assuming all groups are shown | Top N (default 15) drops the long tail. | Raise Top N, use selected_groups, or export from a fuller view if available. |
Caveats & Considerations
- Empty states: “No granularity attribution data available” if the ledger is missing/empty; “No data in selected time window” if the date filter removes all rows—no silent blank chart.
- Monthly labels uses a 30-day approximation for some label paths; weekly/daily tick builders use shared calendar helpers—minor label nuance possible vs strict calendar months in edge cases.
- kpi_drilldown: KPI focus is via classification/hierarchy → target_series_id.
- Causal framing: Use for measurement storytelling and prioritization; not a substitute for incrementality proof where stakes are high.